SaaStr AI London - AI School Academy Stage

Everyone's scrambling, getting their coffee, getting caffeinated.

Everyone getting settled? Okay, amazing. I think let's crack on. I always kind of wondered what the turnout would be for the first session of the day, but I'm pleasantly surprised. Yeah, so great to be here. And we'll do some quick introductions and then we'll dive into, you know, the theme and a bunch of questions that we've prepared. So I'm Deepika. I'm a principal at Northzone. We're a global venture fund deploying over a billion of capital out of our 10th early-stage fund. We've been investing for almost 30 years now, so we've seen our fair share of technological paradigm shifts. And we've been doing that across Europe, in the US, and we— our heritage is early stage, but now we're pretty much a full-stack fund and we do everything from pre-seed to pre-IPO. And we've obviously had the pleasure of investing in some amazing category leaders over our time. Spotify, Klarna, TrueLayer, you know, will be some of the names that you probably recognize. And obviously Personio. So I know it's really great to be here. Personally, I I love talking about product. I love sort of being very product-centric and product-focused founders and operators. So I'm really excited to be here with Li from Personio. But I'll let her do her own introduction.

Yeah, good morning, everyone. It's so nice to be here. So as Deepika mentioned, my name is Li. And I currently lead product marketing at Personio. Personio, for those of you who don't know, we're an HR SaaS company. We're the intelligent HR platform, and we kind of support HR admins across the employee lifecycle journey through our services. So you can think of everything like from recruitment to onboarding of employees to managing their performance and development cycle all the way to helping them get paid at the end of the month through our payroll services. But I lead product marketing, and what I love about product marketing is that we've got one foot in product, but we also got another foot in go-to-market. And I think this conversation is very much about the intersection of those two things.

Brilliant.

So the theme of the day is, you know, what does AI-native actually mean in a SaaS product today? And I think that's something we as an investment team, you know, debate all the time. We see it in our portfolio when we talk to our founders. And also to prospective founders. And, you know, it's something I think that's also top of mind for you and your organization. So maybe if we, you know, if we dive straight in, I think we've got a bunch of sort of, you know, prepared questions that we've discussed, and then we can probably leave some time for Q&A, I think, at the end. So if we dive straight in, when we think about, like, traditional SaaS metrics that we've all come to, I guess, know and love over the past 10, 15 years, as SaaS has been sort of the reigning paradigm, and now that's changing. Where do you see sort of like what is still super relevant when you think about product, and also how you talk about product at Personio, versus which metrics are actually still relevant, but they're not necessarily at the forefront maybe as they were, and what new metrics are coming to light that are really important that you're tracking?

Yeah, I think this is a really topical question, and maybe call me a bit of a traditionalist, but I do think that that kind of the traditional SaaS metrics still matter, right? Especially when it comes to kind of measuring financial health of the company or the actual business outcomes that you're trying to drive. So maybe I'll share an example from, of how we think about metrics at Personio. So we have a discipline of looking at the financial metrics that we care about and also the customer love metrics that we care about in a Driver Tree format. So the Driver Tree format is almost like a hierarchical diagram that outlines what are the key outcomes that you wanna drive and then what are sort of the branching off from it, what are the things and initiatives that you wanna take that you'll measure instead. So an example of this is, like most SaaS businesses, we care about net retention. And for us, we take a look at net retention at the very, very top. Very top, and the things that drive net retention could be things like churn prevention. We are also looking at upsell and cross-sell, so attaching new products into our existing base. And then also we might also be looking at downsells or downgrades where people might be reducing seats or downgrading their packages because that contributes negatively to net retention. So I think that So, that discipline of reviewing those metrics every single month is, is really helpful for us. And I think when it comes to AI metrics, we still look at the AI metrics in the kind of structure of this driver tree format. So, rather than having an isolated AI metric that might be kind of a vanity metric that we kinda look at, we make sure that it's baked into the key business outcomes that we're trying to drive. So, an example of this is, for us, again, net retention is really, really important, and we know that in order for us to keep customers, we need to keep them happy. And for HR leaders, what keeps them happy with their provider is that we are able to save them time in the kind of day-to-day grind of the HR admin work that they do. Do. So maybe a quick show of hands, how many of you have sent an email to your HR person being like, um, how many days of leave do I still have left? Yeah, we've all sent our HR leaders those questions. But if you imagine being in the shoes of an HR leader, they're getting these questions like time and time again. But what we provide that kind of helps them eliminate some of this administrative of work is that we've got the Personio AI Assistant. And so rather than, you know, emailing your HR leader this question, you can kind of type a question and say, hey, what's my vacation balance? Or help me understand this change in my payroll slip. And so when we first rolled out the AI Assistant that we have, we started looking at usage of the AI Assistant through, of HR managers that were using the AI assistant, but actually the true value comes in us saving time for those folks. So instead we shifted to measuring how many employees are actually within a company using the assistant and how many queries are coming from each of the employees. And we know that measuring those things kind of lead to happier customers and net retention. So I think net-net, like, traditional AI metric or traditional SaaS metrics still matter. But I think making sure that the AI metrics that you do measure become kind of a leading indicator to those business outcomes is like an important part of baking that into your process.

Yeah, no, that makes a ton of sense. I think, um, I guess one thing that we, you know, think a lot about, and maybe it's something that we just reflecting in the past year or 18 months that we talk about more when we look at companies or we talk to founders is, I guess, depth of adoption. Uh, and I think you, you know, you pointed out like how many people are using the product, but it's also so much more important of like how often are they coming back to it, how much time are they spending with it, how— or whether it's how much time prompting and whether that's compounding. Because I think there's traditionally when we looked at sort of SaaS products, like depth of adoption, like the proxy for that was How many integrations do you have? How deeply embedded is it? You know, and it sort of, you know, gives you a bit of a pointer of stickiness. Like if something is really embedded, it's hooked up to all of your other tools, you know, it's quite a pain to get rid of it. But I think in this like sort of age of AI, spinning up those integrations, you know, getting them up and running, exporting data, importing data, transforming it, it's a lot easier. So like the amount of integrations doesn't necessarily point point to, um, I guess, you know, it's not necessarily a leading indicator of stickiness anymore, but the depth of adoption in terms of like, is this tool like, you know, is it getting better and better every day each time you use it? Is it learning about the quirks of your organization? Is it learning about the edge cases that you come across as an individual or as a team or as a company? And I think that's— it's sometimes hard to quantify, but that's what we try to get when we talk to customers. We do these types of interviews because it really points to something that's like, it's not just a loss of, you know, the switching cost isn't, you know, the technical pain of having to implement or rip something out. It's like the cognitive loss. It's like you're losing someone on the team that's, you know, actually functioning at a pretty high level. That is, that's just something we, you know, as a team, we didn't talk about as much, and it's really at the forefront for companies today. So if we kind of move on, and it's a related point, It's, I guess, in this paradigm, like, what does ROI look like? And what are some of the, I guess, leading indicators that, you know, marketers, that, you know, operators and companies can, can look for, like those green shoots of ROI before it's maybe even obvious to the end user or the customer that they should look out for?

Yeah, I think this is a really good one from the lens of like if you put yourself in the shoes of the customer. Recently I read that there's a sort of an increase in purchase regret that's happening in SaaS, and this is quite common, especially in where Personio occupies, which is that we're selling to HR leaders. HR leaders has one of the highest purchase regrets in tech than, than other tech buyers, and I think I think what— there are lots of reasons why that is, but one of, I think, the meta theme here is that there's a disconnect between expectation and the reality of implementing software. And I do think AI can sometimes be like a— play a role into this as well, where we're expecting AI to change our world, make all of our workflows more efficient. And then the reality might be that it takes a really long time to implement. Things might not be coming to fruition as fast. So I think in terms of the green shoots of ROI that you mentioned, the antidote to purchase regret, I think, is focusing on time to value and speeding up that moment from when someone purchases something something to when they actually see value from your product. And with Personio, one thing that we've been trying to think about is, you know, AI, our AI capabilities aren't just bolted onto the product. It's actually embedded throughout the sort of the employee lifecycle that I mentioned before. And so one thing that we are thinking about in terms of of shortening that time to value is really thinking about what the onboarding experience and implementation experience might be like for an HR admin. So most of the times when somebody purchases Personio, we are replacing some kind of system, right? Like, for a small business, it might just be that we're replacing Excel spreadsheets, but a lot of the times they already have an HRIS that they're using, that they have all of their employee data on. And so migration and onboarding becomes like a total headache because you're trying to like transport, you know, hundreds of employees onto another database. And I think that's oftentimes when there's like a lot of pain that's caused. And so we've been able to, you know, put AI into the onboarding and implementation process by using things like AI data imports. Another thing that we we have on the payroll side is payslip parsing. So you can upload all of your payslips and all of this physical payslip can actually be parsed into data that can be imported into Personio. And that means that they can get up and running faster. We've been able to shorten our implementation times and onboarding times. And that really does shorten that time to value. And I think that is, you know, A, a defensive measure against purchase regret, but also kind of an offensive measure to showing ROI and making sure that your newly converted customers become like advocates and like are also able to speak to the power of your platform faster.

No, I think that's— those examples are really, um, they're super insightful. Um, I think on just, you know, on that last point of, of getting, you know, users to become advocates, almost like, you know, uh, you're— it's almost like doing sort of— I think AI tools lend themselves, at least today, to like a lot of product-led growth where like individuals need to feel that magic. Yeah, they need to feel like, yes, this is not just my like clippy assistant that's sort of spitting out suggestions or saving me, you know, a few minutes of time. It's really something that is doing a certain task or a process that I as an individual in that time constraint just couldn't do And I think that's when you get the magic moment of like, this is not just performant, but it's exceeding my expectations. And it's, it's, you know, yeah, I can only describe it as like that magic moment. Because when we do, for example, we do customer interviews, when we look at new products and new companies, you know, the investments that we've done, it's like people can't tell you enough about how much they love a certain product. And they really, you know, they're not just sort of very positive of like, yeah, this is great, this, this tool does what it says on the tin. They're like, they're, they're such enthusiastic advocates of it. And I think one example that really sticks in my head is from an investment we did last year. It's a medical scribing company called Tandem. So it helps clinicians summarize their notes, lets them focus on the core of what they want to do, which is helping their patients, listening to their patients, understanding what's going on with them rather than having to write down down a transcript, you know, in real time and doing that admin work. And some of the early users of Tandem that we spoke to, they literally said, if my clinic stops using this, I will quit and go to work for a clinic that is using this technology. And like, we just hadn't heard that kind of thing before. Like, who talks about their SaaS product in that way?

Yeah.

I mean, hopefully lots of you and lots of you have products that, you know, have those types of customers. But that is just like the level of excitement. And I think the other point that's really sort of top of mind for us at the moment when it comes to ROI is the point you mentioned on time to value. And I think it's not just because like time to value has always been a thing, right? The ROI, the absolute ROI that people get in 6, 12 months still matters, still just as important as it was. But I think with AI tools specifically in the year 2025 and going into 2026, like everyone who is thinking about using AI is in the market. They're testing 3, 4, or 5 different tools at any given time. I know, like we at Northstone, we are, you know, and companies and teams have dedicated Slack channels where they're constantly looking at tools and saying, hey, I'm trying this, works really well, you guys should give it a go. Or this was really great, but now it seems to be plateauing. Has anyone seen anything else? So like, and that, that That means, I think, if you're not able to demonstrate a certain level of ROI quickly, there's like 5 other tools ready, snapping at your heels, trying to replace you. I think that's just something you have to be really cognizant about at this moment in time. We'll see how long it lasts, but it's just, it's pretty brutal out there in terms of competition.

I guess the onboarding thing that I mentioned, as you mentioned, is also like a double-edged sword. It makes it easy for people to onboard to you because we're using AI, but it also makes it easy for them to switch to another provider as well. And so that kind of fostering internal advocates is really important. One of the, one of my favorite kind of AI features in Personio is what we've been able to do with performance reviews. So if you have managed people before, you might take hours reading through different peer reviews that people have submitted for your employees, checking back what you wrote for their previous review, and that can be time-consuming, but it can also lead to certain biases as well, because you might be reflecting on things that you saw recently, or you might be kind of thinking about stereotypes that someone might have. But our AI summaries allows you to kind of quickly summarize like all the feedback that your employee has gotten, and that's saving managers from spending 10 hours on a performance review down to like 1 hour. And then I think being able to see that manager is loving the tool back to the HR admins, I think allows them to also see the return on investment faster too.

Yeah, no, I'm sure that's really topical given the time of year.

Yes, yes, yes, there is definitely a spike in usage of our performance and development tool this time of year.

Great.

So maybe changing tack a little bit, and I think this is something that you're going to be, you know, you're super qualified to speak about. And it's about, I guess, you know, what we just said about selling AI. It's kind of the opposite end of the spectrum because I think probably in this room we're all digitally native, very tech savvy. We're constantly willing and open to trying, you know, the latest tool, the latest model. But when you're selling especially to sort of mid-sized to larger enterprise customers, that already has certain hoops you need to jump through, you know, certain, you know, I guess, product specifications you need to meet. But then you layer on top of that a conservative market, whether it's geographic, whether it's, you know, in a regulated industry, and that just adds a layer complexity, I think that, you know, given that AI tooling is relatively new, I think the playbook on that hasn't necessarily been written yet. So would love to get your perspective maybe on, I guess, a geographic lens of given Personio's home base is Germany, which is not, not necessarily known for being the most tech-forward-leaning country on earth. So what have been some of the insights from that?

Yeah, and I've Um, it's very nuanced, right? Like, I think it's easy to be like, uh, you know, a certain, uh, target audience or a certain demographic might not be willing to switch over to AI or might be reluctant. Um, but I think there are lots of things at play. So just for context, so Personio is a European company, but our home base is German. A lot of our existing customer base is German, and we've been around for 10 So I think the AI kind of surge has like predated us. And so we're kind of learning this as we go along as well. But what I found in conversations with customers and also through research is that I think generally there is like a higher standard for things like privacy and data safety in German, in Germany, and especially among sort of the smaller businesses. Size as well. And but that's not to be confused with like an interest in becoming more efficient or an interest in adopting AI, or in German it's KI. But so they are interested, but they just still have this culturally kind of a desire to do more due diligence when it comes to adopting new things that might, you know, influence the, the way their data is treated or how, like, customers might perceive them. Um, so I think building trust is really important, and there's, there's been a couple of things that we've done, uh, to, to build that trust. I think the first one is really to just have an open dialogue about AI with our customers, and that's even way before we even talk about our product. Um, I think the interesting thing that's happening with HR leaders is there's this technological shift that's happening, and the C-suite is actually turning to HR leaders and being like, hey, this AI thing, like, we should take a look at our org design and see how that might change, or, you know, we should really retrain our workforce to make sure that they're AI literate. And so HR leaders are kind of in the hot seat trying to figure out what this means for our organization. And so, you know, We want to provide them with a forum and a platform to bring these questions and to talk because sometimes it can be a very lonely job to be the leader of an HR organization. So being able to engage in conversation, I think, is the first thing. The second is that we've been really open and transparent about our own transformation as like an AI-first company as well. So Hanno Renner, who's our founder, has been really, really bullish about us transforming the ways that we work internally and making sure that everyone adopts AI as well. So, you know, we've brought in internal, external speakers to come in. We've had hackathons, we've had a lot of trainings on various software, and then we've been sharing that really actively on LinkedIn, on with all of our customer base to be like, hey, we're alongside all of you in this journey. And then it also gives confidence for people who buy into us because they're— they know that they are working with a company that is really thinking about AI-first as well. And then after we do all of that, we can like finally get into the product. Um, and I think one thing that's really helped there is that we've, uh, set out a couple of like product development principles that we've been really upfront about to kind of tackle some of the questions that people have. So, our principles around how we develop product, AI product at Personio is that one, we empower people that first and foremost, like, it needs to be useful. It isn't just like an AI-washing thing, right? It needs to be— have augment user outcomes. The second is that it's built fairly. So, you know, we are— we have measures in place so that there's no bias that gets incorporated or discrimination that happens because of our AI tools. Obviously, working in the HR space, that's really important. And then the third thing is that we are building with safety first and foremost. So we talk about, you know, where our data centers are located, you know, what are all the checks and balances we have in safeguarding people's data. People's information. So I think it's like a multi-step process of like engaging in a conversation. And, you know, even if all of you might not be selling into Germany, I think it's an interesting exercise to think about like who amongst my customer base is the most conservative or might be the most challenging to convince and kind of look through your AI features and your AI approach through their lens.

That makes a ton of sense.

Great.

I think we touched a little bit on, I guess, product marketing and how you talk about the product to your customers. But are there any other sort of, I guess, outside of Germany as a geography, any other, I guess, key learnings that you've made over the past, let's say, year or 18 months?

Yeah. So I think there are a couple of key competencies that I look for in a great product marketer. And the, the 4 that I can kind of think of are like, one, that they're insights-based, like they're always thinking about the customer and the market, that they are then able to translate that into like a go-to-market strategy. And when I say go-to-market strategy, I think you might think of like a product launch, but actually product marketers can be a lot more strategic and thinking about like, where does my company play and where can we win and how do we drive sustained growth. And then the last two competencies that I look for are the ability to tell a story. That's really important in a product marketer. And then the last thing is enablement. How do you get that story and share it with other people within the organization? And I think with AI, the latter two, the enablement in storytelling has become a lot more productive and faster. So we've been able to synthesize competitive analysis and build battle cards automatically using AI, or we've been able to— product marketers work on messaging, but we've been able to create lots of new variants of narratives and content pieces using AI. And that gives product marketers the permission to move more upstream and become more strategic. So being able to work on more go-to-market strategy, being able to really be insights-driven. So I think product marketing obviously is a role that is shaped by where product development is going, but I think the paradigm shift of AI is also helping us evolve into a much more strategic function as well.

Amazing.

Super.

Those were the main questions, and I can see we literally have 5 seconds on the clock, so I think we timed that perfectly.

Yeah.

Thank you so much, Lee. And thank you everyone for coming for the first session of the day. Really appreciate it. I'll be around the day, so happy to talk to anyone, answer any questions, talk about your products and companies. And yeah, I hope you have a great rest of the day at SaaStr.

Wow, right on time.

I'm like relatively unversed.

Sorry.

Oh! Hello!

Thank you for hanging on after the talk that didn't happen. I'd get a quick show of hands so that I can make this session as useful as possible. Who's an operator working at a company, not an investor? Operators, great. Companies under 100 people? Okay, companies between 100 and 500 people? Companies above 500 people? Okay, we've got a good showing of operators. Quick show of hands, who's got agents running around the company today of those who are operators? All right, now keep your hand up if you feel— keep your hand up if you feel that there's roughly 90% of the company using them on a regular basis and well. All right, okay, some work to do.

Jason M. Lemkin.

So, uh, I'm the co-founder of a company called Dust. Uh, we provide a platform that helps fast-moving teams design, deploy, and most importantly, maintain over time agents that are focused on internal productivity use cases connected to the best models, connected to internal data sources, and, um, quite crucially, easy to build and easy to have, um, getting used. Used by the teams. And over the deployments that we've seen, thousands of teams have used it, some fast-moving like Clay, Cursor, Vanta, Spendesk. We seem to be seeing similar patterns that I thought I'd share with you today in terms of success, maybe more often traps that then become success stories over time of adoption at scale on what I consider to be obviously one of the most important journeys that your companies are on for the decade to come. We'll basically talk about 8 traps that I think you can all avoid that seem to systematically hurt adoption. With this gold standard goal in mind, what if, what if 90% of all employees at my company got to use augmenting software in a way that actually made a difference? And also in a way that meant that they could continue to stay on top of the many, many changes to expect from this technology over the weeks and months and years to come. Number one, I think that basically, um, just before going into trap number one, I think what we'll see is that it's essentially always that people do things but not necessarily in the right order. Uh, and so, um, Who thinks that a top-down approach is the best approach to adoption within a company? Who thinks it's bottom-up? Who thinks it's a bit of both? All right, I think we'll all agree, and I think the order in which it happens is going to be quite crucial. The one thing that we see systematically happening is people getting excited about the wrong type of adoption plan and then sort of wondering why things aren't quite going according to expectations. Trap number 1, it's the whole like the CEO says we're doing AI and now we're doing AI and now what? I think that the AI washing fatigue is creeping in. We're starting to see the messages repeated. We're starting to get told to use AI more. Now it's agents. Is it the same thing? Is it different? But it's not quite clear how it's supposed to be rolled out internally. The number 1 mistake there is that I think the executive mandate is not accompanied with an explanation of how long it's going to take and how people are able to learn and not just know from the get-go. So this sort of executive mandate without the description that it is a cultural shift and not just a once-in-time decision is trap number one. Understanding that it's going to take weeks, months, years, that we're here on a journey together, that the technology we're talking about now is not the way collaboration with agents is going to look look or feel next year is just as important as saying, hey, presto, we're doing AI. I think availability of a platform is crucial, but availability of the space to explore, make mistakes, and actually celebrate the lessons that we learned along the way, not just the successes and the amazing ROI stories, is absolutely crucial too. That's trap number 1: the executive AI-washing mandate without the accompanying cultural space and time to learn, explore, and reward and recognize those who are trying to actually adopt the technology. Trap number 2. Trap number 2 is when IT is considered the gatekeeper. I often try to use the analogy when people ask me like, how do we get these to be used in the right way often enough with valuable use cases? I just, just replace the word agent by spreadsheet and imagine a world where spreadsheets were controlled by the IT department. Government. That's like, we've got a Department of Spreadsheets that reports to IT. If you have a spreadsheet need, just please come to us. Spreadsheets are pretty gnarly, they're pretty complicated, you've got all these formulas in them, you could get it wrong, so we'd love to actually help you. 2-3 weeks delay to get your spreadsheet out. Don't worry about us, we've got all the templates, we'll make it work. And just see how that might have slowed down the adoption of something that I think we all agree is synonymous with decently concrete productivity in the workplace today. So picturing IT as a gatekeeper is the trap that I see often. Again, they're obviously welcome to the table. They have amazing expertise in providing some of the guardrails and some of the reviews of the most complex connections to internal data sources, but they don't necessarily have the business acumen to understand on the ground how small productivity shifts from the technology might be the beginning of a broader adoption story story in those business teams. Okay, so I think that having them in the conversation as participants rather than owners of the topic is a pretty important one to not miss. Number 3 is on training. Who had mandatory LLM training at their company? Mandatory? Yeah, I think that making training optional is a bit of a miss. I think that people feel they have have surface-level understanding of what the technology can do because they've used ChatGPT on the weekend. Unfortunately, the technology is quite different when you're actually looking to plug it into internal, potentially sensitive data on scenarios that may be automations or augmentations, and that are going to get used by multiple teams or many members on the same team. I always take the simple example of of agent memory. It's a very trendy topic. People are talking about agents having memory and getting to learn. When you're on your own using ChatGPT on your phone, memory is a pretty simple problem. I'm Sandra, I'm logging in from London. This is how I prefer to travel, this is how I prefer to book my trips, this is how many kids I have because I've been asking you about them all weekend. That's memory for a single-player use case on AI. If I'm in a team and I'm using an agent agent. Should that agent learn from the last interaction it had with me? Should it learn from somebody else on the team who actually has the rights to modify its memory? Should it modify the memory of the agent, or should it modify the underlying documentation that's used to provide the truth and the policy about how many vacation days we have in this country? All those details are really the difference between surface-level single-player AI usage and at-scale adoption of agentic scenarios that are really going to power teams. So I think that mandatory trainings should be a must-have at companies with the basics of LLMs. What do they do? What are they really rubbish at? What can we expect progress to be over the, over the coming weeks and months? Who here feels comfortable explaining what RAG is? All right. See, that's the sort of indication that we still have a lot of progress to make, and it's totally fine. We're in a room of people who are interested in the topic. Who actually showed up to the talk and want to make it work at their company. So what do we think about the people who aren't in this room? Learning about the basics, learning about the Lego bricks of how agents work, trying to demystify them before people feel like they're completely left out and left behind is, I believe, a crucial step. And understanding how we're going to do this over time as the technology continues to evolve is just as important. We started building Dust at the very end of 2022. ChatGPT had only just launched, and my co-founder was coming out of 3.5 years as a research engineer at OpenAI. Back then, the context window of some of the frontier models were— was a few thousand tokens, right? A few thousand words. We're now working with models that can take all of Harry Potter in without a blimp. That fundamentally changes the architecture of the tools we're going to be able to imagine and work with. And we should continue to learn about this progress Yes, because the current time to double capabilities on reasoning is 7 months. By the end of next year, agents are going to be able to autonomously work on tasks that could take up to 7 hours. Right now, they go off track more than 50% of the time if we leave them on their own for more than 20 minutes. So just the shape of the solutions that we're going to be able to imagine with agentic software is moving at an incredibly fast pace. And so trainings as a mandatory set just to make everybody feel like they're part of the story is, I believe, a really important part of that. We all have conversations about the champions. How do we find the champions? I think one of the big traps is to pre-list the champions. All right, IT, you've got 2 champions. Product management, you know how to do new stuff, you've got 3 champions. Somebody else on the the team that was excited to talk about it last week, they should be a champion. Let's list people who asked for a ChatGPT license, they're going to be champions. What we see time and time again is that pre-identifying agentic champions is a mistake. On every single team, people who don't have the technical skills, who don't know that they're going to be great at this, actually discover themselves as amazing iterative creative partners to building very productive agents. It takes confidence, curiosity, a transversal mindset. It takes skills that aren't necessarily knowing how to code, but really deeply understanding how the business works and how people's day-to-day can be improved. So I think that understanding how people are using, discovering, and progressing is by far the more successful way. When the basic tools for agentic software are rolled out pretty broadly with guardrails and security that's provided by IT, with a mandate from leadership to explore and identify use cases that can be powerful, we see emerge the people who are actually going to define on business teams how this is going to change stuff for real for them in weeks, not months. Everybody's heard about the MIT report that says that 95% of POCs are failing, and when you go you go into the details of that report, what you see is that often they were talking about large single team projects with 3 to 6 months to get ROI instead of actually thinking about, are we teaching this team to understand how this technology works? Are we building a mindset of exploration on our teams? Are we on a journey that is going to make us an attractive company for the kind of talent that's excited to work with GenAI over the years to come. So I think that some of the failures aren't really looking at the metrics that matter over the longer term. Around 60 to 70% adoption rates. I'm talking about broad wall-to-wall adoption. So, um, one of the metrics that I like to talk about is high-value use cases. So not just using it for summaries, not just using GenAI for translation, but actually trying to connect it to internal data to produce assets or to move on decisions that actually mean productivity. When a company sees 60 to 70% adoption, which is already great, things change radically. The tools and the mindset of the leadership team has to shift. We have to move from a bottom-up adoption that's been approved and encouraged by leadership to a more structured, structured approach. Now, mind you, most companies never get to 60 or 70% adoption. So if you get there and stall, it's already progress compared to most companies at this stage. But I think that the last layers of actual in-depth productivity with agents comes from being able to identify that you've reached a point where bottom-up is not going to cut it anymore. And past this, structuring and implementing a certain number of interface spaces and, and processes across the company is going to be quite crucial. So let's name the mistakes and traps that we see happen after that. The first one is to actually now call out a dedicated AI team. I think sometimes people call the AI team before they know what they want to build, before they know when it's going to be useful. They have a committee where legal and IT represent two-thirds of the votes, and they don't even yet know what's actually going to be valuable to the business. But once you've had at-scale adoption, once the business teams have understood which Lego bricks they can play with, once they start picturing what can concretely happen on their teams, you get much more of a baseline to structure a team that is going to help the company on internal use cases, on customer-facing use cases, on the kinds of technology that you want to adopt, on the kinds of data sources that are going to make a lot of sense to connect to these leading LLMs, for example. At that point, I'd expect most of the members of that team to either be builders or decision makers with authority to actually roll out agentic scenarios quite quickly. And it has to be a mix of both. It can't just be people who have the authority to say it but don't really know how it works, or people who are like eager builders but don't really have a say in which data sources get plugged in without it taking weeks or months. All right, so only when you reach that level of adoption should you name the AI team. How many leads should you have on AI operations? I think what we generally see is that per business team, having one identified person that can relate to the company— hey, this is the data that we need to make stuff happen, this is where we expect ROI to be coming from, and this is how much time we feel we need to actually structure our efforts, is the right scale. They don't all need to be on your company-wide AI team, but they do need to be identified. I think this is shifting from a bottom-up builders go nuts to, okay, we now have an owner per team that is responsible for saying where they're most excited to build this. Those people may be identified only at the end of a couple of successful peer OCs. They may be one of the builders, but I really think that at that point they become the anointed person to also take responsibility for training. Again, models are going to change, capabilities are going to evolve, the way in which they interface with the software we already use or the software we're going to be buying is going to be changing. So those department or team leads need to be invited. So I think at a company of 200 to 500 people, which is what I roughly saw, I'd expect at at least half a dozen, maybe a dozen people to be identified in that group. By the way, this is one of the most efficient ways to share best practices. I've seen marketing teams understand that the same bricks can actually be used to do amazing market research that the legal team was using to qualify contracts and make sure that the sales team wasn't accepting indemnity clauses or liability clauses that were through the roof. It's the same bricks. You just need to hear about how the and how the outputs work, and which parts of the process can be done by the agentic software that you've heard about. I think that level of connection is really useful. This is where it does become important to have a shared roadmap. I think there's going to be different levels of agents in companies. Talk about roughly 3 levels. One, personal productivity agents connected to my email, connected to my calendar, connected to some of the ways in which I like to do things that are actually controlled by me. Maybe they're provided as guidelines by the company, but I get to customize exactly how they respond to my needs every day. The second level are going to be these team agents. I really think that they can serve ROI cases at a team level. They have to be able to be sort of ranked and scored on those business goals, and they have to provide concrete ROI to the those teams. And the third level are going to be either company-wide agents— you don't need 7 HR agents at your company— or critical use cases where a technical team is going to really be useful to come in and actually help increase the performance and robustness of these agents on business-critical use cases, right? That is best served by a shared roadmap, giving visibility to the technical teams on which have the highest priority and which data sources might be useful to multiple use cases. We see time and time again that some scenarios are blocked on one or a few data sources that haven't been approved, that haven't been vetted, or that haven't been properly connected, especially with the latest technology. Maybe some have heard about Model Context Protocol, MCP. There's a lot that can be done quite easily, but the guardrails to be put in place by businesses are important. So they will be blocked on those steps if there's not a common understanding of which use cases are highest priority. And then thinking about metrics, one of the traps that I think we fall in— and I mentioned this before with the MIT report— is trying to get to ROI before people even know what the technology is capable of. Until 60 to 70% adoption, waterfall, I'd argue that the main metric is is actually adoption. How many people are using GenAI in some way, shape, or form on a daily basis, on a weekly basis, on a monthly basis? Of those people, how many used it in a scenario that wasn't just summarization or translation or meeting note-taking? All right, that'll give you a first version of breadth and depth of adoption across the company. And then, as we mentioned before, When you start hitting that stalling point, 60 to 70% adoption, you started seeing sprawling of agents, people weren't sure which ones to use, people weren't sure which builders should be prioritized when you need technical teams to help assist on the most critical cases, shift to ROI metrics. At that point, it's okay for the business to say, okay, how much are we spending on this and how much do we feel it's going to get us over the months to come, over the year to come? One of the metrics that's probably hardest to measure, but that I would argue should stay top of mind for the leadership team, is are we more or less attractive as a company to the most empowered talent that we have and to the most empowered talent that we want to hire over the years to come? Any 25-year-old that's graduating today and looking for a job, if they're top of their class in agency, in initiative, in creativity, they're looking to join a company that's going to give them the sandbox and the tools to explore this technology that they're being told day and night is going to change their entire career? If you don't have an answer to that, you're not going to get that talent. And by the way, if you don't have an answer to that in your company today, you don't get to keep the talent that's already in your ranks and that's wondering, why am I still at this company if we don't get to use the tools that are going to define productivity in the future? So I think that's a metric that's pretty important. How do we think about about the top-performing talent, the highest-potential talent. Which tools do they have access to? How have they been able to reinvent their work and get rid of the tasks that everybody should be wanting to get rid of today? The mindset for those people is that they're not afraid of actually automating some of the tasks. They don't think it'll get them fired. They think it'll get them promoted, and they're right. And I think leadership teams agree. So making sure that those metrics, that those dashboards are also accessible to the leadership and people team is, I think, very important. That's it. I think that the stages we have ahead of us are really playing out right now. So as operators, as leaders in your companies, you have a choice to make. Are we going to roll this out in a super centralized way where IT is the gatekeeper, where we take weeks to get back to people who have an idea that they want to explore, where we don't recognize or reward the mistakes or the explorations that are going going on, or are we going to have a rollout that actually lets emerge some of the builders and champions of this technological shift that's not just going to take the quarters between now and the end of 2026, but is actually going to define if our company maintains a competitive edge over its market over the years and decades to come. Um, yeah, you're going to get pressure when you go back to say it's not safe, we have to centrally manage We need to slow down. I think some of the ways in which you as leaders can push back is to just come up with the examples of the kinds of people that you want to be promoting over the years to come. What are they thinking about the ways in which your company is approaching it? What are they thinking about the ways in which their toolkit is evolving? Thank you very much. Happy to take any questions if there are. Do you think we have time for questions? Right, I think that was one. So if you do have any questions, feel free to raise your hand. And then we— yep. Which teams have I seen drive the biggest adoption? So I think there's a— I'll sort of answer for high— higher growth companies, which is one of the areas that we focus on. Plus the 3 teams that I found most most reliably implicated in some of the sort of highest average use cases: sales enablement, support enablement, and engineering operations. And I think within those teams, the ops functions tend to be quite critical. The mindset that I feel is the biggest predictor of eagerness to deploy useful agents tends to be transversal thinking. People that don't just have a siloed view of how they're inventing their work, but understand the inputs and outputs that their team or function is providing to the rest of the company. Productivity to me is 3 things, right? You can do the thing faster, which basically means doing it cheaper, or you can do it better, which I feel that we've seen this technology be quite useful for. You didn't have support in Japanese and now you have support in Japanese. Thank you, GenAI. But the third pillar that I'm most excited about is doing the things you couldn't quite do or didn't have energy to do previously. One concrete example that we've seen come up quite regularly is agents that take data from one part of the company and provide a translation of it to another part of the company immediately and for free. So agents that will translate the codebase to the marketing team and say, hey, we just pushed to main. You don't need to know what that means. But basically, this segment of customers now has this feature. And I've directly gone and read it from the codebase. Or the other way around, hey, I've been looking at core transcripts coming in from prospects and these two features that the team is working on, we're probably not thinking about the sizing of the feature correctly because there's a systematic piece of feedback coming back that I don't see in any of the design docs that the engineering team is sharing on. So those I feel are really best transmitted by people who get up in the morning and think not just about their tasks but about the goals for the company and how different parts of the the team might benefit from it. Yep. Yep.

Yep.

Absolutely. Yeah. So the question is around training. Many companies treating it as a checkbox, but given the pace of change, what are we seeing as successful ways to actually train people up on a continuous basis? I'll answer bluntly. I think agents are an amazing source of training. I think the way in which the product itself can be customized by companies to provide quiz cases, examples, and a sandbox for people to explore is by far the, the, the, the most effective way that I've seen training happen. It doesn't have to be a slideshow, and it definitely shouldn't be a video. And so I think the broad availability of tools, again with the sufficient guardrails to make sure that the use cases that are going to pop up aren't suddenly leaking data that shouldn't be going one place to another place, is I think one of the most concrete ways to get people excited. It also allows for very discrete and granular monitoring of who's genuinely excited about going further and exploring, which goes back to identifying the champions that I think are going to be local relays of training and keeping people abreast of the latest changes as it continues to evolve.

Yeah?

[FOREIGN LANGUAGE] what it means to properly measure adoption. So to me, adoption— I think about, uh, 4 stages of, of agentic AI in a company. I think stage 1, 2 years ago, was like, let's fight shadow IT. People are using ChatGPT on the weekend, we actually don't appreciate the way that's potentially leaking customer data, so we'll have like a license for them to use it in the workplace. Level 2 has been connecting this technology to internal data sources. We have chatbot connected to internal data. You know, if you're working at ACME, chances are it got called ACME GPT or, you know, uh, Chat ACME or something like that. All right, that's stage 2. Stage 3 is agents, actually adopting them with productivity use cases where it's not just providing an answer, but it's maybe used in a continuous way or in an end-to-end way, and you're able to put an ROI number to the tasks that it's taking into account. I think stage 4 that we're looking down the barrel of now is actual in-depth collaboration with agents on long-lasting tasks. Going back to how long these agents are able to work for engineers, software developers are really acutely aware of the fact that it's now possible to throw agents at complex coding problems for minutes, if not hours on time. And so I think that stage of adoption is like, am I able to take on a project with team members and agents? And so measuring Knowing which stages and which types of agents are being used is to me just as important as the frequency of usage. Because if you use an agent that prepares a very complex meeting on a weekly basis and it does an amazing job, that is very high ROI, even if you're only tapping into it once a week. I'm seeing this thing blink, so I think that means I'm off. Thanks ever so much, everybody. Have a great SaaStr.

Uh, I'm Shalin Jain from HappyFox. Glad to be at SaaS Tour London. Today, in the next 20 minutes, I'm gonna make sure that you leave well with 5 questions. I know every talk wants to give you an answer on what to do, but I'm going to leave you with questions that you can take home. And these are the questions we asked, and I think every founder needs to ask themselves to get better at deploying AI. So our focus is everybody wants to implement as many agents, chatbots. How do we get it done in a way that it feels all streamlined? Client. So just a little bit about us. Happyfox is an AI-first software stack for support teams. We also help IT and ops team to manage their work. We have over 2,200 customers. We are about 100 employees. We are in 70 countries and over 30 languages. We are $20 million plus in revenue, which we crossed earlier this year. And we have not raised any capital. So we are completely bootstrapped. We are 100 employees, which also means we are extremely lean in how we operate it. And we have 8 products, everything from the system of record to analytics, workflow automation, and AI. So 3 of our core products where we have started doing AI, chatbot we introduced in 2018. So the pre-LLM era when AI was not a hype train. We did Copilot, which was launched last year, which basically augments agents to do work while chatbot helps customer to not raise a ticket and get their problem solved. And sure, everybody's fairly aware of that. Autopilot is our latest product that we launched last month, which is a collection of AI agents for support teams. So we built over 60 agents and deployed. And one of our core premise for having AI agents was I don't want anybody to sit and build anything. Everything should be ready to ship out of the box. So if it takes less than 60 seconds for you to deploy an agent, we won't release that agent. We'll figure out how to do it within 60 seconds. So essentially, literally everything from triaging a ticket to making sure you're automatically closing a ticket when a customer says thank you, to doing everything complex in between. And this has led to a complete rewrite of the job description of what a support rep really does. And we went one step further. We extended that to success team, product product team and engineering team. So now a support ticket which has a bug instantly gets a bug description sent out to our QA team that doesn't even have access to the support ticket, but can get access and gets the summary of what the reported bug is. And it can quickly escalate to other teams as well. So if we heard on a ticket that somebody needs an integration with a new app, There's already a product requirement document that can be created. Or if a customer says, hey, I want to export my data, we know it's a potential churn risk. It automatically flags the success team to get into action. So one ticket created can have implication not just for support team, but for everybody in the organization. Now, being lean, it's very hard to stay doing the best in the market. And generally it's shown that if you have a large team and it's organized really in a good structure and you have good hierarchy, you will make sure everything gets checked. But we have a flat team of 5 support engineers, which we had at $1 million and we still have at $20 million. And we are doing support at 98% CSAT score. We were at 97% for many years and post our chatbot and Copilot, we finally cracked the 98%, which was a huge deal. And next year, I don't know how we can get better than this because some customers do love to complain, but hopefully our agents will reduce work and stress for our agents, for our support reps, and they can do a lot more than just doing support. So being lean is a huge advantage, but keeping the financial discipline, making sure we are scaling efficiently, and by hiring generalists, we have kept silos away from the company. Instead of hiring too many leaders in different departments and creating different teams in a very hierarchical manner, we have focused more on hiring generalists. So how do we scale from $20 million to $100 million? $100 million, and it's every founder's dream that this happens at the fastest possible pace. So with AI, you can ask this question once again and say, what can AI do to accelerate this, and can I actually get to this without even wanting to raise any more— any funds, right? So being bootstrapped, staying bootstrapped all the way to $100 million It's not done that often, but it's a hard challenge to take. Now, Shopify founder put out this tweet. It's a long post, but I put the parts that I felt was relevant, was always hire, always prove that AI can't do this job before we hire a person. And the last sentence was specifically interesting that this can lead to a lot of discussions inside the company and projects and what you can do about it. So my takeaway from that and how we are thinking is that AI is not an optimization step. And I think we are falling into this trap very easily where we think AI, like if I had a screwdriver but I was now have a power drill, it would make my job much easier. But I think it's a mindset. Much bigger change, and history has proof for this. So, in the last 100 years, Industrial Revolution, computers, and the advent of internet has completely reset what an organization chart looks like, and I think AI is about to do that. We are just at the tip of the iceberg trying to put some AI into hands of every team member, but what if we have to actually redesign our companies and our org chart for AI. So we thought about 3 core pillars that we think needs to be thought about. How your work is designed. We are no longer just human-based work, we are human plus AI. How is org wired? Job descriptions are going to change because AI is going to take over some and humans are actually going to do a lot more judgment than they ever did before. And how we govern and learn and keep the symbiosis with AI is extremely important. So as we thought deeper, we felt that we need to find the building blocks to deploy AI at scale. And we thought about, we thought about playbooks, steps, and checks. So think of when you hire a you always think about what projects they are going to do, how their task is going to be divided, and when the person takes over the job, he thinks about all the subtasks that actually will make him successful. Similar to that, at an org level, we need to think about playbooks, the steps, and then checks we need to have in place, especially when AI is involved. So we have hired a lot of Jason M. Lemkin's people. Generalist and a few specialists, but the new org structure will not be the typical org structure that we know of from the org diagram, but it would be better classified when you think of it as a team of generalists, a team of specialists, and agents. The generalists are the orchestrators. They know what to do, how much to do, and what is a good outcome that it could would lead to, and the specialist's job is to figure out how to do it best. And agents will actually take it to scale where you can execute that task thousands of times in a repetitive manner with predictable outcome. So these are the 5 questions we asked ourselves. What is the highest volume activity we do in our org? What does this activity mean for every team in the company? What are the judgments that we are taking and what is the work that is highly repetitive and what governance do we do right now? And if AI was involved, how do we up our game with governance? And we started answering these questions and we took an example of our sales team and this we have actually implemented at HappyFox that for one of our products, we do about 2,500 demos a year with a 15 to 20% conversion rate and it means something to everybody in the team, not just sales. Sales is usually the people who are involved, but most of the times even the directors or the VPs involved don't get to see what actually happened on the sales call. But if you think this has an org-wide implication, then there are judgments that will be taken by sales, marketing, product. There'll be repetitive activities each of these teams will do, and there'll be governance which will be involved that managers and the superiors will try to act upon. Now, so when you think of all the judgment and management currently happening, and if I just look at the sales org, there is everybody from the rep to the VP or the CRO is making different judgment and trying to make sure are they all on track? So pre-AI, sales reps had to write notes, very laborious task. Post-AI, we now have agents available that can actually write great summaries, right from the vanilla Zoom summaries that you have to something that's augmented by sales enablement tools. And but what would be the next way of implementing or thinking about AI would be how do we structure value for entire org? And in this case, for our sales demo, more around sales, marketing, and product. So imagine a world where you finish a demo and once the demo is over, the sales rep has the notes already written. The marketing team has already created collateral coming out of the demo as the demo actually completed. And the product team, if they, if there was any mention, hey, we need a specific integration with your product which your competition has, that we have already run an agent which deploys a bunch of playbooks and the product team already has a product requirement document already written for them right after that call. So This would be crazy speed that after a sales call, the entire organization is in action without anybody doing anything. And if I could run at that speed, I think getting to $100 million would not be that hard, and I would have discovered my reusable rockets to go to Mars. So to operationalize this, I mean, it's all great when it's an idea on paper, It all sounds good, but can we actually operationalize this? So we built a product. Me and our CTO locked ourselves in a hotel room for 5 days. We had all of these discussions with us. We made sure we chose a neutral country where no disturbance can happen. We are not connected to anybody. So we found a middle spot to meet and we met up in a hotel room and started— I started designing, he started writing code. And we actually built a product in 5 days and we have been working on it for the last 2 months ever since. But we built a general purpose AI acceleration and orchestration tool, which in this case I'll continue to use the sales example. All the playbooks and the AI agents we dreamt about was now built on this tool. This tool, first job it does is orchestrates what work needs to be done. When you define a product in a non-AI world, you always think about what are the breakup of this project, who will handle it, what would be their task and steps to do it. So just in that manner, every agent will have its set of assumptions, instructions, it will operate on certain playbook playbooks and give us the output. So every output of that agent run, and this is an actual output of one of the sales demos that we did for a company where we wanted to make sure that that sales call and every sales call here on at HappyFox is turned into all of those 9 playbooks that we, we saw. And this is how detailed our team wanted really detailed analysis of everything. And there were clear guidelines to AI that anywhere where you don't have inferred, please mention do not infer. So we created playbooks for medic to have medic summary. We created a playbook that will automatically creating marketing assets, everything from, hey, what were the good taglines that the customer mentioned that was really valuable? Marketing team should be aware of. And it's very common in companies, the marketing collateral, the sales reps don't want to use because they think it's already outdated and disconnected from what really happens on a sales call. So now you have all the marketing asset completely in line, generated for each of the 2,500 sales demo instantly as that demo is completed and Zoom sends the transcript. So we do multiple agent runs. Across all the teams and we have different playbooks that are implemented and we have also built complete product feedback. So if customers said, hey, I couldn't figure out how to use this and that even happened on an onboarding call, that will instantly turn into a spec for product team to look into. So what are humans going to do more of? They're going to do more of judgment. Post that sales call, I would already have collateral based on the templates I have and the use case I use. We have a lot of customers who are moving from Zendesk, and oftentimes we just send them a plain vanilla battle card on how we are better than Zendesk. But now with this process, we actually, if it's an IT use case, the battle card is already a new battle card battle card is created for that customer with inference from the original battle card, but customized to look like as if the sales rep sat and built the battle card for them. But it all happens within minutes of that call. So we want to bring it all together. And so to use this orchestration, build all the playbooks and agents, We want to operate our org differently. So instead of thinking of teams and large hierarchical structure, we are now moving from siloed teams to cross-functional teams, which we call it as pods. So a specialist and a generalist and AI agents are all considered all over again by the leadership team on if this was a pod, what would all they— what would what would be their job descriptions, what, what do humans not to do anymore that can be taken over. One of the challenges implementing any third-party AI tool is it's a black box. The way that tool is written has some core prompts that you are not able to modify. But with Grid, we are able to turn that into a more deterministic AI where you're not only defining determining what outcome you want, you're also determining the steps that you think will lead to the outcome, and you're using AI to augment that. And then we have a weekly leadership ritual to go over all of those, whether it's number of sales demos, whether it's tickets that are created, and so on. We are looking at that every week and trying to take one part of our org and reorg that structure. And I think this is what is going to keep us lean and make us more AI native so that we can deploy more agents, more AI playbooks at a record speed while staying hopefully at $100 million. We should not be over 200 employees. Thank you. So we have some time for questions if anybody has any questions.

Questions?

Do you think bootstrapped companies have advantage or disadvantage at this age? So I think there's a huge, definitely huge advantage to bootstrap companies, especially if they are lean, because you have not bulked up your organization just to to look right on paper, right? You're doing only the things that are required and in most bootstrap companies that I know of, lot of people wear a lot of hats and which means you have a lot of generalists already in place which are gonna be very important. For specialists, you can always hire an agency, say, show me how to write the best ads and help me augment that in Grid and then you have the best ad creator and your generalists who know what is a good outcome could take over. But even funded companies that operate in a lean manner and have not added too much of hierarchy would still be at an advantage and they can accelerate even faster.

Yeah.

Hi, we're looking at doing something similar and kind of systemizing it across the company.

Do you use orchestration tools like NA10 or have you built all all of your orchestration layer yourself? And how do you document all the different automations you've got running?

[Speaker] So there are two note keepings we are doing. One is essentially rewriting our org structure, redefining how we are supposed to operate. Those used to be just diagrams earlier, now they are living documents on, hey, how— and we are, again, one of the things that I showed here was we are not driven by teams, we are driven by activity in the org. I think that's a very different angle to look at it. If 2,500 sales demos happen, what does it mean for the org? So that's a living document. So every such activity in the org is becoming a living document. And to keep all of our playbooks and agents in order, again, Grid organizes So that product automatically auto-organizes everything. It has how many times something is run. Anything goes stale, we can immediately get rid of that. So it's a self-managing thing. We have no folders, no tags. It is automatically organizing everything, keeping in mind that nobody should be configuring this tool beyond the specialist writing the playbook and the orchestrator, which is the generalist actually executing the agent. So we're up on time, so I can take more questions. We are at the Happy Fox booth right at the entrance. I will be happy to answer any questions. And thank you everybody for coming on this talk. Thank you.

Cool, we good? Nice one. Hey everyone, I'm Simon. I'm the co-founder and CEO of Alloy, and today we're going to be talking about the death of the design handoff, one of the dreaded phases in the waterfall software development lifecycle that's pretty much gone as of right now. What I'm going to be doing today is showing you 3 new ways that we're seeing top teams evolve their products and take advantage of a new superpower that's coming about with this new category of software, uh, called AI prototyping. So a little bit of background, uh, if you can't tell from the accent, there's a bit of Australian twang, and that's because, uh, my co-founder and I are from Canva and Atlassian, essentially design and and product planning companies that have sprung out of Sydney. And, you know, when we were working at these places, what we could feel, what I could certainly feel in my role as a product manager, was that planning was sort of this archaic process that was stuck in the Stone Age. You know, you'd get in a room with people, try to put a bunch of sticky notes up on the board about what you wanted to do, argue priorities. You couldn't really visualize any of this work, and then you'd get into 200-line spreadsheet where again you're sort of theorizing about the order and priorities of how you want to evolve your product. So we thought there's definitely a better way to do this, way that's a bit more visual, kind of Canva-like, but brings together all of the enterprise workflows that Atlassian was really well known for. And so we got to work on that, and after 2 years, our first product looked a little something like this. Uh, and so there's a little bit of usage, but nothing too groundbreaking. Uh, and that's because, uh, Alloy 1.0 was pre-AI product for planning and, and prioritization. We weren't actually moving the needle forward that much. Um, and then something really interesting happened. As people were using our tool, these product managers weren't just planning and, and prioritizing in the sort of old school pre-AI way that our product was built for, but they were starting to attach prototypes, uh, in tools like Lovable or Replit or v0. We thought, huh, that's really interesting. There's sort of this behavior shift that we hadn't seen in, uh, 10 or so years that we'd been working. In product, uh, but once we went in to these product, uh, prototypes that they were attaching, we noticed that they didn't really look anything like the products that people were working on. So if you're attaching a, a prototype and you work at Pylon, you have this sort of purple gradient, uh, prototype that doesn't really look anything like your product. Certainly not something you could show to a customer. So we set out to build a prototyping tool that was for existing products and professional teams. And we launched that about 2 months ago. And then the chart went like this, uh, which was pretty cool. Uh, and you know, what is kind of important about that discovery is it's kicked off now what we're gonna be talking about today, which is new category of software for AI prototyping. With your real app. And I'm going to show you exactly how you can start doing that, the playbooks and workflows that people are using to implement this in their orgs. Um, and of course, everyone is now going out and, um, shipping tools that enable you to do this. And so, uh, you're welcome to use Alloy for this. You're also welcome to go and use anything else. Um, Alloy's completely free to use. Uh, and so you can just try whatever you like and, and see what fits you best. So it's not really about the specific tool you use today, but it's more about how the workflow is changing and what it's gonna look like headed into, to 2026. And by the way, as soon as we went out the door, we basically had signups from all sorts of companies, from small startups all the way through to the Fortune 500. And so, um, regardless how big or small your company is, this category of software, this new way of working, working and designing products is the kind of de facto way forward that you're going to want to have a play around with and get acquainted with. So these are some of the things people are saying about AI prototyping. The most fun 5 minutes I've had with a product ever. Blows away other tools. It's a hack for founders to convey ideas to their teams. All sorts of holy shit, dude. That's a good one. Um, and so people are really excited by this new way of building, and it's a superpower for evolving your product that you're gonna want to try. Um, Aaron Levy, who's a great friend of, of SaaStr, uh, put it really well in this tweet. Uh, more and more product managers are now using AI, uh, for functional prototyping. And we definitely think that it's gonna be changing product management forever. All the benefits like increasing the feedback loop, uh, identifying problems that previously, you know, wouldn't have been seen until you went many weeks into your design process, your workflow. Um, these are all solved challenges now, uh, that you can take advantage of with this new technology. So how do I think about this? You know, a bunch of you came by the booth and, um, have either been using tools like Alloy already, um, or have been thinking about using them. And the way to think about it is that 2024 and prior, we had sort of the legacy design tools where I like to say you were drawing every rectangle yourself. And so this was very much catered to the waterfall, waterfall rather, way of working where you'd write up a long document. You would then go ahead and sort of argue for resources with your team, get back into those planning spreadsheets, get a designer's resources, finally see a prototype or a visual that you can't even really interact with yet. So this year, that kind of well and truly went away. And then we saw this secondary kind of interstitial category of these consumer app builders that professional teams were trying to sort of shoehorn into their product process. Then the challenges is how do you make it look like your app, uh, if you're pasting in a screenshot, getting these sort of random purple gradients, um, and UI elements that, uh, sort of make it hard to actually immerse yourself in what it would look like if you evolved your product in this way. And so that's where AI prototyping comes in. Uh, and we're gonna show you exactly how to, how to do that. So easier than a Loom. It's only gonna take a second or two to do as we're about to see. Looks exactly like your real product, and it gives you feedback and input in minutes that you otherwise would've never really, uh, been able to, to get. So this is a whole new superpower, new technology. It's available to every product team, and there's 3 ways I'd love for you folks to think about how to make use of it. And so let's get started. The first one is, is instantly exploring new product ideas. And so what I mean by this is anytime you're thinking about, you know, a feature that you'd love for your product to have, or I wish it did this or looked different in XYZ way, you used to have to sort of put that to the back of your mind, again, because you have to fight for resourcing. Maybe you don't have a designer this week, but now you can and actually just visualize it straight away. Uh, so no more coming up with the idea, waiting, hoping you like it. Just prompt your change and then you're done. You can share a link just like a Loom. Some of you might be familiar. I'll go back. I saw a few people taking photos. So 2 weeks down to now 2 minutes. There is another way that you can prototype that some of the more technical people, um, in the audience might be familiar with, and that's to actually get the code base in your hands yourself, run it locally, um, and then start prototyping with something like Claude Code or, or Cursor. Is anyone here technical enough to, to be doing that? I get a show of hands. Awesome. So like a quarter of, of the audience. Um, and that's exactly what we've found. You know, if you're really strong, um, at your sort of development toolchain and you want to run Claude code locally and work with your actual codebase and you want to make a straightforward change, um, that's still a great solution for prototyping. There are some challenges with sharing the link and commenting, um, but they're like somewhat solvable. Uh, but the real opportunity that we're seeing now, um, is that you don't You don't have to be a design tooling expert. You don't have to be an engineering expert to be prototyping, visualizing your product ideas. If any of you can interact with your browser, with the computer, you can actually now start showing people how you might be able to evolve your product and generate better outcomes for your business and for your customers. So would you like to see a demo? [SPEAKING CHINESE] Just gonna flick over to the AV here. Sweet. All right, so Linear is our sort of reference product today. So put on the hat of maybe being a product manager or an account executive at Linear, not really knowing how to use a professional design tool. or, or use a, a development environment, but, um, wanting to, you know, evolve this product in, in some way. So previously, you know, maybe I could screenshot this and put it into a vibe coding tool and, um, get back something that doesn't really look right. Um, or I could again get access to the repo. But what we think is much better is just opening up the Alloy extension here. and then clicking the capture button. And in a few seconds, what our system is doing is downloading all of the context of this application, the icons, the components, the layout, the library, and then taking me to an exact replica of my app that I can now just talk to, uh, with text. And so here's that original Linear page. And then here it is in my AI prototyping tool, ready to, to now go change it in any way that we would like to. And so let's say my idea is, um, one of the common ones for these task tracking systems, which is, hey, the list is cool, but wouldn't our customers love if we had a whiteboard? Uh, and so I could go and fight for weeks of design and engineering resources to have a prototype of that. Or what I can do instead is I can say replace this list with a set of whiteboard cards instead. I should be able to drag the cards around on the canvas. And then I'm just gonna go ahead and submit that. It'll take a few minutes, uh, and, and get to work on implementing the task. But if we fast forward a little bit here, you'll see that it's done in this other tab, and I've now got a whiteboard canvas that Alloy has designed for me. I can drag these cards around, and now I can maybe imagine how my customers might be planning their issues in Linear with this new 3D format. And of course, I can come in and give it a subsequent prompt. Maybe I want to change the color or the layout of the cards, add a new feature like the ability to zoom in or out. Uh, but you can see that it's already pretty much nailed the design system of my product. I didn't have to connect GitHub and get IT approval. I didn't have to run the code base and, uh, sort of jump through hoops in Slack with my engineering team cuz there's some error with my API key. Uh, it just works in, in one or two clicks and, uh, couple minutes of, of processing. If I wanna now make some more fine-grained changes like I would in a traditional design tool, I can always do that. So maybe, um, we want to go ahead and, uh, like make these round, add some sort of, uh, border to these. It's not the prettiest of changes, but, uh, you don't only have to use AI. You can use it like a traditional design tool as well for the for the final polish. And so that's one of the really big ways that you can start using, uh, these AI prototyping tools. Just visualize your ideas instantly, share them just like a Loom, um, and get some instant feedback in 2 minutes instead of kind of 2 weeks or, or months. Second big way that has sort of surprised us us about people, about how people have been using AI prototyping to improve their business is visualizing not just their own ideas, but actually customer feature requests live with their customers in sort of sit-down sessions. So is anyone here in a sort of customer-facing role like sales or— nice one, couple hands. You'd be very familiar with the process of getting on a Zoom call, having the customer say, "Oh, I wish your product could do this. Like, can it do that?" And then you'd have to say, "Oh, you know, cool idea, but like it's on the roadmap." You try to understand what they mean. Maybe you'd misinterpret. The customer's in the dark. This would get delayed the next quarter. You're half a year down the road. There's been actually zero progress on this customer ask. Ask, you've just put a sort of line item in, uh, your ticketing system. The new way, the 2026 way, uh, is while you are on that call with the customer, just prompt your change in the background. And then when they're done talking, just share your screen and go, oh, by the way, is this what you mean? Is this that like new reporting feature that you've been asking for? And the customer will, like, light up. And we have people that are now closing sales based on this. We have people that are increasing their customers' satisfaction, uncovering requirements they previously wouldn't have been able to understand because they're showing the customer something visual, not even, um, 24, 48 hours later, but live on that initial customer discovery call or sales conversation. And so So we can take a look at what this looks like, and I'm going to use a real example from a customer that came by the Alloy booth today. I hope they're okay with me doing that. But this is Scoro. I've just logged into a dummy workspace that I created here, and let's imagine that one of the feature requests Scoro has received from their customers is the ability to add an extra pending column to this view and add some more details about each of these specific clients in the Kanban board. And so enriching it with, you know, some logos, some additional information. So basically, very common customer requests you'll get on a call, they'll say, hey, we don't have a 4-step workflow, we have a 5-step workflow, workflow, I need to be able to see more fidelity about my customers without having to click in. And so, let's again, just open the extension, click the capture button. And then if we give it a few moments, I'm going to pop into, again, an exact replica of my app. There we, there we go. So, So here's what Alloy's picked up. Here's the original. Back to Alloy. And so I'm going to go ahead and ask it to add a 5th column after Proposal and before Confirmed called Pending. It should also, along with all the other cards, Have more information in each card about the company. Ideally a logo, other enrichment, any other useful ideas you have. There's also a good way to, um, put your AI to work, 'cause sometimes it might come up with stuff that you haven't even thought about. Uh, and then again, just go get a coffee, talk to your teammates down the hall. Uh, when it's done, you're gonna get something like this. So I'll pop this away. It's gone and added the column. It's picked the right color. It's matched the branding perfectly. Uh, it's gone and generated some, uh, demo customers. So you can see Slack, uh, Slack here, Atlassian. I have a soft spot for those folks. Stripe, Shopify, pulled the icons, added some margin data, probability when, um, this item was created. And again, One prompt, I can add some more, but I've like already skipped weeks and weeks of waterfall work across multiple stakeholders, multiple teams, multiple different functions. Um, and I can just show this to the customer and go like, hey, is this what you want? And if not, um, how can we improve it further for you? Third one. Last one. Uh, this is something that we've started seeing over the last week. And so definitely going to be a big trend heading into the new year, uh, is going from not just your quarterly plan in a static spreadsheet, but more of a quarterly playground of what your product's actually going to look like in the future. So what do I mean by that? We've all seen these lovely things. I spent a lot of time looking at them as product managers. They're useful. They have a place in prioritization, but they don't actually help you visualize how's the product going to work, how's it going to look when we do all of this stuff, how's it going to feel, do we want to head in, in this direction. It'd be great if there was some sort of playground where you could get a feeling for that. Um, and that's exactly how we're seeing customers start to use AI prototyping tools and Alloy, uh, as they'll take their plans. You can integrate with Notion, Jira, Jira Product Discovery, Linear, et cetera. Uh, you put those into your AI prototyping tool. Most of them now have integrations, and you're gonna get out an interactive prototype or a playground of your products that includes all of those roadmap items essentially built for you as a first pass that you can then start to iterate with. So you're sort of traveling into the future, getting a sense what your product's gonna look like when that roadmap is done, and then you can use that to, again, share with the customer, drop it in a Slack channel with your internal team, get a feeling for how your product should evolve, and maybe come up with some more ideas off the back of that. So what's the key takeaway that I want you to take home from this? Just do it. It's a whole new way of working, and it only takes a couple of minutes to try. And I think like all of the AI talks here today, the best way to get acquainted is just to get your hands dirty and give it a try. So again, doesn't have to be Alloy, can be any of these new AI prototyping tools that are popping up out there like wildfire. Uh, just jump in, get your product into it, start to play around and experiment with some of those things that maybe design hasn't had the resourcing for that you've always wanted to build. Uh, and oftentimes you'd be pretty floored by the results. So get out of this sort of archaic planning environment that caused us to leave our jobs to go build something new. Um, Say no to people saying, "Yeah, cool idea, but let's schedule it next quarter. We don't have the resources. We don't know if this is going to resonate." Just do it. Enjoy the new superpower, and think about how you can take advantage of this new future. Thanks so much.

Dale Williamson.

And then trying to score them to determine how they land. This is, by the way, workflows that—

Good afternoon. Great to be here. And so I'm Dale Williamson. I'm from Databricks. So Databricks is a private Series K company. Yeah, you heard me right. Uh, we have, uh, over $100 billion valuation about 2 months ago, 20,000 global customers, a $4 billion run rate. Of that $4 billion run rate, about $1 billion of it is attributed to AI. Uh, and the thing that a lot of companies don't realize is how many non-enterprises run on us. So there's just a screen, an example of those companies. Um, one of the things about us us. Our mission is to democratize data and AI, and one of the ways in which we do that is through the open source community and building and donating a lot of open source tools. So you might know Databricks via tools like Apache Spark. You might know about us from, um, things like MLflow. You might know about us through Delta, Linux Delta. You might know about us through Apache Iceberg or Apache Parquet. Each of these projects, the founders work at Databricks and they work on growing and scaling these communities. Um, most organizations, they struggle, and the reason they struggle is because, fun fact, there's about 400 proprietary database engines on the planet today. There's actually a cool StackRank on this. Now the problem that you have if you want to get access to that data is you have to pull it out, you have to transform it, and you have to load it into something else. So we do a lot of taxable plumbing. So the proprietary lock-in of data is a huge problem that we've been working to address. Like, how do we take that into a standard format? How do we make that standard and unifiable and accessible? And our definition of data is all data. 16.8% of data that processes through Databricks's 20,000 customers is images. Uh, there's a good non-trivial proportion that is PDFs and, and DOCXs. We even see email, we even see Slack messages, a lot of JSON. So huge numbers of different data that come in in raw formats and get converted into a more sort of unified open format and therefore is available. So we have limited lock in. One of the other problems that we see is everyone has these siloed security policies. These— I have security here, I have security here, I have security here. And what that basically means is, how do I navigate security to get access to data? The third one is, it is impossible to automate because you have to automate all of this plumbing in order to get access to your data. There's a standing joke that originated in the machine learning world of 90% of the time is just finding data, 10% of the time is actually building the ML models. We see the exact same thing with, uh, with different modern agents. Um, same problem. So imagine a problem where you have a fleet of agents that work for you and they can't access data because you can't access data. Do you see a problem? So the approach that we've pioneered is to unify the base through open formats. So how do we get all data into a common standard open format that's accessible? All data— image data, communication data, sound data, and structured text. How do we unify the governance layer that sits upon that data so that you have lineage graphs and you have access policies and you effectively have real accessibility, but it's still secure. And then how do we build composable agents on top of that? And what do we mean by agents is— our definition is an autonomous unit that has agency over a task. And we work at the task level because the context is narrow enough for us to get high quality. So we've started to build a series of bricks called agent bricks. And this is one of the reasons we did our Series K, so that we could actually invest in building these out. We're starting to see a proliferation of common templates, common templates of agents that can be built out and you can contain and maintain the IP. There's a famous Warren Buffett comment about about him owning a textile factory, and Charlie Munger came over to him and said, "There's a guy over here trying to sell us a loom." And Charlie— Warren Buffett immediately sold the textile factory, 'cause the value gets attributed to the loom, not the factory. So AI has looms. So how do we protect from the looms? How do we maintain the IP? How do we build IP? So this is our platform, and a lot of this is open building blocks. Blocks. We effectively are just a really radically fast toaster, and your focus is how do we do really fascinating things without having to build out all this commodity infrastructure. So 3 fundamental areas that are driving our impact today are how do we simplify agents in production, uh, that whole random sort of amount of vernacular that 85% of agents don't make it to production. Well, that's kind of obvious because it's science and we haven't quite proven this yet. But there's also a lot of other very, very fundamental primitive reasons as to why these agents don't run in production and how long they can run in production. So that's the first thing. Embracing unified analytics. So how do I give this agent access to all of these different databases? Databases. I need unstructured sources. I need to be able to look at historical archives. I need to be able to do time series. I need a transactional database. I need— these agents need to be able to store things to do work, access things to do work. And then the idea is to scale that intelligence. For us, data intelligence is about moving from very general models to building models that are far more specific to a task, far more specific to a purpose. So I'm going to walk you through what Agent Bricks does. And the whole basis is, how do we manage quality? What is the latest AI technique that I should have? Because new models come out all the time. DeepSeq just dropped a new model this morning. And there were about 12 models that broke out last week. How do you rapidly plug in and plug out so that you get the benefit of these new looms but also build IP on top of that? How do you balance cost and quality? So a number of these different factory of agent types are the things that we've been building out, which we call agent bricks. And these things, each of them effectively creates a scaffolding that different people can create different agent types and start to sequence them and stitch them together using sort of supervisory agent types. Some of them are text-to-SQL, so they basically plug in— we call that Genie Spaces— that plugs into your databases and actually can read and sort of structure information. Some of them are actual entire personas, like data science roles. One of my favorite ones is the data science, uh, the data science agent type, because it actually creates ML models for me much quicker. Some of them are around business functions like fraud protection. The way that we work is we have automated a lot of the plumbing for you while still giving you that deep visibility. So the first step is we create a series of judges for you that helps us to create guardrails. Creating an agent can be done in about 3 minutes. Creating an agent that works and continues to work takes much, much longer. There's a lot of evaluations that need to be built. There's a lot of judging that needs to happen. It's a bit like parenting. You have to kind of guide them and continue to reinforce how they work. And then you always have to be benchmarking the cost and quality. And it's a loop. It's got a lifecycle. The shelf life of an agent is anywhere between 3 and 6 months. So it's not something that you can write an algorithm, slap it in a mainframe, and leave it for 40 years and it still does the same thing. These things change over time, so it changes the entire management discipline. And what we're seeing is some very, very fascinating scaled implementations. So the idea that a company like AstraZeneca could take 40,000 documents on clinical trials with a team of 3, scale that to 1,400 medical reps in the space of 3 months, and continue to iterate and build those out. Out. So the reason I love this use case is, one, they're doing it very cheaply. Two, they did it with a very small team. Three, it is highly sensitive data that you can't get wrong. So forget this whole hallucination story. We need stability. evals is how we offset hallucinations. And the other thing is, how do we run this thing cheaply? And how do we continue to run this cheaply and swap out for cheaper models as new models emerge? The third thing is how do we collect the telemetry so that we can retrain? Open telemetry is going to become a really big deal over the next few years as you start to see how we need the exhaust fumes that come out of AI, and that effectively becomes your new moat. So let me just do a quick demo for you. I'll walk through what Agent Bricks does.

Okay.

So we have a multi, um, sort of multi-service agent. And what this agent is going to do is it's going to look at two little agents. Those little agents are looking at a whole RAG system. I want to look at unstructured data. I want to look at a whole bunch of data that's specifically, uh, to a purpose of feedback, you know, recommendations, you know, the staff and this is a bookstore that we're running. Now, what we also want to do is we want to be able to, in live streaming, look at sales data, look at supplier logs. And these are the two agents that we want to stitch together. And what we want to be able to do is ask some very, very simple questions and build this agent out. So what's super important, the GenieSpace agent, that plays a role in looking at my real-time supply system. I want to know that when I ask this question, the latest copy of a particular book, it's going to tell me when I can expect that and how— what I should recommend in return. Now, the problem with this is we need to have the real-time information from the transactional systems. We also need to be able to ask this question and know what the actual dates are and what the actual data that's coming out are. What we also want to do is then go to the recommendation system and start to look at that agent and get actual real-time advice. And what happens with this recommendation system is it's going to look at multi-sources. There's a lot of unstructured text. There's a lot of documents that it's going through, a lot of data that's stored in many, many different places. And then when you look at the output, you start to see this insane spill out a prompt response. Now imagine I could then go in and I could start to evaluate that. I can start a labeling session, and notice I'm not writing any code. I can then start to look at permissions. I can start to find SMEs in my business that can help me to answer this. I can then start to really build out what the actual feedback needs to look like. I can start to put in real feedback guidelines and instructions in a very simple way. I can write in there, please make sure that like some of the recommendations are like only 3 max so that I don't end up with this very vast verbose spillout because my agent, my, my people working in the stores are like limited in time and they want to actually have the best recommendations. I also want to make sure that certain recommendations are from actual staff in the stores so that you get that personalized touch. I can then start to bring that feedback together, merge it, reset my, my whole question and answer, and come back with the same sort of level of response. And what I end up with is a very condensed recommendation system. Now what I've effectively done here is very quickly bridge some incredibly complex datasets that you have to try and coerce together. Now, most of us have never seen this before, um, happen in a real-time system unless it's on Amazon or something like that. Most retailers take weeks to build this. And I can then go into the trace routing and I can have a look under the hood and start to build further evals. I can start to examine what's going on. So the key takeaways that I want everyone to get from this, and I wanted to save a lot of time for Q&A because I figured there'd be lots of questions. You have to ground your apps and agents in actual enterprise data. And in order to do that, you need the system to be able to give you access to all of those different datasets. Otherwise, you're gonna spend more of your time doing data preparation and less of your time actually building and making agents work.

Work.

The second problem is you have to unify governance, because if those agents are not going to work, or you don't have an audit trail or traceable data, or the way that you can actually demonstrate that this agent has the same access permissions as the person it works for, you're not going to be able to scale this. No procurement department, security department, or CIO will ever let you sort of scale that out in a material way And then the third one is this idea of investing in evaluation and optimization and having that ability to build a feedback loop in a way where you're not locked in to a provider and being able to scale that and have that open telemetry. I will promise you, you will see a pushback from people around the world about AI telemetry over the next year as they start to realize how valuable this is. Good example of this is if you look at what Cursor just did, shifting from Claude over to their own model. How do you think they trained that? What do you think they built it on? So thank you, and, uh, I'm happy to go into open questions for the next 10 minutes. Anyone around? Come on, don't be shy. There we go.

Thank you for the presentation.

Maybe I'm the ignorant person in the room. What do you mean by AI telemetry?

And that I own the IP and the code in your box?

So imagine you have— imagine, okay, when we built software before, Software always had trace logs, StackTrace. These are probabilistic systems, and what they effectively do is they spill out usage logs. So you have the prompt, you have the, the model reasoning, you have the steps it took. Where are you capturing all that, and how are you using that to reinforce and make your models more accurate? Because that data is the most valuable data that So where is it? How are you storing it? How are you capturing it? How are you turning it into making these agents and reinforcing these agents over time? So if you look at software, the past, we used to build functional software. We used to use StackTrace to be able to figure out support. We would answer questions, you know, and be able to figure out where the bugs were. These are stochastic, probabilistic. They don't always behave the same way. If you ever look at a kind of standard agent, ask it the same question 5 times, you'll get a deviation in answers. So how do you narrow that deviation? One, how do you start to make those models behave better over time as you gather more and more usage data? So this is where we shift from functional building software to behavioral building software. And the behavioral thing means we need this exhaust fume, we need this telemetry, and the telemetry becomes super important. And if that's not in an open format and accessible, you're going to find that there is a very, very hard thing about making your models work all the time, making your agents work. Another question? Okay, one more. Only shy person not in the room.

Um, you mentioned the success is having enterprise data. So if you're a third-party vendor trying to create agents onto other customers' platforms and they don't use Databricks, I mean, is that the entry point?

If they use Databricks, we can take their data and—

well, we're, we're not in every enterprise in the world.

I know, I know.

But if you do have Databricks or another third party, is that a good way to get into the organization's data?

Yes.

Okay.

So, so there's a series of fundamental building blocks that you need that are open standards. So everyone's been making a big deal about MCP this year. MCP effectively gives you some narrow context, um, data, uh, sort of protocols like Delta Sharing. They give you far more of those, of that kind of data, like real-time read-write. So if you think about, um, Salesforce and companies like Salesforce, they've been pushing zero-copy data movement. The under, under the hood plumbing that they use for that is Delta Sharing, and they use Iceberg and Apache Parquet. That means you're reading the data in an open format. You're not copying the data, but it also then means you have access to that data to be able to build it into your agents. Now, if you don't, if you don't have those those sort of open building blocks. It's a bit like when Linux took over in the, in the OS layer. Like, you couldn't run a Kubernetes cluster in certain places until it became a standard. And until Linux became a standard, you couldn't run it on DOS, you couldn't run it on an old-school Unix system. This is around the same thing in the data layer. So in order to be able to access, the conformity needs to be onto these open standards. and one of the things about the telemetry is if you, if you, if you use MLflow as a standard, you effectively have open telemetry for model lifecycle management. Uh, and then there's open standards like Unity Catalog, which has open APIs for moving governance data around. So each of these almost stack up and you can use all of them without buying Databricks. We just create a more management framework around it.

Thank you.

Delta versus Iceberg?

OK. We'll move on to that one in a second.

Great talk.

Thank you so much. You talked about auto-evaluation and auto-optimization. Can you explain a bit of role of open source research like DSPy and JEPA?

Are you utilizing this, or you are having your own optimizers and evaluation framework?

So, yeah, I think it's too early to bet. So you'll have noticed in one of those agent, custom agent types, we support DSPy. DSPy is awesome, by the way. Uh, I build a lot of stuff on DSPy. Uh, LangChain is also important. Like, so I don't think we're at a standard in terms of these protocols. CrewAI is another standard. Like, there's so many out there. Um, I think we're in this kind of fragmentation zone when it comes to agent orchestration. And we are nowhere close to that starting to ground out. In the lower layers, the data layers, the sort of storage formats, we've got a lot more kind of 1 to 3 standards. Uh, he's about to ask me a question on 2 of the standards in the access protocol, but in the orchestration, it's chaos still, and we are just choosing to support as many are that, that are open standards because we one of them will be the, the sort of horse that we'll bet on. Do I answer your question? Cool.

I mean, it was exactly that, right?

Like, how do you see managing all these different open standards, and how do you see Delta versus Iceberg?

And yeah, so Ryan Blue works at Databricks, he created Iceberg. Uh, Michael Ombrus created Delta, he works at Databricks. So we actually don't care. They have different communities and they almost have different purposes. So if you almost actually go look at the origin story, you know, Iceberg came outta Netflix, it was far more in the consumption layer of the stack. Delta came outta Databricks, it was far more in the translation layer. And actually the sort of redhead stepchild of the three is Hudi, which happens more in the ingestion layer. Suggestion layer. But I think those standards are getting increasingly unified by the open foundations. There's active work to kind of bridge the gaps. There's a lot of really clever things that happen in the one layer, a lot of really clever things that happen in the other. The more we can unify that into a common standard, then it doesn't matter. So my bet is in 2 years' time, no one's going to care.

Just as a follow-up, because you mentioned transformation, so what about things Kafka and Flink?

So Kafka, Flink, and Spark grew up together. Increasingly, that's starting to coalesce more and more. You know, Kafka is very much a movement. Do we even need data movement if now we can do zero copy and everyone standardizes on Parquet? Like, it's a good question. Like, data movement costs us a fortune. Like, copying data from A to B B to C to D and then doing the integration checks. That's an absolute nightmare. So do we need to do that? And remember my comments about the 400 proprietary database engines? Why do you need that if like you have Postgres and you have Parquet? So why? So the question will become, will Kafka have a future? I think it will for a long time 'cause we still have a lot of on-prem heterogeneous data. Will Flink have a future? Of course, it's very good at the stateless kind of rapid data transformation. Will they unify over time? I think so.

Thanks.

Anyone else? Great. No more? Well, I think I'm over anyway. Thank you very much for your time today.

Alright, now I'm curious. I'm curious, does London work in office or what? How many of you are actually remote? Okay, so we'll assume everybody else works in office and we don't need them here today anyway. Let me ask this, how many of you are fans or listeners of the 20VC podcast?

Okay, a couple.

Well, you know later they're gonna be recording a live 20VC. I think it'll be pretty fun. We love the banter between Rory, our dear Jason, thank you for having me, and Harry. But the reason I bring it up is not actually because I'm excited for later. It's because recently on one of the episodes, I heard Harry say, you know, my mom, she gives me a hard time. She says, Harry, you give away your secrets. You tell everyone what you do. Like, why do you do that? This is the thing that's your superpower. Like, why are you sharing all of that? And he was pretty He goes, "Mom, 99% of the people won't do shit about it." And that's the reality. So I'm asking all of you to be that 1%. That's kind of how I like to think about my career trajectory. I always listened in to the people that were doing things a heck of a lot better than I was and thought, "Cool, I can do that even better." So today we are gonna go a little bit tactical. You either left the last session last session feeling a heck of a lot smarter, like you knew nothing at all. I sat somewhere in the middle, but it was fantastic. Today we're gonna dive into some tactics, but also the actual layers of how I have built a remote-native leadership organization and grown a company from employee number 5 to 150. So I'm Marcelle Mooney, VP of Sales at Mango Mint. We are vertical SaaS for salons and spas, and not only do we serve an SMB market, I like to say it is very SMB. So our software alone is roughly a $4,000 ACV. We also have payment processing built in, so I'll talk a bit about that. But today I'm going to show you how I built a remote revenue organization with such rigor that we're creeping up on a 7.2x ARR. RR to OTE, which is unheard of in the very SMB market. So how many of you would be interested to know how I did that?

Cool.

Hopefully that's why you're here. So we're gonna have some fun today because I have seen into the future and I know a thing or two about AI. And if you haven't heard, AI will make you more productive, right? We've all heard it. Well, the reality is The reality is I haven't seen into the future. I have no fucking clue what's going to happen with all of this. But what I do know is that the majority of us could be a heck of a lot more productive and have a much cleaner layer of communication and data within all of our remote organizations. So we're going to pop into it because the reality is AI is here to expose you and it's going to show your entire organization what you're good at and what you're absolutely terrible at. And I'm sure many of you— let's see, who's in the room? Do we have founders in the room? CEOs?

Okay.

What about some revenue leaders? Any VPs of sales? Anybody know a VP of sales that you kind of like? Fantastic. Give them my number. I'll be hiring soon because what I've heard is that you're not promotable until you're replaceable. So give them my number.

Okay.

So what we're gonna dive into today is gonna be beneficial to many of you. But if you are a founder, if you are a CEO, it's probably gonna leave you wondering what the heck is your leadership team doing and how do you get more information to you without having to ask for it. So how do I actually know what's happening in my company? How many of you have asked yourself this? If we're being honest, hopefully all of you. It would be a little weird if it wasn't. But the idea that the revenue is growing, but we don't actually know what's going on, not necessarily in the day to day, but those things that keep you up at night. So my CEO likes to say, I only like to think about one thing in the shower. And I know if he is not deeply clued in to what's happening with win rate, what's happening on the layer of if he wakes up and notices that one AE's win rate has plummeted, he needs to know where he can go without having to send me a message. So this is the real fear, not productivity, but actual visibility into the organization. So this visibility gap has only grown since many of us not natively moved into remote companies over the past 5 years. Some of us are hybrid. How many of you are hybrid? You've got some remote, some in office. Okay, great. I'm opening a sales office in January, so that might be contradictory to everything I'm saying right now, but we're going to employ the exact same rigor around that in-office experience where we're incubating BDRs into our sales team that we do in the remote revenue organization. So we need a single source of truth. That's true. Nobody's really got it from what I've experienced. I sat at dinner last night with some fantastic people hosted by the lovely team at Momentum. How many have visited the Momentum booth? Okay, make your way over. It's great. We're customers. But what I've heard is that meetings are happening everywhere, especially in a hybrid environment. You've got some people who are in meeting in office, you've got some trying to Zoom in. All of this creates decisions that happen in people's heads, in those 8 inches between their ears, rather than on on paper. Tasks live across different tools. I like to say, if I walked into your office and I slipped you a note, you weren't in there, when you return to the office and you read the note, you're probably going to get back to me at some point or acknowledge that you've received it. If I send you a damn Slack message and I never hear back, it is very rude. So we've got to build rigor into how these updates go. Things are scattered. Knowledge isn't captured in any one place. So we've got to rethink the way that we build this. So I, of course, I went all in on AI. This is also why I've been asked back to speak at Saastr, because I'm pretty hardcore about it. I'm like, why, why would we ignore this thing? So I'm now 2 years into my AI journey into go-to-market, and I I realized that me going all in on AI was the actual problem my organization had. I installed everything. I was running literally like the AI kitchen sink. It was just like, what can we throw in? And what I learned from that is that the more AI I added didn't help me understand what was even happening. For example, this was the first year I brought in an EA. I tend to be somebody who's like hypercritical of every single move that I make. I'm thinking about things 6 months in advance down to the point of like what shoes I'm gonna wear on stage. So bringing somebody into that brain, it was more of a vulnerable state of like, oh shit, do I really wanna show somebody what goes on up here? And, uh, I have this moment where we're onboarding team members and I've I've got Fellow app from Chelsea trying to join. And then I've got this person sharing their granola notes. And I've got my notes over here on a notebook from 1999. So what I realized was, actually, what it made me feel like was that all of us were walking into this meeting. I loved this. Do we still call it a tweet?

What do I call it?

But Toby, Jason M. Lemkin mentioned this idea of showing up to a meeting and you don't know who's joining what and what's triggering this or that. And honestly, it was embarrassing to me that my internal team couldn't figure it out. We're paying for these tools and we're not using them with any sort of disciplines attached. So how does it start to feel? Well, like this. And the problem wasn't the tools, it was the fragmentation of how things were happening. Now many of these tools have only been in our hands for 14 months. So I'm here to tell you that these things move quickly and you've got to move even faster. How many of you know about the Red Queen's effect?

Okay.

What about Alice in Wonderland?

Ever heard of it?

Seen the movie? So there's a scene in Alice in Wonderland where she is running and she is running so fast and she is racing. Her heart is beating and she realizes the trees are moving right there with her and she hasn't gotten anywhere. So the Red Queen comes in and explains to her, that's because the trees are moving with you. So in this day and age, you've got to run twice as fast as you did before to keep up or even get ahead of where you thought you wanted to be. So I cut. I started cutting. I knew I had to debloat.

Very delete, delete, delete.

The path to rigor, as I like to call it, is always going to come from subtraction. It's gonna be, you know, again, we're talking to the person who folds their socks, so might not be for everyone, but I digress. I go full ruthless. Some of my team members like to call me the pit bull. I go ham. We consolidated everything. Into one operating system. And today, I'm gonna show you that new reality. We're gonna dive into the layers of how I created visibility. And this is really where we started to see that everything changed. We have increased across our team our win rate 7% in 2 quarters just by creating complete clarity around how to access access the things that you actually need to be disciplined within the organization. So everything flows into the systems where we work. I don't know who said it first, but I was recently listening to a podcast with Kyle Norton. He's got the Revenue Leadership Podcast, and he mentions— he repeats this all the time, but it's no new surfaces. You're going to walk around this floor and you're going to be be very turned on in the way your brain is lighting up about the new capabilities we have, the things we can do. And what I urge you to consider is what new surface would this introduce, especially— bless you, allergic to new surfaces, so am I— what new surface would this introduce to someone who is attempting to actually produce revenue in your organization? The main thing. So I like to say let's keep the main thing the main thing. Earlier I mentioned what are we able to do? We are on the path in very SMB, like I said, an ACV of $4K and we're able to create reps that are producing 7.2 ARR to OTE. Is that good or is that great? Thank you. The AI rigor stack. This is what I'm gonna introduce. This operating model is all about first analyzing where we work. So for you, if that's in office, that's where you start. For me, it's not in office. In 2020, I was employee number 6 at MangoMint, and we sat in this dinky little office where I annoyed our co-founders, the 3 of them, 4 of them at the time, and basically ran demos while they were trying to build the product, and finally they locked me into a little room, and I realized, you know what, I'm not gonna commute to this office anymore. And then the world shut down and none of us did. So from employee 6 to employee 150, we've been completely remote. And what I realized is I had to recreate the idea of an office for people to understand how to integrate themselves into that type of environment. I'm seeing in even more now with bringing up a team that's fresh out of university coming into the sales world, coming out of sales programs, and I'm going, are you sure you were in a sales program? Because I don't know what's going on here. So that is why we're opening a sales office. But I digress. The operating model needs to have 3 layers, and eventually these will be shared out. Uh, you're also welcome to shoot me a message on LinkedIn, I'll send them over to you. But the clarity layer is all about the playbooks, the definitions, the things those ICs need to know, but they're never gonna know inside out and backwards. So they need to be able, for some, with a click of a few keys, for me, by dictation, to access everything that I need here. Number 2 is the cadence layer. What does that mean? That means when do things happen? Happen and how do we know when they're reliably going to be happening. So these are your reviews, your team updates, the rituals that make your organization connected. And finally, my favorite, because this is where I've been able to give my reps 16 hours a month back, the Copilot layer. So that's your AI triggers, your automations, the agents you're employing within this stack. So Now we're going to dive in a little bit deeper and I'm going to show you like literally what it looks like on my day to day. So I'm in a meeting. We're recording this meeting, right? We're transcribing it whether you know it or not. And then from there, that meeting needs to go somewhere because I said I need to complete my slides. I'll have my EA Taylor follow up on that. This is where it goes. So working in Notion, that has become the main surface of my work. This is where my team's playbook lives. This is where the revenue organization's playbook lives. We even have our leadership playbook here, our handbook here. Let me tell you, 4 years ago, I thought Notion was for nerdy bullet journalists who like to meet up on the weekends and talk about Notion. Today, if you— you couldn't pry it out of my cold dead hands. So when I'm done We're done with the meeting. By the way, I was chatting with the, the Notion agent this morning, and soon I hope they will be able to automate this. This is still manual. I do have to dictate, tell Notion to add this to my database, but this meeting information is going straight from my meeting into the database. What happens next? So the meeting's over. It now lives in that database. And let me tell you, I didn't build any of this. The agent built this. So this is about saying 2 sentences and getting something done. But what's beautiful is immediately post this meeting, which I laughed because my slides were very late in my opinion. So when I made these, I was like, let's remove the timestamps so Amelia doesn't get too mad.

Just kidding.

So we've got that task. But if you look on the far side of the screen, you can see these notifications. I did not set any of these notifications. I'm simply telling the agent or talking in the meeting saying, yeah, I'll have Taylor follow up with that. I don't have to take another action. By the time this has processed, added to the database, I will often hop out of that meeting, check in with my EA at the end of the day, and these tasks are done. It's not just because the automation has pushed the task for her. What happens in the next couple of slides really shows how the richness of this type of database is important. So rest in peace, Todoist. I never used you anyway. Everything lives in Notion, and we've got tasks created. What's in progress? What's the priority? Priority. All of the notes are automated. What's really important though is that we have more richness to the data. So let's say I set a task for my EA based from the meeting. We need to follow up on 2026 incentive planning. How the heck does she know what 2026 incentive planning even looks like if she wasn't in every single one-on-one or she hasn't spoken into those revenue leaders that are working on their incentives for 2026. This is where we go back to the layer all about cadence. So again, we talked about rigor. Rigor is how you handle the disciplines you have in place. So one discipline we have in place here is our weekly updates. So within Notion, every functional leader from every part of the organization, not just the revenue organization, post a weekly update. This can be a template that you create. Of course I created my own, but it can also be what works best for them. What are the main things? If they just had the main thing in that update, I'd be okay. As long as the main thing, if that gets done, doing that one thing will then eliminate the need for all the rest of my questions. I'm good with that. Like that. Sometimes they're very lengthy, but either way, the access for this is granted across the people that need it, which would be Taylor, my EA. So she can now chat with her agent on that side to understand the task at hand, the context at hand, and be able to take action autonomously without me involved. So what does the Notion AI agent do for me? This is going to access to my updates, it's access to my meeting notes, it's cross-functional communication. So we use Notion, you know, if you've got a Notion inbox with 30 unread messages, there's a major problem. This should be your main push-through of notifications if you've got them. This is going to be my task management, and it's also going to be where we escalate emergencies. So finally, let's move move on to that final piece that lives in that communication layer. This is where we have our decision changelog. This is something that I find gets missed heavily in remote organizations. Somebody changes something in the product, and then one person trains over here. Four of the AEs have actually read the product changelog. A few people showed up to the lunch and learn, but the other 50% of the organization has no idea where to go to see those internal changes. This could be something in the product, but if you take a look at here, these are real things that I've put into the changelog from my original. So compensation plan updates, and then access is granted for those who need it. So you can access what you need through the AI agent, but nothing that you don't have permission to access. Access. The decision change engine has been a game changer for us because it's this idea, you know, I'll never forget in my first job, it was actually with a British man. He was a hairdresser. I was a hairdresser for 15 years and I trained in a Vidal Sassoon salon. How many are you familiar with Vidal Sassoon? Rest in peace. So in that salon, I'll never forget at asking him a question that was in our handbook. And I mean, let me tell you, this handbook was like this thick for a hair salon. He was the type of man that as he's parting the hair, my job was to hold a section with one finger and to stand far enough away where I did not get in the way of those scissors moving quickly. And if the finger slipped, I got hit with a comb. So I'll give you that much. Loved that job, learned so much there, but the one thing Something he told me that really stood out is, "It's not my job to tell you, it's your job to know." Sounded harsh in the moment, but I opened up that dusty handbook and there was my answer. So this decision changelog needs to work the same way. This is us providing the information that you need to do your job damn well, and it's expected that you go there should you need this information. So remote companies don't die because people aren't working. They die because the decisions and the changes aren't communicated clearly, and your team does not like that feeling. They want this information. So here's the total stack if we zoom out here. Something that needs to be true, and there are founders in the room. If you've got shitty data, your job is to go fix that and to find the people you need to make it better. The biggest challenges I hear from RevOps, from CROs I speak with, from our own head of data when he was looking at joining our company, was purely around the actual integrity of how our data was set up. So we had the benefit of from day one having really beautiful data infrastructure. So that top layer is the deep, knowledge, all of the communications, that Slack and Notion. But when we go down into the system of record, we've got Salesforce for our sales team, we've got our playbooks, tasks, follow-ups, and Notion. But the data layer and the analytics are so important because none of this can run if people don't have access to the information they need. So Snowflake, which we are using the Snowflake intelligence, we've got all of our our revenue, customer data in one place, very clean. We do all of our enriching, our entire TAM database is in Snowflake. So all of that's living there. And then we've got Sigma. How many of you are familiar with Sigma? So Sigma, a BI tool that allows you to take this, put it into dashboards. We build everything there. I'm actually quite allergic to Salesforce reports, sorry. But my thing is we really need to live live and work all in the same place. Use it for something quick, but let Sigma be where we build out, and let the source of truth from our data be that Snowflake layer. The automation layer is where a lot happens. Uh, I mentioned everything in Notion, but it also is Slack. Slack is the second surface and the most popular surface for our ICs and for much of the revenue organization. But I also popped on Momentum here. This has become a new layer of automation, but it doesn't provide a new login. All of the work done through Momentum pushes into Slack to keep us working in the office. So this is the whole system. Slack and Notion are where we live. Salesforce is that system of record for the accounts, you know, We could go into many paths here, but to the point where in Salesforce I don't even use opportunities. I said that to a RevOps leader last night and he just about keeled over. But, you know, I realized for an organization like mine, our sales cycle is 5 days. We've got our sales team selling, our onboarding team onboarding on day 3, our sales team closing on day 5, our onboarding team taking the next 1 to 2 weeks to complete the onboarding. I needed all of that in one place. So we've built it out custom on the account so we can see everything in one place. So I really think deeply about the way you look at your data, the way you consume your data, and to be obsessive about it. And if you're not obsessive about it, find someone who is and bring them in. So the magic is here. All the insights start with the AI prompt. They get pushed to the team. ICs do not have to go hunting for a dashboard. All of that data comes to them through Notion and Slack. This is the way that the information actually flows through the teams. So product and revenue data that's living in Snowflake comes into Sigma. I can build in Sigma. Our data team builds in Sigma. I don't love to build in Sigma, so I have a list that goes to that team. They build there. Everything that needs to be pushed to an IC then goes into a Slack alert. One example would be card fails. Where does that live for you right now? Does anyone know? Okay, interesting. I know exactly where that lives in my organization when a card fails, something doesn't go through, you know, the invoice is past due. This is going to push directly into Slack to the person person who needs to see it that needs to follow up and handle that. Notion is also that next surface where for data flowing items will end up in Notion that need to be. Customer call data, this is Momentum, this is Snowflake, Slack alerts automatically to Salesforce and Notion. This does not take a human to make this data flow. This is all automated. The execution flows happen with those Cadence layers. So if we have an internal meeting, that's going into Notion. Somebody posts an update, that's going into Notion. I can literally have a conversation with my AI agent to understand how this layer is flowing and what emergencies need to be brought to my attention. Finally, should anything need to be pushed into an executive summary, we do use Momentum for that on the revenue side. We also have that built out to where from Notion or Slack, you can have that information float to the person that needs it. In our company, we like to think of data and context flowing downstream automatically. There should be no need for this thing you cannot find. Uh, and reps don't need to go into dashboards. I cannot repeat that enough. They will not go and find out what their win rate is you need to push it to them. So where the team actually works, I thought this could be useful. This is kind of starred down the majority of the team. This is kind of secondary to their core function. So if I'm a support team member, I live in Intercom. However, most communications, alerts, escalations, we're living in Slack. That's the office. We show up to Slack. Notion is going to be where work happens. So tasks are completed, decisions are made. Salesforce, system of record. Reps are in there, I don't know, 10% of their time. So very slim there. Snowflake and Sigma, very rare for most of the team. But even for the leadership team, this is generally where we're going because the data we need that is pushed to us isn't there. So recently I decided to look at a trailing 30 and trailing 60 day. I didn't have that on our win rate, so I did have to go into Sigma. We did have to build something new, but it's not common. Okay, Slack as well. I'm gonna dive into this just a little bit because I mentioned earlier, I put a note on your desk, you don't get back to me, that's rude. Rulebook on how to use Slack. One of the first things you'll onboard into my team is what happens when I send a message at 2 AM. Well, I've had many suggestions. How many of you are like, "Hey, for when you get back." Okay, good, 'cause I won't spend the extra 2 seconds to say that. I also don't schedule send because that takes time to hit I don't hit schedule send, and half the time I end up hitting enter anyway, so it's all screwed up. Here's the deal. When I onboard team members, I teach them how we work. We work when we work, and if you aren't working, go ahead and set your notifications off. We're all adults here. You need to manage your time, your energy, your attention. That is not my job. So having this type of rigor around the discipline you expect needs to be explicit. If you are assuming and it is not explicit, it will not happen. So for that fast async using Slack heavily as the office, you've kinda gotta think in terms of that and then communicate that to your team. So overall, this whole thing is about being really obsessive with having that high level of discipline, really low low headcount for the volume that you're doing and maximum leverage among that. So I mentioned, you know, we're roughly at 150 people in the organization. They always say organizations break there. I've felt that. This is why I went so hard into this. And over the next year, we've employed a process where we're not going into 2026 with a headcount spelled out. I'm bringing on 2 AEs here, 2 AEs there. We've decided to to be completely selective in our hiring process. So using these types of tools to ensure that we can maintain that ARR to OTE ratio that I mentioned, how do we actually have conversations among the leadership team where everyone understands the information necessary to select team members to add that really make sense, that make sense for the business make sense to the main thing. So with that, a little takeaway, some things to think about. You wanna go back, you wanna run an AI rigor audit. What does that mean? Rigor is how we handle the disciplines that we've put in place. The disciplines are the actual actions we have in place. So where do your meetings live? Do you have some people using Zoom, some people using Google Meet? How do you condense to sense it. Now, in an enormous organization, think of it by function and break things down. Find the blip in the radar that you don't know anything about and keep zooming in until you figure out what's going on there. Where do the decisions live? So we try to make as many decisions asynchronously as possible and only have a meeting should it actually be to sum up the solution. We're not going to sit in a 2-hour meeting just like hashing through things that we could be reading or dictating to each other. So what tool owns tasks? Are your updates consistent across your leadership org? And are you using AI to enforce that rigor amongst the organization? If the answer is no to any of these, I'm here happy to answer questions for the next few minutes. Thank you. We have a mic if anybody wants to dive into a particular, particular question. Or we can go eat lunch too. Go ahead.

Hi. Thank you very much. This was really useful because again, I run a fully remote team. And again, obviously that issue around communication where you've asked the question and it takes forever to get feedback in. And again, you're trying to like reinforce that. Um, just kind of curious about like how do you— again, a couple of really interesting things started coming to mind when you were talking about maybe getting to your point around ensuring that people maybe change their status that they are not in, they're not available if they're going to be away from their desk. Because again, sometimes end up calling the individual like, oh, I stepped away to get something done. But how do you kind of reinforce that culture where there's a lot more like responsiveness, especially around that communication? Because that's the entire breadth of the organization, actually, when you are fully remote in that sense.

So if I'm hearing you correctly, you've got a completely remote organization at this point. You're already witnessing where you've got ICs who are not available when you assume they would be. Maybe they're on a customer call, but you're not getting the responsiveness that you expect. So like I said, once you've established something, that becomes the status quo. So the two things that I would look at is first, change management. So you're going to have to create a change. What does that change look like? If, if I were me in your organization, I'd probably I would probably take a couple dark days and I would just plan out what do I expect. Because anytime we have an expectation, if it's not met, it's probably because we haven't been explicit about what that thing is. So the first thing I would do is sit down, like what is my actual expectation? Well, in my organization, all calendars are public. Even mine. Two ICs, they can see if I'm free or if I'm busy. 3, I will respond to them, even if it's a, hey, I'll get back to you in a few minutes. For the ICs, I can see every customer-facing call they have. This was extremely important when we were really small, and now we have pod managers that have a pod of up to 8 ICs. They need to be able to see their complete calendars. Again, this is definitely specific to the high velocity we run. Our AEs are doing 8 demos a day. Close a deal to 2 deals a day, like they have to be able to be rigorous and like really know what's going on with them. So first thing I do is audit calendars and truly understand, like I've done the math. I know to close 20 deals at my organization, it takes roughly 4 hours per deal max. And these people, 30 hours a week, they can get into their quota, but they've got to be disciplined about it. It won't just happen. By chance. So I know the math inside out and backwards. I know what the calendar should look like. I know what responsiveness should look like. And then I write that down. This is all written down. This is now documented. At my team, one of your onboarding courses is going to be Slack hygiene. What channels should you completely mute? I have all of our social channels completely muted. Occasionally I'll unmute I mute like the dog channel. That's cute. But then I mute it again because I need to be able to focus on the main thing. I do the same for my team. I teach them how to mute 55 Slack channels. Then I go in and I say, these are the channels that expect highest response, especially if it's like async DM channel, that doesn't need a response right away. If I DM you directly, that's because I am walking into your office and barging through the door I'm like, hey, here's this thing. If you're not there, you're not there, but you're gonna get back to me in a DM as quickly as you possibly can. So I'm writing that out. I am literally making this explicit, and then I'm creating that change model where we have a database. Here's the change that's going into effect. This type of change, I think you need multiple touchpoints. So I think you need to write it. I think you need to publish it, and I think you would have a meeting with all of the people who are involved in that change so they understand explicitly what that's going to look like going forward. And then you have to enforce it. That's what I would do. What else?

Hi.

It is very nice to listen to you.

I'm a founder of a software service for marketing technology, and we work completely remotely. And we don't use Notion. I don't much know Notion. One of my colleagues brought it into the organization, but we use usually Google, the Google Workspace. Yeah.

So my question to you is that what is the extent that you use Notion?

How is that your everyday workplace? You know, again, this goes back to, we hired a COO a little over a year ago who was an investor of the company, fantastic man, had come from Fleetio, and the first week on the job, he sent our CEO and I an email, and neither of us got back to him. We both just slacked him because we realized we hadn't taught him where we communicate. So it really does have to be this thing that you onboard even the executives and even the leadership, and you make a decision. Now, I can also say our Our CEO did not choose Notion. You know, we had to like pull him, we were dragging him into Notion. And like I said, 4 years ago, there's no way you could have got me to use Notion. I thought it was so annoying. So I think it's about finding that champion within your team that deeply wants communication to flow and that is going to like run this for you. For us, that was Iris, who is on our VP of customer operations. She knew information needed to flow cross-functionally better. So she championed Notion, building it out, making it our hub for all things. And then once it got good enough to where I could just literally talk to the AI agent and I don't type anymore, I literally dictate with Whisper Flow everything like I am. That's how it needs to work for me. So once I realized how easy that was, that made sense to build everything there. So I think it's find the champion in your org that wants to do this. If that's you, find the tool that works well. Today there are occasions where somebody sends me a link to a Google Sheet. Sure, I'll open it, but every time I have to go search for it, I'm cursing. So I usually then copy that to a database in Notion so I know I can search in Notion to find that Google Sheet. Otherwise, there's there's nothing happening in any other parts of the org. Again, when organizations get large, usually this kind of breaks down by function is what I've seen, and there are different functions and different tools. Of course. What else?

Sweet.

Well, with that, I'm happy to hang if you guys want chat in person. Otherwise, um, I am on a panel with some fantastic— Amelia from Saastr, Ashley from Momentum, um, and we just added one of the new AI team members at Salesforce. So it should be a good panel around 2:30. Hopefully I'll see you there. Thank you.

I think— okay, great. Thank you so much for the microphones. Hi everyone, thanks so much for joining us today. My name is Todd Garcia, VP of Client Success at Cleverbridge. We've got about 15 years of experience here on the digital commerce space. My team and I focus primarily on retention, strategic growth, and Optimization. And I'm joined by Megan Hemp. I'll allow her to introduce herself.

Hi everybody, my name is Megan Hemp and I, uh, lead digital transformation for PTC. For those of you who don't know, PTC is an industrial software company primarily in the CAD, or computer-aided design, and product lifecycle management space, among other things. Uh, PTC is roughly 30 years old, so we have a lot of processes, uh, manual processes in place. And before we fix those, you know, we were looking at AI and that, that just didn't make sense for us. So looking forward to this discussion today and how we're navigating that.

Yeah, great. Thank you so much. So obviously we're talking today about how AI won't save a broken sales process. And I know this is an AI conference, so it's kind of like, hey, maybe this is a bit of an anti-AI presentation, but I promise you that's not what this is. The idea here is talking about what the future of AI needs to look like and what you need to do foundationally to make that future as beneficial as possible. So Cleverbridge, coming from 20 years of experience in the digital commerce space, has a, has a lot of things to say about that. And I think one of the key things that we've seen over the past, past several months is that the AI market itself is becoming incredibly saturated with solutions, and these solutions are there to improve processes, optimize margins, and ideally create a, a better customer and employee satisfaction score for, for you. And I think you've, you've met a lot of the people here. You've heard about this. Despite the fact that we are in the early stages of AI, the value of these solutions are real. Absolutely. But the crux of what we're talking about today is that, you know, that alone is not going to allow you to see the transformational change that you want to see with AI. The, the usage of AI by itself is not going to fix processes that are already broken. And in fact, as we've seen, you know, AI can sometimes even scale broken processes if not managed properly. And so for Cleverbridge, one of the key things that we've identified and that one of the biggest issues that we've come across is that sales processes, traditional business-to-business sales flows are not as optimized as they could be. And one of the key things that is driving a change within that is actually the customer experience. And this is really exciting because AI needs a foundational element to be built upon in order for it to provide the most value that it can. And we're seeing that actually the industry is shifting in such a way where consumer preference is actually driving some of these core initiatives, where traditional B2B buyers are looking to actually spend more online, engage in a more traditional sort of B2C digital-first buying experience. And if you marry that with the actual software provider, PTC in this example, their decision to want to expand faster into new, to new markets it's at a lower cost. You know, there's a really interesting sort of marriage between the two of those. That all comes down to, I think, the, the core element of what we're talking about, which is that in order for these companies to see improved margins and manifest sort of the data accessibility, which is not available to you in a traditional multi-tier sales process, you need to consider a digital-first buying experience to create a more optimize internal and external engagement with your customers. And so to talk a little bit more about that, I want to actually turn it over to Megan, and she can give you an idea of what PTC themselves experienced and what brought them down the path that we're talking about today.

Yeah, thanks, Todd. So as I mentioned earlier, PTC is roughly 30+ years old. And with that, and as well as the size, it is an enterprise industrial software software company, they're— it's very, uh, human-led, uh, process-heavy sales cycle, which works. PTC, uh, does about $2.75 billion worth of revenue annually. So obviously something's working, but all of a sudden it wasn't. So what we saw is we had high-intent customers coming to us wanting to buy a couple licenses, do a quick renewal, add a couple licenses to their, their stack. And we would say, great, absolutely, can't wait to help you. Let me go get you a quote. And I have to go get an approval and I have to go get another approval. And what did you need again? Oh, maybe you should go talk to one of our partners. And then the cycle would start all over again. And that obviously is not a great experience for our customers. And so this is really where we started to hone in and say, what are we doing here? How how can we better serve these customers? And I think Todd's going to talk a little bit about why digital first requires more than just an online checkout because simply putting the button on our website isn't going to fix the problem. There's a lot that we had to untangle to get there. So yeah, I think you can take that next.

Yeah, absolutely. So we talk about digital first and it becomes, I guess, almost a buzzword or buzz phrase is this idea of what does really digital first mean. And there's a lot of solutions out there that allow you to go digital very quickly, right? I mean, Stripe gives you a couple lines of code and suddenly you're selling online. But the reality is that actually doing a digital experience, especially in the international market, is a much more complicated and potentially risky affair. So you can kind of see here your traditional iceberg where you've got the tip of the iceberg and all the other work that you do behind it is we've got taxes, we've got that, we've got compliance, we've got PCI management, regulatory management, and there's a lot that can go into it. And all of this is also data points that you, your company, you need to manage, or you can leverage to be able to optimize certain experiences. And so we see that if you're not taking a foundational and future-facing look of how to sell digitally to your customers, you're going to be missing out on the long long-term solution. So again, fast is not always better, and the key thing here is for you to identify a, a solution that you can actually grow with. And when I say solution, I wanna be very clear, I'm not just talking about technology. I think solution here is the marriage or the coupling of technology services, whether that be internal and external, and then also expertise. So the idea is that you either need to build this and maintain it, it or find a partner that can support you in this. And the more that you do, the more that you take internally. And I mentioned this earlier, is this sort of foundational data and margin optimization that you're getting, the more that you can leverage that data to take advantage of the future of agentic commerce. You know, there is AI that exists today with regards to B2B buying, fraud management, payment optimization. But the future of agentic commerce is kind of unknown at at this stage. We don't know how it's going to interact with B2B, but we do know that if you don't have the data, if you don't have access to those customers, if you're not taking all of this as part of the broader picture, you could theoretically be missing out in an area to optimize and expand. And I think one of the key things here when you talk about the difficulties that PTC ran into is, is talking about how do you take a step back? And I want to turn that over to you, Megan, to talk about what you did there.

Sure. We— and to hit maybe Maybe briefly on that, the last iceberg slide, we also went through the, should we just build this ourselves and do it internally? Which I mean, was sort of a non-starter for me, but we had to entertain the conversation. But the amount of like additional people we would have to hire to maintain our own in-house built processes was just unfathomable. So, uh, going with a, a provider like Cleverbridge really helped us to scale quickly. So when we took a step back and like looked at what we were been dealing with, there was a cohort of customers who we wanted to target for digital-specific sales motion. And so segmenting our customers was really where we had to start. And that doesn't mean that only those customers are who can transact with us online, but that's who we want to go after because there's a segment for them, there's a segment that should be serviced by our partners, and there's a segment that should be serviced by direct. And those are not hard lines. They're pretty soft. We started from the basics and we're still, I'd say, at the basics of just trying to get a baseline for what digital commerce looks like at PTC. I think we're almost a year into our journey, I'd say, and starting out with that simplification process has helped us a lot. The structure automation and AI is where our partnership with Cleverbridge came into play and being able to have one list price that we're using for our customers, being able to unify our quoting and payment at the sort of, you know, snap of fingers was great for us. We also were able to remove a lot of friction that the customer was experiencing with self-service portal within Cleverbridge as well as automated renewals. And so it's a one-stop shop for that customer that's like, I just need one license, please let me buy a license from you. We have that now, which has been really great for us. And then I think we're now getting to the point where we are able to scale and build repeatable processes beyond that. We have through CloverBridge payment recovery and fraud screening, which we can show on the next slide in terms of the what it looks like and then renewal triggers as well as recommendation models for our additional products.

Yeah, just sort of alluding really briefly to what Megan was mentioning, you know, this is the PTC checkout where you can go online, purchase a PTC license, give your details, understand what you're getting, sign up for a subscription model. CleverBridge then supports multiple currencies, multiple payment methods, the actual lifecycle of that customer as well. And then after the fact, you have the ongoing engagement with those customers, trying to automate that experience, identify when is the upsell potential, when is the cross-sell potential, do they need additional licenses, are we managing consumption billing selling to a certain degree. So there's a number of things that come into it. And again, we keep coming back to this idea, the more data that you own for these consumers, the easier it is for you to optimize and automate that experience. And so we're really excited ourselves at Cleverbridge to kind of see how the world will change with agentic commerce, specifically in that enterprise space. Uh, because the more that we learn, the more that we're experiencing, we're bringing these things to market for our, for our core clients. Before we sort of get to the Q&A, though, I think it's important just to talk about what PTC has learned and what they've seen.

So I'll hit on some of our big metrics. I'd say, like I said, we're coming up on a year of having this live and we've seen a 25% growth in year over year for e-commerce specifically. We have automated renewals now through our digital means, which is which is, like I said, we didn't have that before. And then faster quote to cash. So all of these things have been material for our improvements in this space and meeting the customer where they're at and what they need. But I'd say even more importantly are some of our lessons learned. And the first two go hand in hand, and that's really about sort of the organizational change management of it all. We did a lot around helping our partners and our sellers to understand that this is additive. We're not trying to take anything away from this. And so I don't want to undersell how important that was, the organizational change management of it all. Additionally, making sure the incentives were aligned for both our direct sellers and our partners. And so we have things like a partner referral program in which they're able to get commissions through our digital sales motion as well. And then lastly, you know, you can't automate ambiguity. We have to be really focused about what we were going after and what portion of the of our customers this made sense for.

So really it boils down to what we've identified as this digital transformation checklist. And I think Megan pointed out a couple of key things here. Your stakeholders, your incentivization. It really does come down to change management. You have to know what you want to accomplish in the long term and then step by step approach the process to be able to to take advantage of what you're laying in front of you. I think from our side, we think it's fixing your data. That's a key aspect of this. Identifying what are the KPIs internally as well. I think this is just sort of this— the business case side of this. One of the best opportunities with this is margin enhancement. It's increasing your margins by reducing complexity and multiple layer engagement. And so really understanding what do you what do we want to see, what defines success, and then tracking that throughout the lifecycle of all of this. And then again, you know, the concept of applying AI after you've done this is important. There are opportunities for AI throughout this entire process, as I mentioned, fraud management, payment optimization, you know, potentially even some of the people here are selling AI-driven SDRs, BDRs. These are all value propositions that can feed really nicely into a digital-first buying experience.

I agree.

Yeah, I know. I think for me, I feel like we're kind of just at the beginning of this journey in terms of taking this and scaling it with additional AI. So it's been really helpful to sit in some of these additional sessions to see where some of this can sort of plug and play. But I agree, I think that Cloverbridge has been a great partner in all of this and a thought leader. But I don't want to underscore the organizational change management aspect of it as well.

Absolutely. With that in mind, thank you so much for, for hearing us out. We appreciate you sitting, but we have some time, so we'd love to answer any questions that come up. If you'd like to ask Megan or myself anything, please feel free to do so. I don't know if there's a microphone for the questions over there. Oh yeah, great. Thank you.

Does that work?

I don't know.

I don't hear it.

Okay, so, uh, My question is, you say 83% of enterprise buyers are happy to go digital first. Megan, the, the segments that you tested this with, were they enterprise buyers or was it the, the lower end?

It's a combination of both. So I would say we're targeting the like lower end, but we have enterprise buyers that are also interested. It's all about what makes sense for the customer, because what we see at PTC, as you imagine, our enterprise buyers tend to have more more complex needs. And so we are purposefully not putting some of those offerings on our, like our web store because we want to have a conversation about that because there's potentially more opportunity there. That being said, I do see a future in which we would work in tandem with an online buying motion as part of those conversations.

And so how do you triage that process? So how do you use AI to know, okay, this now needs to go to enterprise, needs to be dealt with by buying enterprise?

Yeah, it's through our prospecting motions primarily. I think even today we've seen a couple enterprise customers come through and we flag those for our AEs to say, hey, FYI, they just bought a license of like our training on the store. And so that's like less of a concern. When we start to see them buying other things, we're like, this may not make sense. You may be getting better terms if you go direct. So it's not a black and white thing.

Yeah.

Thank you.

Thank you.

Hi, I had a question. Did you experience any cannibalization in your revenue doing this?

Did you see any cases where that persona previously would have paid a larger fee and now because they went digital it ended up becoming a small deal for you?

No, in fact we were looking at transitioning some of our direct renewals to e-commerce and our our average difference from list price is basically 0%. So meaning there are people that are paying more and there's people that are paying significantly less because they haven't been brought up to list for whatever reason. And so this is almost a mechanism, a forcing mechanism to get everyone at parity.

I think one— while we're waiting for another question, if there is one— one comment around the 83% We also see that there's some statistics on what people are willing to pay for online. And I think most people are saying $50K and below is that nice threshold right now. But Clover Bridge works with a number of enterprise businesses and we do see $100K and plus transactions through our system because it's not just credit card, right? There's other payment methods that exist that are more enterprise focused and those are supported by some of these digital-first models.

I also think that it's like a larger cultural shift that we're just seeing to change. I mean, you can imagine your average engineer is probably not like super gung-ho not to talk to someone, but then people that are, I would say, more digitally inclined don't want to talk to someone. So you're seeing a mix of both.

Yeah, absolutely.

Thanks for the presentation.

So I heard a lot about direct customers.

How do you handle the channel partners and are they part of this program?

They are.

Yes, our channel partners partners are part of this in two different ways. The first is as a referral partner. So basically as like an affiliate, they have links that they can use to send to customers who would otherwise not be particularly valuable to them from a growth perspective. And they can get a one-time commission if they send that to a customer to the store and the customer purchases. And then the other part is we're transitioning some of the lower value renewals to partner renewals to the store. And that's done through like an opt-in program. The customer has the choice at the end of the day as to like what, how they want to work with PTC. If they want to go through the partner, that's fine. But we're trying to, to offload that low value, high churn population to e-commerce.

Yeah, I love the— it's, it's about customer choice, right? It's customer preference, meeting them where they want to buy. And again, that requires a more complex ecosystem to be able to differ from a traditional model to a digital-first model and back again. And so you have to have, again, the solution that we talked about earlier is not necessarily an individual partner, but also understanding that it's more than just technology. It's about the entire experience and expertise that flows into that as well. I think we have time for one or two more questions if there are any.

Hi, so I have a maybe two-part question. I'm curious if before launching your digital-first model, you had your pricing available on your website, and if not, then like what was that change management, especially for the legacy customers that you had? What were some of those steps to kind of transition them into the new?

Yeah.

I have lots of thoughts around this. So first off, we— PTC had not published our list price, but some of our partners had their own e-stores. Whether or not they were actually working, I couldn't tell you, but our prices were out there. If someone wanted to go find it, they could. So it already existed. Now, I started to talk to the GMs of our various products, and they were very scared about putting the list price out. And ultimately, Ultimately, the people who— the enterprise customers weren't going to go predominantly through this motion. So those bigger deals are reserved for direct. And so that's kind of where the— how we made them more comfortable with it. If they were going to come and complain about what the value is on the store versus what they were getting, then we would have to have a conversation with them about it. But so there's some gray there. But yeah, that's how we handled it.

Any other questions?

Well, thank you.

Thank you. And if you were at the Cleverbridge booth, Megan will be around, uh, at the event. So if you have any further questions, please feel free to, to come up to myself, Megan, or any member of the Cleverbridge team. We're happy to talk about it further.

Yep.

Thank y'all.

All right. Enjoy the conference.

Good afternoon, my name is Juliette and I'm delighted to be here today to talk about how we at Glean real— rebuilt our high-performing sales org to be an AI-first the best— org, without losing the human craft, because I very much believe that selling is both a science and an art, and the emotional side of selling is equally as important as the science behind it. I was moving house and I came across this photo, so I thought I'd share it with you to demonstrate the changing times that we're living in. So this is my first manager job. I've worked in tech all my life. Do you— some people might remember having an office, so look at that monitor, the telephone, the wall arts. Look how happy I am. I can't believe— thinking how lucky I am to have my first manager job. But I do think it's a very exciting time. You know, we've gone through the personal computing age, internet, smartphones, cloud, and now AI. I spent the last few years actually at the heart of AI, so I was at Google and then I've recently joined Glean. And in fact, Glean was founded by a Google search engineer to try and solve the challenge of enterprise search, and we're now within 3 years built a $100 million business providing a work AI platform, not just for search but it gives every employee an AI assistant, an agent grounded in enterprise context. And you'll hear me talk a bit more about enterprise context, but it's super important because it's connecting your interactions to your data and providing you with deeper, better insights. This isn't a pitch about Glean. Very happy to talk to you about Glean afterwards. I just wanna share with you some of the things that we've done to superpower our sales org using our AI. And I'm going to share 6 use cases and repeatable workflows, and then I'm going to leave you with a 30, 60, 90-day plan on how you can implement it and create big impact. So AI is flooding the sales cycle at every stage stage of the cycle, and not always in the right way. So what once felt tailored now feels automated. How many people have received those emails and thought, hmm, did a human write that? Um, however, the admin behind it hasn't disappeared, it's just hiding in new places. Discovery calls are sometimes more rushed even though they're AI-scripted. They're missing the human nuance. That uncovers real needs. And while managers have more data, coaching has actually become harder. So the fix is not more AI, it's pairing that emotional intelligence with artificial intelligence grounded in what I was saying before, in the perfect context. So what is perfect context? AI only works if it knows what matters. So perfect context means real-time understanding of your buyer's world. And it's, it's not just about automating tasks, it's about serving the right context at the right moment. And when it's done right, reps move faster and sound more human. And this is what I believe unlocks the workflows I'm going to show you later. So human-centered AI across the sales cycle, it's more than just speed or automation. It's relevance, it's timing, it's empathy, it's precision. So this slide shows how AI can technically plug into every every stage of the sales motion, from prospecting to renewal. But that alone isn't enough. We've seen firsthand that AI without human-centered design doesn't elevate reps. It can actually overwhelm them. So that's why the real unlock is designing AI workflows that support reps in the moment of need, quietly, reliably, and with nuance. And when we do that, AI becomes more than just a task doer, it becomes a trust amplifier, helping sellers move with clarity, confidence, and care. So I'm going to walk through the 6 workflows now. Tell me if any of these resonate with you. These are— oops, let me move on. Each one is tied to a real moment in the sales journey journey, and they aren't experiments, they're daily workflows that we use day in, day out at Gleam. So the first one is research. So we all know that in order to have a great client engagement, we need to do the prep, and we found that a lot of our salespeople were spending nearly an hour building a full picture of their accounts. We have an agent called Account 360, 360 that now retrieves that information with the right context in seconds, which means then the reps can spend their energy crafting the message. So the toil is taken out of all that research and they can spend more of the time on the art of selling. We've structured prompts and templates to balance automation with human judgment so the reps can still adjust for their tone, time, timing and message. And the biggest shift is that we're seeing outreach that actually lands, and that's not because it's AI written, although sometimes it does come up with some quite good sentences, but because it's grounded in that real timely insight that reps can verify and use confidently. Does anybody use Command of the Message? Or Medpic. We have embedded that in agents to create value hypotheses and to help BDRs with their research. Who likes updating Salesforce CRM hygiene? It used to be one of our biggest bottlenecks, and I want to say it wasn't that the reps didn't didn't care, but it's the updating that took them time away from selling. So we built agents that craft CRM updates automatically based on trusted data like Gong calls, Slack messages, and calendar events. And then every Friday, the reps get a draft summary of what they did that week and what they need to do to prep for next week. It's very actionable it's very accurate. And this simple shift gave managers clarity that they never had before. What is in motion? What are the blockers? Who needs help? Who needs support? And the reps saw the results. They saw how that better context helped them sell smarter. Third one, Is a competitive research. So I think every team has got a folder somewhere with loads of decks and proposals and they try and reuse them for, for the next one. We trained AI on the actual winning decks and responses so reps could start from a solid draft, not just a blank sheet of paper. And crucially, every claim had a source, so it came from a customer quote, a usage stat, or a validated proof point, and that's what built trust. It also meant we weren't always reinventing the wheel. RFPs and battle cards were standardized but flexible enough for real-world nuance. However, the human layer did stay critical to this process process because the reps still had to shape their own story, but AI got them 80% of the way in a fraction of the time. We do have one of our agents that we use regularly is the RFP response generator, so that again takes away a lot of that grunt work that we've often faced when completing RFPs. We also have a business case slide generator, which is also a very neat way of summarizing the value that your clients are going to get from whatever you're, whatever you're selling. Next stage of the sales cycle is the meeting follow-up and customer questions. So again, doing those meeting summaries and recaps take a lot of effort. You might have to rewatch a call or start an email from scratch. Now we have a quick post-call summary that's drafted automatically with actions, items, questions, and next steps very clearly outlined. That's super powerful. However, the real unlock came when we tied this into our product knowledge base. So when a customer asked a detailed question, say about an integration or roadmap, we could generate an informed draft using using trusted content. So the reps still reviewed every message, but they weren't hunting for the answers. They were spending that time building the relationships and building the trust with their client. The next one is about expansion and monitoring account health. So once you've won the client, how do you grow the business, and our reps had at least 6 different tools to manage the business and the health of their accounts, which means that inevitably some things get missed. So we use AI to build a single view of each account. It was updated in real time with alerts tied to expansion potential or churn risk. So we get a flag if a customer is likely to and it wasn't just about a dashboard, it was about getting the right signal at the right moment. So if the champion went silent or support tickets spiked, then the rep sees it, sees it immediately. And we believe that's how you keep trust alive in customer relationships, because you're showing up proactively, informed, and that's what partnership is about. Out. Sales excellence. A lot of sales playbooks don't often stick, so reps can often go back to the muscle memory, especially when you're under pressure. So we've embedded our methodology, like Medpick, into the daily rhythm using AI, using AI. Agents now score calls automatically. They can flag missteps and prep reps for the QBRs or performance reviews. I'm actually living at the moment for performance reviews. We're using AI to capture the whole year's performance. Um, again, judgment is needed, so we've got all the data points, so there's always going to be a human element, but it also takes away some of that recency bias that you might experience when you're doing performance reviews. So as a manager, now I don't have to manually review one of my calls that my rep's doing. The rep is actually getting real-time feedback so they can actually react to that faster. And I'm learning how to coach better and coaching the, the parts that are the most important important, like focusing on the message and the tone and using my judgment, not just relying on the data. So in summary, a few of our lessons learned: if you want to scale fast, you have to balance the science and the art. You can't rely on art alone. And standardization actually beats the individual heroics every time. Strong discovery inputs made AI smarter and more relevant. Governance was non-negotiable, especially with scale. And no matter the tech, reps follow what managers coach. However, my biggest lesson was that empathy wins out and trust still closes the deal. So just coming up to the plan, so what you can do in the next 30, 60, 90 days. Don't feel like you've got to boil the ocean. The first 30 days with your sales teams is about building trust with some quick wins. So like anybody, if you experience some early wins, you can start to trust AI as a partner in the workflow. So think about what are those quick-win AI workflows. The ones that we chose were meeting prep summaries, recap drafts, personalized prospecting nudges. Standardize your, the dual-layer personalization across outreach. This requires both account-level and persona-level relevance in AI-generated messages. And have dashboards visibility, and that will give you a baseline of what the reply rates are, the discovery depth, the CRM hygiene. So at least you've got something that you can measure against. Then at 60 days— oops, go back— at 60 days, you start to normalize some of this usage to the standard workflows, roll out some of your core sales use cases. Again, We use the snapshot prep, the recap follow-up, pipeline hygiene, objection handling, and account health reviews. It's also important to define some ownership, so have people in your teams that are going to be the champions, that are going to create dashboards and, um, advocate for this process. And then enable AI-led coaching, so automate Scoring and surfacing call quality, improve your next step rigor, and adhere to your sales methodology. And then at 90 days, you've got it sorted. You scale the impact through governance and orchestration. So AI is now embedded in your sales org and it works, it's reliable, it's governed, it's coordinated. And by this you can add it, um, and orchestration triggers. You can set SLAs and adoption goals, but you must mandate traceability. So, all AI outputs should include links to source material to ensure the accuracy and trust of the data that you're using. So, finally, I'd just like to close with saying that AI doesn't replace the human human sales org. It rebuilds it and it elevates it. It's certainly not the end of the human seller. We've still all got super important jobs to do, but it is our next chapter. I'm very excited about how it will superpower us, and the best teams will be the ones that do pair the emotional intelligence with the perfect context of artificial intelligence. And with that, that's how you and your teams will move faster, sell better, sell smarter, and build real trust. Thank you.

Please welcome Amelia LaRue, CAIO at SaaStr, and Marshall Mooney, VP of Sales at MangoMint, and Ashley Wilson, co-founder at Momentum, and Greg Belzer, CCO of AI for AgentForce at Salesforce. Hello. Hello, friends. Come on in. Thanks for coming. I know Jason is still speaking, I think. So we're flattered you came to sit with us and learn a bit more about how AI is rewiring B2B sales. Um, for— we're going to cover things like quota, we'll cover retention, we'll cover a bunch of different topics. We'll try and keep it fluid so that we can also answer questions for you guys because I know it's already been a long day of a lot of agents and content and things. But let's start with intros. So I'll, I'll start with you, Greg, if you want to give a quick intro. And then maybe also, in addition to AgentForce, what are your favorite AI Got it.

So it's interesting. I came from financial services. And so when they say CCO, I think of Chief Compliance Officer. That is not me. That's not me. Chief Customer Officer for AgentForce for Salesforce. So all things AI. What are my favorite tools? Look, I'm Anthropic on day one. I always have been, mainly for ethical Reasons as far as I like their stack. And then from a CIO standpoint, which was my job before this, I think that they're best when it comes to coding.

So those are good ones. And then we'll go to Ashley.

Hi everyone, I'm Ashley. I'm one of the co-founders at Momentum. And I would say one of my favorite tools, of course, is Momentum. We get to dogfood it all day long. And I feel like as a founder, especially as we've started to scale and hire a sales team and hire a CS team, I can stay kind of— I don't want to say creepily connected to my company, but I do kind of get to stay a couple of steps ahead of my team because I always know what's going on in the business without waiting for my team to tell me. I also use a lot of, of course, Claude and ChatGPT, and we've recently rolled out Gemini, Gemini 3 to the company and kind of our are moving the company to that. And I have been really impressed with some of the just like built-in agent behaviors, even asking a question, writing up a memo and having my Gmail right within there. And I think that's such a great way to kind of make AI more accessible in the organization. And I'm excited to, to keep using that tool at our company.

I'm Marcelle, VP of Sales at Mangament. We are vertical SaaS in the like very SMB space for salons and spas. I run our revenue organization. So everything from outbound sales, full cycle AEs, onboarding, implementation, and, you know, favorite AI tools.

The staff, when you asked this question, my list was like, well, bam, pretty deep.

Yeah, yeah. But I would say that the surfaces I work on the most, really quite committed and dedicated to Notion. Notion is AI. If you asked me 4 years ago, I hated Notion. Today you could not pry it out of my cold, dead hands. I also really like using the way I think they're like editing. All of that is based on Claude, which I also love and use quite heavily. And then, you know, Gemini. I actually had an interesting time on the way here. I stopped off at the— in SFO. In San Francisco before I flew out, I pulled up to the bar and I was like, "Oh, there's a gentleman. He seems to be like working hard. Like he won't bother me. I'm going to sit next to him." And then I sit next to him and I was like, "He's not working hard at all. He's making images on Nano Banana." And I was like, "Hell yeah. That's the world we're in." But yeah, those are some of my must-haves.

Nano Banana, the new one's pretty good. I've been like— I'm team Higgsfield. As many of you know, it's one of the agents that we use for image creation. I had a slide in one of our earlier presentations. We didn't get to it, but this actual backdrop that you see us in front of and then on the podium, it was something where I told Reeve and Higgsfield, okay, I want something that's futuristic, but here's what we did with Saastr AI. I'm a little bored of it. Will you make me something new? Made this like portal opening thing, which maybe is not that inventive and it's like kind of Star Trek-y, but I don't know, it looks cool. Um, so I gave it to our designer and then our designer made this, so that was fun. Um, but I think it's a fun use of AI too, in a way where like, I feel like the whole day too— and I know this is about sales, we'll get back to sales— but like the whole day we've been talking about like sales and quota, which we're going to talk about more of. Um, but maybe just really quickly from each of you, like, that's like one of the most creative things I think we've done where I've seen like AI get creative. And I know somebody, like a lot of the speakers today have been like a little bit anti-AI when it comes to creativity. Like, just what is your guys' hot take on that? 'Cause I do feel like it ties into sales and the motion of like, you know, how creative do you get with your sequences? Like a lot of that used to be, okay, we're gonna hire for the, I know like, Companies I used to work with pre-Saastr, where I worked, it'd be like, okay, for the SDR team, we're gonna like find the most creative people out of college and those are the people we hire.

Yeah.

So now that the AI can do a lot of that, like how do you balance that? Maybe we'll start with you, Marshall.

Yeah, I mean the one that pops to mind sticking on the idea of like photos, images. So again, I mentioned we work with a very SMB customer, so salons, Spas. Many of you look like you've got a haircut at least once in your life. Uh, that business you went to is who we sell to. So everything from booking to payments. Early on, uh, 2021, we really pushed into like a PLG motion. So customers could go to our website. We were the first SMB salon and spa software that you could start a trial, literally be set up and using the product in days. And over the years we've allowed for the product within it, customers to upload their logos, customers to add their colors. Well, we realize our customer is not sophisticated enough to use a dropper tool and to figure out what their brand color is anyway, and they have no idea what a hex code is. So one of the things we rolled out this year during a hackathon was our engineering team, again using Cloud Code among other things, was able to take that process, which currently when a customer would upload a logo and it was like the wrong format, it was like horizontal where vertical looks better on our online booking. Rather than a designer on our team having to take that and completely transform it, it is now all completely done automatically using AI. And then the designers just check a box, you know, manual, make that happen. What's really beautiful about that is is we don't use PLG in the traditional motion where it's like you get in, you sign up, you're all on your own. It's really a tool to tell the sales team how quickly and how engaged this customer is. So we can score that quickly and get that in front of a rep to reach out to them. So within minutes of a trial starting, somebody uploading a shitty logo, it is now fixed and that rep can text them the link to their online booking booking that shows their logo done properly with the right colors within minutes. So that's like one of the ways that we've really like pushed on the creative side to make things like minutes, not hours, not days.

And it's instant value though for your customers.

They— it's actually still a little— they're a little freaked out by it. Like magic.

It's too good.

Yeah, it's kind of like the Disneyland effect. And I like that.

Yes, I like that. Greg, I know. So you're at Salesforce. There's obviously you've You guys have a massive, massive workforce.

Yes.

How are you tackling this at your scale?

Yeah, I mean, so look, I think getting your sales force, get it, to be agentic is easier said than done, right? Yeah. 'Cause it is a cultural change. It's interesting, the word tools keeps getting brought up and people automatically assume just because it's AI, it's an easy tool. It's not, right? The first time I used a chainsaw and I tried to cut a log and it didn't cut, I was like, well, this is a piece of shit. I didn't take off the COVID to the chainsaw because no one was there. I obviously didn't read the directions. I mean, what man does? But it gets to the point you have to get the same culture when you're trying to make your workforce or your folks to be agentic. It is not just about turning on a license for Claude or OpenAI or whatever it is. It's actually about making sure they understand the tool and its possibilities and how they can use it in their daily life. At Salesforce, obviously we have to eat our own dog food as far as what it looks like for us. What I've seen from a Slack perspective of how the sellers use Slack—

I saw a little preview of it.

—is transformational for them. But again, I think the point that I would leave with is sellers want to sell. If you give them the tools tool that's going to help them, they will figure it out. But don't leave it to them to figure it out. Have a strategy around it. Make sure you're getting the right training, making the right education.

Yeah, we've been saying that all day. And then Ashley, I want to get your take on that because you are an AI-native company. You are newer. Your sales team is still doing a lot manually, actually.

Yeah, I was going to say, I think we're probably further ahead than a lot of companies, but I think to make everyone feel better, there's no holy grail that you just get to. And I still think that a lot of the workflows are human-assisted. And I think that's probably where it should be right now, because I think in your talk this morning about all the training that goes into it and the time. Also, we're in a unique position where some of our team is non-English speaking. So our SDRs are in Argentina, some of our AEs are in in Argentina. And so it's been— this is one of my, I guess, favorite creative uses of AI to kind of democratize language and to make these folks be supercharged. Yeah, folks who are right out of college, but not— they don't live in San Francisco, they didn't go to a U.S. college, but they're smart. A lot of them are actually coming from technical organizations, technical universities. They want to break into SaaS. And so the fact that we are able to empower them to do we as a company, but really because we have AI to make them able to write sequences. And they still have to do a lot of the manual work though, because I think that's where to break through, you got to personalize, you got to be human. Our SDRs send a lot of video messages on LinkedIn, and that seems to kind of break through the noise of just all of the inbound that we tend to get from people in those roles. And so I really look at it as human-assisted, start to just like knock off the manual tedious tasks that we could do. Um, and I do think eventually we'll get to agents, agent force, like the world will be run in the next, I don't know, maybe 5 years, 10 years, who knows when it'll happen. Like so much more things will be powered by AI and run, but I just don't think we're there today. So it's more about like bring it in and start with kind of some basic use cases so your organization can also feel like it's a win. Yeah, those are—

I think something you've all brought up, and something it was also at the, um, I was at an Oakland event, um, with Salesforce before literally getting on the plane to London. And it's— you guys have like this tagline, it's like humans and agents working together. And part of the talk too is like, you know, we're— again, I don't think, uh, Mark or Salesforce is saying to fire anyone, but you definitely are retooling people, retraining people, pivoting them to new roles, right? Putting them into FDE roles, putting them to SE roles if you can. So I feel like so much of this is table stakes right now on the rollout. So let's start with Salesforce, because you guys have had one of maybe the most public rollouts of AI. Like, just walk us through it.

Yeah, so I know this is not recorded. I see the camera right in front of me. And no one's actually on social media. Look, the rollout is bumpy, right? The roll— it's not like a day one everything worked. It's not like your data was in a perfect place. It's not like the processes we were trying to enhance were all there. And also ways of working. Yeah. That, that is constantly being rewritten. So to your point, it is not about, again, if you run a call center, yeah, you probably do want to cut heads and there's a there's an absolute ROI. It is about revenue growth. Yeah, right. And then if you look at most of the AI that's, that's out there, people tend to focus on service-related, tend to focus on operational-related. Why? It's easier, um, than sales in, in my opinion. Um, but why is it easier? Just because it's operational? No. Usually your data's in a better place. Usually your process is more advanced because you're already doing that service activity with humans today. On sales, what are your relying on? For the most part, you're relying on your sellers. You're relying on your people to go out there and sell. And so I think as we looked at our evolution, which we're not evolved, so we're not there, but it is to think about rethinking roles, rethinking the tools you're making available. For us, some of the low-hanging fruit and what we publicly expressed was, guess what? Salesforce gets a ton of web outbound leads. Uh, most of those leads are pretty low quality and, um, they're very pick and choose on which ones are being followed up on. It was very low percentage of those are followed up on. Again, so that's a great thing for an agent. Let's, let's turn that over to an agent. Let's go ahead and get those curated, get those to be enriched, get those to— we can actually get a human to pay attention to them, or just get them out of the funnel. Just let's just get them, get them out, get them done, say that we did And that percentage is probably the most that's led to direct revenue that we would have not had because nobody was working those leads.

It's funny, throughout the day we've been talking about this, of like, you guys did that, we did a similar thing where we had a bunch of leads, and then I was like, okay, 'cause we're gonna deploy AgentForce, I'll put it on that agent. We have all these leads, literally, even at, I mean, we don't have the level of inbound you guys have, but like— You mean trash?

'Cause it is, like, it's not just trash, to say it's a whole bunch. That just makes the problem worse.

It makes the problem— but yeah, to your point, it makes the problem worse, right? So it's like, okay, we have everybody, everybody— when we've told the story today, or like when I see other like founders and revenue leaders bring it up, I'm like, everybody think— like, I don't know who can honestly raise their hand in this room and say, I think every lead has been perfectly followed up with. Like, maybe if you have a, a— then like, then be up here Next year. But like, I feel like everybody today was like, no, I feel like I probably should double-check with my team because there probably are some leads that we haven't thought about. Or if you haven't checked, like, you know, your, your org or your RevOps in a while, there's probably a bunch of leads in there. Like, everybody has this use case of everybody has leads they never followed up with. And it's honestly, it's embarrassing for all of us. I'm like, it's embarrassing for Tasker, Salesforce. It's like, okay, We have all these like big brands, right? And it's like, okay, we should be getting back to people instantaneously. Maybe. I don't know. Maybe. Maybe. And it's like, for me, I'm like, I find it so embarrassing. Like when I found all these leads post-annual that I'm like, okay, our sales team is not that big and I get it. They're like, they don't have, you know, maybe infinite time. I'm like, but they do have time post-annual. Like this is the best time to call people, right? Just happened in May. These people wanted to know about annual. You didn't call them yet. Maybe call them now. Like the event just happened. Maybe they came. You could sort that way. And so for me, I was like, this is just like—

I was like, we have to fix this.

I'm like, we have to just do it with AI so that it's consistent. Because part of the problem beforehand was, you know, we were giving these leads manually to humans and they would just, like you said, they would just cherry-pick. Pick. Like, they'll just cherry-pick the ones that they want to work, and the rest will either say, oh, not a fit, or they'll come up with some, like, phony baloney close loss reason. And then you go back and you're like, there's no notes. Like, there's literally no notes in the record. So I have no idea why this, like, got lost.

And I think, well, it's important to, to, like, start that process now. So we are about 80% inbound, 20% outbound. I just spun up our outbound team with our director of outbound this year. So beginning of this year, no outbound. Now we've got it. And what I've noticed is, yes, within— like we started with how do we layer in AI from the top of funnel to make outbound at our ACV make sense? So it has to be data first. So we ingest all of that lead information, enrich it in Snowflake that gets pushed out to our BDR. So it's already good data there. But when it comes to inbound, we're like, okay, we want to keep all trash out of the AE's hands. So we actually have a team that, like, human team that actually looks at every single lead. Because if you're thinking of a salon or spa, well, Amelia could open a salon or spa tomorrow and she's using her Saastr email to sign up for a trial and we're like, DQ, that's definitely not a salon or spa. But if it a human actually calls you and you're like, oh yeah, Bay Area loves getting injections. Like, I'm opening a med spa tomorrow. Then we would actually want to talk to you. So there's these like so much nuance. But what we're doing is cool. We now understand the nuance on a human level. As soon as we can do that with an agent. Yes, we will.

Yeah, right. Agreed. Yep. The agents do take a lot of training and I feel like I want to hear your guys' quick perspective though on like resistance to the rollout, right? Because we are all at different stages right now. Maybe actually we'll start with you. Like any resistance with your— I mean, you're an AI-native company. Maybe they can't resist AI anyway. Yeah.

But like, why are they doing? Yeah. Yeah.

Not in our organization, but we are selling to companies that are dealing with their own internal resistance. And I think it's been really interesting to see. I mean, I was on the stage earlier with Contentful, and we talked to them twice over 2 years before they became a customer, and it was just a matter of internal teams getting on board. You had someone like Ashley who was championing it from the beginning, but they're European-based. They have a German works council. Their legal team was very, very reticent to bring AI into the organization. So I think that what we see is that it's great that we've moved beyond like note-taking. We're like, okay, we can have bots on the call, we can take a note. So I feel like it's easier to sell AI to organizations than it was 2 years ago. But I'm still kind of amazed at the internal resistance when you bring in legal teams, CIOs, like it, it is so dependent on the personality and the mindset. And I do believe that we will eventually push through all those barriers over time because it will be so obvious who's leveraging AI and who's not. And I think the companies that are will be so far ahead. But I, I've come to just appreciate that some things really come down to timing. And another angle of that is that the company that's going to adopt adopt AI internally needs to have a strong team championing it because you can't just give these tools to your team and expect them to just magically use it all day long. I think there's some crazy stats out there about how many people are actually using AI in their orgs, and they probably all have access to ChatGPT or Claude. And it's just, it's very hard to get people to change their behaviors. So I think that's equally important is that there is just somebody raising the flag internally and who's really committed to the cause of saying this is going to be transformational, but it's going to take time and you have to have buy-in.

Well, it's like, it's, it's the psychology. Like, as humans, we will do far more to avoid pain than to gain pleasure. Like, it's just what we do. So like, I noticed we rolled out Momentum with our team, started with our onboarding with the marketing team, worked backwards to AE, soon BDR, and what I noticed is I could quantify that we were going to save 15 minutes per deal. For our AEs, that's 16 hours a month. They, their 20 wins is their quota. I can quantify that, but what I didn't teach them at first is how that will change their workflow. So they are literally 30-minute demos, demo, closed the deal, immediately have to get off of that call, build notes for the onboarding team, get that into Salesforce. Well, what we learned immediately was we don't have that readily available. It's going to take 15 minutes to get that information back to them. So the workflow changes. You're now stacking demos, and then you have an hour-long break where you're going through your Momentum tasks in the, the Slack that you're already already in, so you're not having to change surfaces, processing them much more quickly than it would to do it manually. But actually, like, implementing that and being very explicit about the way you work is changing. It's going to look different, but the net gain is going to be so much greater. You have to, like, be ahead of that. And so you've mentioned a couple of times, like, I'm really active in the onboarding of Momentum into our team. I want to build, I want to be in in the tool. If I'm not, that's where the resistance bubbles up. Yep. Because it's like the, the greater than they, like, I'm here, I'm in this, I, I know how to use it. It won't be that way forever for every single tool, um, but I think you have to, like you said, have that champion, that leader that is in it.

Yep. And then anything you want to add from the Salesforce side?

Yeah, I mean, um, look, throwing CIOs under the bus, I'm a former reform CIO. And the problem with AI, it's not just AI, right? It's not just saying, oh, I'm anti-AI or not. Again, there are data to be concerned. What is your company IP? How are you going to turn that over? I mean, there are other concerns that technology experts have of why we are perceived that way. It's just, it's dangerous. By the way, the CIO that, you know, gets a data breach, like, that's That's gonna be fine, that happens. The CIO that gives their data away to AI is fired. Yep. Right? Like, that's a big difference. And also you gotta think about the psyche of those folks. So I'm just a little defensive. All right, got thrown on the bus. But you at Salesforce, look, several things. Number one, we have a product, right? Yep. So there's a product that they want to learn and use because they're going to be selling it. Yep. Even if they're not a seller of Agentforce. Yeah. Anecdotally, like they're going to need to know it 'cause that's the questions asked. So, so you, you did have that buy-in. But I, I also think that when you're talking about, um, is AI gonna replace humans, you know, that's the fear. Maybe that's part of the adoption, or maybe the tool doesn't live up to the expectations. Maybe that's part of the thing. But your analytics is, is your best friend, right? Like you need, whatever your agents are, you need to make sure you have defined analytics. Yep. Now KPIs, ROI, I could probably go a little off on that. I'm not, not as big a fan on that, but the analytics around it, you need to be able to prove that it's actually is working, what you're getting out of it, because you're going to want to improve it. I also think that's pretty lacking in the industry as far as when you look at your agents that are built, getting those level of analytics to say, yeah, why is it dropping out in process 3 at 21%? Is that because it's bad data? Is that because there's latency and the customer's not waiting around? Like, what, what What is causing this to happen? It's been a big part of Salesforce's roadmap since Dreamforce, but I also think it's something you need to have in mind as you build out your own agentic layer.

Yeah, it's interesting 'cause I was like, okay, pre-May, right? Before we went on this agent journey and a lot of other folks did too, like, I was like, I don't think our data's that bad. And then I started to onboard some of these tools and agents and I go, oh man, like our sales team He never put half the stuff in there. Like, you know, there's no close. It's just close lost. I'm like, well, where are the notes? Okay, no notes, no calls, no nothing. They just, you know, never did it. And why that sucks for so many reasons now is because you're now— your agent can't go back to those things. So like, if it never existed as like a record previously in your data, it's not going to exist there now magically with AI.

It's not going to find it.

It's not going to find it. So I think there was a little groundwork we had to do when we first rolled out all these agents to say, OK, what should we implement? And it's part of how we picked some of these tools was like, OK, well, what should we implement so that going forward— I obviously can't change the past, so that is what it is. But going forward, what can we put in so that we make sure everything is always in Salesforce as it should have been? We've had this system of record since I've been here, 9 years now, like alongside our marketing automation platform. But why does my marketing automation platform have like 10,000 more data points than Salesforce? I was like, Salesforce should have more actually, because our, our marketing automation pushes to Salesforce, but then the AEs are pushing to Salesforce. So it should have more, but it didn't. They were like almost even. And so part of this process for us was making sure we could implement the right tools in the right ways to say, okay, what are we going to pick so that everything goes back to at least one source of truth? So right now that is Salesforce. Like, we use Momentum and like that goes back into like, that was why, like, after annual I told Ashley, I was like, I need your help because I went into Salesforce. You're like, there's nothing there. Yeah. I was like, okay, close the lid.

But I think you don't always know what you don't know. I think we're just starting to wake up.

Like, in the back of my mind, I was like, well, you You think it's being handled.

And then, and then the, the old way of working is you would go to the reps and ask what's going on. Yeah. Then you realize, talk to them. You could go to the data itself. Yeah. And make your own decisions. Yeah. And I think that's just the change now that AI is unlocking for us. Yes.

That's a, that's a big call out because I feel like to part, it's not that I don't trust humans anymore. Obviously like we've worked with a lot of humans to make this happen and I've gotten to know a lot of humans in this process, right? Like all these people here, I'm like, I've met so many really great humans in this process, so don't take this the wrong way, but a lot of our humans also just didn't do their jobs. The basics of put your, I think it's, you know, AE SDR 101, put your contact and notes in Salesforce or whatever your CRM is.

It is, but it's also like the personality type. Yeah, yeah. The top seller has a persona, and that persona at my organization in our organization at least is a whatever forever, you know, anything you've got to throw at her is gonna roll right off and she's closing that deal. But there's no notes. Yeah, there's no notes. She could screenshot her text messages all day and send them over. But it's like, it just is what it is. Another one of our reps, I love him, but he shared with me after working for the company for a couple months that he basically had to, he'd been in sales for a while. He was like, I I explained to my wife why I had all of these women in my text messages because he had never sold to salon and spa owners. And I was like, no wonder you have no notes. And like, everything's in your phone and you're having to explain yourself to your family, so you're not going to explain it to me, right? But this— the— what shifts when you do get something like Momentum and you get everything into Salesforce and these fields filled out is like the signal-to-noise ratio shifts so quickly. It really does. That then you have to, because again, like I notice even for myself that you become addicted to the noise or sifting through it, and then you have to like stop, regroup, and go, it's all here. Yeah, like we're okay. Yeah, now what? What next? Now, like, how do we then level up and do more with the team you have? Yep.

Yeah, and I think that's just such a big unlock for most people, right? It's like, okay, once you finally deploy your first agent, one agent, whatever it is, sales, support, retention, like wherever you're going to deploy it, it's like, okay, getting it to work is one. I know it sounds like a big hurdle. It's a hurdle. Yeah. But then once you get there, I feel like everyone has this like aha moment where like, even for me with AgentForce, I was like, I've rolled out so many agents already. I was like, I have to now do another one. But then, um, like we, like, we started like going through the process together, and then you guys brought on like an FDN. I was like, well, that one, that helped a lot. And then two, once I like got over this like mental hump, I was like, I don't know, like it looks kind of complicated to be honest. And then I got in there, I was like, no way I understand this. I was like, I just couldn't like, I was like, I just need to sit for like 10 minutes and read it and read the prompt and like how I can train it. And I was like, okay, never mind. Like I get this. It's like not that different from a lot of things that you prompt and build and tune. And like some of the— once you learn how to do one agent, like a lot of the core principles you learn are actually transferable, like vendor agnostic, are transferable. Yeah. Which I feel like is a big thing for folks. And then before we run out of time, I want to talk about two more things, which is one, because it's in the title, let's talk about quota for a second because a few people have asked me about this today of like, what's the quota of your AI? SDR. I was like, well, it's more than what the human quota was. Also, it doesn't ask for, you know, a percentage of the AE's commission when the deal closes, so that's nice. But let's open it up to you guys, whoever wants to start on quota.

I'll talk about quota all day long. Yeah, I was gonna say. I actually, I just announced to our team last week that we are not raising our quota in 2020. And I could have. Yeah, that's interesting. Yeah, I could have. Maybe you would do, you know what I learned is what I've seen happen this year. This year was like the proving grounds, right? So I've been in 2024, I implemented AI using a tool called DemoDesk for my team. So I was able to analyze, they're based in Munich, great company. Unfortunately, we're no longer using them because we moved to Momentum for some other things, but I can't help but shout them out. I think they're fantastic. I almost cried canceling that. That was weird. Like, it's rare to feel emotional about a company, but I think what they've done was so good and they rolled out AI early, but the Red Queen's race gets everybody, right? Anybody know? Like, you're running at the speed you've always been running, but suddenly the trees are running with you. You've gotta run faster. So when I think about that, this year we roll out Momentum. You know, my team, like I said, we're very SMB SaaS. My team closes 20 logos a month. When we roll in our payment processing, this is very SMB doing up to the like 7x, you know, ARR to OT ratio. That's crazy in very SMB. But it feels good to us. What I learned is we've got these top performers that are almost those like overperformers. Performers. Uh, there's usually, you know, a couple boxes you could check that make you realize, oh, that's possible for you. And that's great. So you're closing 35 logos in a month, absolutely wild, but good for you. Like, to make that OTE in very SMB sales creates an absolute, like, raving champion of our team. They want to run to RepView and put a review on there and say great things about the company. But I am pleased as pie, as our COO always says, and I love, with that person who's running at quota because you're still incredibly efficient. What I'm going to do is spend this next year really making AI allow you to push up into those top performers. Very curious to see what happens with those top performers once we're like, once things feel truly native, but we're not there. And so when we talk back to like, how do we make these people feel safe in this new environment? That's it, is like reap that benefit and that reward to where it's so like an intrinsic motivation that it feels like part of you, then we move up. And we will get there, but it just felt right. I think that's interesting too.

It's making me think about when we were writing our, the book where we interviewed a a bunch of leaders, Kyle Norton from Owner.com, he, he talked a lot about the systems that he's put in place to support the team and how they have a lot more RevOps folks and ops folks than a traditional team would have. And that theme actually came up 4 or 5 times from Demandbase, some other companies that we interviewed. And our takeaway was that we're going to have a lot more operational folks running the agents, running the data, which will allow the teams to be more efficient. And I think that instead of thinking, okay, we just gotta like double quota now because everyone's going to be a lot more efficient, that will only happen if the business is supporting those folks with all the underlying, hey, do my SDRs have enough accounts? How do we get accounts? Where does that come from? It's like you need to orchestrate create and architect all those systems from top of funnel on down to post-sales. And so, and I don't think we're in a like industry-wide place where we've all like proven that out. I think we're all much more like you guys, us as well, where we're saying let's start to just build, bring in more people on the system side, add the headcount as needed. But I, that's why I don't really buy yet into, okay, we're gonna 2 reps and we're going to get to $100 million. Like, sure, that works if you're lovable. Like, that can work with the right business. But a lot of us do not have that business. So to hold ourselves to that standard just feels impossible and unrealistic. And better to think about, okay, what are the underlying systems we can put in place?

And I just said I'm not raising quota this year, but I was just talking to a CEO of a native AI company and he's telling me about about his company and how his head of sales is like, "Oh yeah, well, we've got this particular quota and everybody's hitting it." He's a new founder. He's probably 26, 27. He's whip-smart, but this whole world of building a revenue organization is new to him. So he's listening to this head of sales that's like, "Yep, that's industry standard. This is the standard playbook. This is where we leave quota." And I said, "You got the wrong person." That's not it. What you don't want to do is take the old playbook and try and plug it in. Like, you actually have to strip things down to, you know, first principles. Like, what is your business doing? And that's where it's like, he could probably double his quota and people would be motivated by it. Like, they feel like they're reaching for something. So yeah, yeah.

So from a Salesforce perspective, it's, um, Salesforce is always going to raise quota. That's just what they, they do to the sales Salespeople. But I think it's interesting when all of the panelists, and I'm going to echo it, didn't focus on raising quota and increasing sales. It talked about what happens when you do that. Yeah. And I think I can point back directly to your story in that, fine, we're selling the shit out of AgentForce, and that's great, and it's wonderful. But again, in some of these situations, it's not It's not just click, click, click, and you're done. Right? Right. Like some of these flows are actually very complicated. Yep. They involve a lot of pre-work from your data perspective, maybe your workflow perspective. Can you actually service your accounts afterwards? Is your operations also scaling? So it's not just a sales issue itself. For us being the forward deployed engineers is what's helping us as, as a bandaid, because it's really, it's a bandaid until product makes it where it is just quick, quick, quick. Quick, right? But to get quick, quick, quick, it still has to have a lot of maturity. So yes, you can definitely increase your quota if you'd like. You can set different ones. I've seen everything at Salesforce. But you also need to make a holistic view. And I think that's what all three panelists wanted people to think of. It's not just about sales. You need to look at your business holistically.

Right. Yeah, I mean, if you're not doing that, it's just not—

There's going to be repercussions. Yeah. Yep.

All right, we'll end on the topic of retention, just because we have a CCO here. I know a lot of people aren't using it for retention. We've kind of mildly used it for retention in the sense that because we're so seasonal, right? If you think of SaaStr as, okay, we have sponsors like these guys out here for SaaStr Annual, then some of them do London, some of them do year-round stuff because we have bigger partnerships. Cool. Most of them do one at a time, and so we We have this seasonality that's just natural to us with our event cycle. And then we have media, which is a little bit different because that's year-round. But for that, the event cycle is very seasonal. So we have this seasonality. And so one of the first things we've done for retention is load them up into the agents because I was like, because one, we had new people who raised their hand for this last SaaStr and said, "I want to sponsor literally one of these booths out here," and we never replied light to them. But I also had, you know, 200 existing vendors who had just done this like last Aster in May. And I'm like, you know what, I know the team says they're going to follow up with all their current customers and current accounts. Why don't they let the agent do it anyway? Like, like, no harm, no foul. Like, it's not going to take credit. It's not going to sell, you know, whatever, you know, these booths were out here today. Like, uh, this is a or I don't even know the ACV on this, but let's just say $50K, right? It's not gonna sell a $50K booth, but at the very least it can keep emailing them if they wanna be emailed, get their feedback from the event, check in on them maybe. Like, you know, half the time I'm like, hey, did you talk to all your sponsors and your accounts? And they're like, oh, I saw some of them and then I got busy. I'm like, yeah, I get it, but like, what are you doing tomorrow? Now that the event's over, I'm like, 'Are you reaching out to your accounts? Are you doing things?' And they're like, 'Ah, you know, some of them were, some of them weren't.' And so I was like, 'Okay, enough of the 50/50, some yes, some no.' You'll find this with your sales team no matter what happens. You're always gonna get a mixed bag of results. You have top performers, you have low performers. That's just how the bell curve works. And so for us, we were like, 'Okay, we're gonna load these existing accounts into one of of our agents, some of them are in agent force, and I'm gonna, like, I've literally, that's why it has such a high open rate too. It's a mix of, you know, people we ghosted and then people who are customers. Yeah. Are getting into this sequence where we're following up with them. And I'm like, in some cases they're like, oh hey, I actually didn't get to connect with my AE post-annual. And I'm like, yeah, they told me that you did.

The gig is up.

The gig is up. Yes. Uh, but Greg, just any quick thoughts on retention with AI? Yeah, so it's, it's interesting.

Um, there's a, a stat from McKinsey deck I saw, which was, uh, 40%, and this was in financial services, 40% of financial services customers want increased, um, outreach from their firm. Yeah. Uh, but I think when you looked, but, and you got that study, it's like, okay, fine, yes, they wanna be talked to. To, but they want to be talked to on their level about products or services that interest them. So it's not just about creating a traditional, like, marketing campaign where you're dripping. Like, that's not what an agent is for. It's not— you can, you can use plenty of off-the-shelf software to do that. It's exactly what she said as far as mentioning about getting some kind of action, and you're trying to build that relationship digitally so that your human can close it. That's all they're doing is trying to build a little bit at a time. Maybe it's a survey, maybe it's a question, maybe it's thought leadership. Doesn't really matter, right? But think about an agent in that standpoint of being able to continue to keep the relationship fresh. Because as a human, you can only talk to so many people all day.

Yeah. I mean, until this specific— again, to my customer— until there's an agent that texts with the blue bubble, It's just, it's not gonna happen. The bias on email for us is so extreme because salon and spa owners don't open emails. Like unless it's like, I'm booking a haircut tomorrow at 3 PM or, you know, need my nails done. So I think that's the biggest piece we've run into there. We have like incredibly high like net revenue retention to begin with. Like our customer, our product is so sticky. So to me, the way I think about it is more like it goes back to how quickly we need to move deep into our vertical to make the hard decisions for them. That's what drives retention. And so, yes, we do that through AI, but it's not through an agent reaching out.

Yeah. Ashley, any last closing?

Yeah, I think what's been interesting, I mean, because we have some customers that use us much more heavily post-sales, and I think CS has tended to be underserved in terms of tooling and process. And I think the expectation has still been high on them getting data back into Salesforce or get the data in Gainsight.

And it's still down. But yeah. Yeah. So it's—

there's like a lot more opportunity actually on post-sales, I think, in disrupting and not just the customer support chatbot experience, which is like great starting point, but actually again to help help CSMs and CS leaders have a better pulse on the business. The moment that somebody is expressing dissatisfaction, it's— we all know that it's a slow— I mean, unless you're like egregiously messing up your product, it's kind of a slow drip to churn. It's, yeah, hey, I didn't get this answered. Hey, it's— this is not working. And it's— I think it's really hard for CSMs, and I oversee customer success, so it's like a close thing into my heart where I'm like, it is hard to know in the moment what needs to be escalated. Unless somebody is being like, I'm going to leave. Yeah. That kind of soft, slow dissatisfaction. But now that I have momentum listening for those moments of dissatisfaction, which almost sometimes swing the other way where you get on, you see a clip of the customer saying something and you're like, what is going on? And they're like, it's actually kind of fine. Like it's not a big deal.

I've had to mute mine since I've been here. Yeah.

So it's like, but I think that on the CS side, I'm very excited for what AI can do for retention and keep doing. Not to mention something like, you know, we have a customer that does 3-hour support calls in another language, and then they had to go write that into Salesforce. Imagine the time savings and the ability now to bring that all into a common language, English, in the CRM. I think that we should all be orienting ourselves to thinking about AI for customer retention and CS just as much as on the sales side, because I think it's all going to have such a huge impact on our businesses.

Take it. Okay, we're out of time, unfortunately, but we'll stick around and answer questions. And then if you guys want, the very last session of the day is— I get confused up and down— is downstairs. So Harry will Moriah and Jason are going to do the live podcast. Live from London right now. But thanks for joining us. Thank y'all.

[MUSIC] [MUSIC]