DEPLOY STAGE

Our program is going to begin in a few minutes. It's time to find your seat. As a courtesy to our speakers and fellow attendees, please take a moment and silence your devices. Let's give a warm welcome to the Chief Product Officer from Vercel, Tom Occhino.

Hey everyone, thanks for joining today. I'm really excited to be here. My name is Tom Okino and I lead engineering, product, and design at Vercel. Uh, at Vercel, we're focused on empowering companies to build and ship agents securely and at scale. I'm not from go-to-market. I'm an engineering leader and a product guy, but what I'm seeing is that the agent era is quickly collapsing the distance between go-to-market and product. And I believe that the go-to-market teams that win will be the ones who are able to innovate and iterate really quickly, much like a successful product organization. And that's gonna require having the right infrastructure and the right tools, which is what I want to talk about today. But before we get into that, I want to provide you with a background on both myself and Vercel. Seeing this logo, some of you maybe for the first time, and I just want to talk a little bit about that. Before my current role, I spent over 12 years at Facebook where I co-created React and React Native, uh, and helped create many of the open source tools and shared technologies that underpin the JavaScript ecosystem. And I partner closely with our CTO Malte Ubel, who spent 12 years at Google where he was a principal engineer for search and created some technologies hopefully many of you know, like Core Web Vitals. One thing that always struck me about these big tech companies was the size of the investment that they made into the foundation underneath the products that they developed. The tools that engineers used and the teams that made sure that products actually reached customers and scaled and stayed up and secure. These companies invested in that foundation at roughly the same level as product development itself. And this is surprising because product is what drives engagement and pipeline and revenue, but the foundation underneath is often seen as overhead. It's often seen as a tax. But what's most interesting about the investment that these companies make in that foundational infrastructure is that none of it had to be purpose-built for either Facebook or Google or Microsoft. And in fact, you could run virtually any globally distributed product on top of what companies like Facebook and Google created. Malte and I often joke about the fact that you could very easily build Google Search on top of Facebook's infra and build Facebook News Feed on top of Google's frameworks and infra. And so that's what made me obsessed with this idea that every company that spends— every minute that companies spend configuring infrastructure, building bespoke frameworks, and configuring, uh, and, and dealing with DevOps is energy that isn't going into what differentiates them: pipeline-generating activities, product, um, and their customers. I call all of that foundational investment undifferentiated heat loss. One of the reasons I joined Vercel is because I wanted to build a general-purpose version of that foundation. A way better version, obviously, based on everything I learned over my almost 13 years at Facebook. Um, but I wanted to build general-purpose frameworks, SDKs, and managed infrastructure that could be used by the greatest companies in the world. And that's what Vercel actually is. We're a managed infrastructure platform as a service, uh, trusted by over 16 million users, including hundreds of thousands of paying users. Um, the thing I love so much about our customer base is how diverse it is. We have everything from financial institutions to e-commerce, um, to just kind of everything you can imagine. And the reason so many customers choose us is because we actually enable them to achieve better outcomes for their business. When you use Vercel, you get to focus on what makes you, you, rather than the undifferentiated tools infrastructure and DevOps that every company has to invest in. So the first era of Vercel is what we call the frontend cloud. And the key unlock for the frontend cloud was framework-defined infrastructure. It was sort of the frontend cloud's claim to fame. I won't go too far into the details. If you search framework-defined infrastructure, there's a great blog post from our CTO from a while back. Um, but it worked really, really well for static websites, dynamic web apps, and even APIs. But we've been expanding our suite of tools and the capabilities of our platform pretty meaningfully to support modern AI applications and workloads. We call the second generation of our platform Agentic Infrastructure, and it has three key pillars. First, we're extending framework-defined infrastructure to become self-improving infrastructure. It's not that the infrastructure is automatically configured for you from your application. It also goes beyond that to actually recommend improvements both to your codebase and your infrastructure configuration itself based on the unpredictable nature of the production environment. In this way, Vercel itself is becoming an autonomous DevOps agent. The second pillar here is that we're ensuring Vercel is the best place for coding agents to deploy software to. Vercel already works great for websites and web apps, whether they're deployed by a human or an agent. Um, and we're making the platform work even better with the tools that agents love. But this last point is the one I wanna spend the most time on today. We're ensuring that Vercel is the best place to build and run AI apps and agents in production. In the AI era, the way we build and run software is changing pretty fundamentally, and hopefully you all have seen this by now. Claude Code, Codex, and other coding agents have really fundamentally changed how we do software development, but the shape of the software that we're developing is also changing and we're going to need some new tools. While we're still building plenty of static and dynamic websites and web applications, we're in the middle of a massive shift that we call the shift from pages to agents. And that also includes AI-powered and agentic experiences. Agents are different from traditional software in at least two 2 big ways. First, the way we interact with agents is different than the way we interact with traditional software. For the past several decades, the primary way we've interacted with software has been through purpose-built user interfaces. While the form factor of those interfaces has varied a bit, from websites and web apps to native desktop and native mobile apps, the model has largely remained the the same. Everything you access and update generally happens through some dedicated UI. But much of the software that's being built today is starting to look a bit different. It's actually not always UI first anymore. The way I think about it is that until now, UIs have been sort of the tree trunk from which, uh, of most of our software systems, from which any action you can take is a branch off of it. Most actions would start from the user navigating to some page, taking some action, and then that triggers some effect in the system. But with agents, UI is increasingly becoming more of the leaf nodes of our software, these lightweight surfaces that help operators make decisions and understand outcomes. And the new sort of tree trunk of our software systems is this headless autonomous software. That's an agent. But the second way that agents are different from traditional software is, is more important to acknowledge. Agents can increasingly act autonomously on our behalf. They'll wake up when something happens, like an incoming sales lead, a fraud signal, or a monitoring alert, and they'll take action following the rules and goals that we provided with them. Them with, and then they can escalate to us only when necessary. Sometimes agents will have a conversational interface, but in many cases they'll have no interface at all. We think every company on earth is going to need at least two agents, one for your employees, internal knowledge and actions and workflows accessible via the collaboration tools tools you already use, like Teams, Slack, or any other internal tool, replacing the need to navigate through complicated legacy systems. And the other agent for customers, whether you're handling support, helping with transactions, education, self-service across any device or channel. And while it's tempting to try to purchase agents off the shelf, call these agents as a service. Um, in our experience, agents as a service don't really work. Just like it's really hard to have a one-size-fits-all website, it's even harder to have a one-size-fits-all agent. So over the past 2 years, we've come to this conclusion for ourselves that every single team is going to need to build their own agents, and that AI has fundamentally changed the equation on build versus buy. Vercel uses Vercel to build Vercel. Um, and so we've actually built hundreds of agents over the past year, literally hundreds. And one of the, uh, I want to talk, walk through some of the agents that we've built just to illustrate kind of the types of things that these, this software is, is solving for us. One of the agents we built is a content agent that turns Slack threads into drafts of blog posts. So you can have this 100-message-long thread in Slack and you can send an agent in and it will summarize it, do a little bit of research, and then draft using our voice and our narrative, a blog post draft for us that a human can go in and edit and review. And most of our marketing content now, I think 96% of our marketing content is generated this way. Of course, there's still a human involved, but the amount of work that goes into that first draft dramatically reduced. And we can turn these things around a lot faster, so they're much more relevant. Our lead qualifying agent replaced a large SDR team doing manual repetitive work, and all of the folks that were working on this type of work have now been redeployed into the rest of of the business into much more impactful roles. And another of the agents we built, you've probably heard lots and lots of examples of this one, but it's a customer support agent and it now handles over 90%, as of this morning, 93% of customer inquiries without human intervention. What's so great about our customer support agent is that anything that gets by it is actually a real opportunity either for improving the product or for fixing improper configuration. And the team gets to work on, you know, real issues, real opportunities, rather than just mechanical, you know, sort of anything that can be handled by an agent. But the agent I want to deep dive on today is an agent that we call Deal One. Deal One is one of our go-to-market agents. It ingests every sales call, generates notes with action items, and posts coaching selections, coaching suggestions directly into an internal Slack channel. It also proposes CRM field updates, tracks competitive mentions and objections over time, and runs postmortems on every single closed, lost deal. And all of this happens where our team already works, in Slack. So I wanna dive into what's under the hood of this thing, because this is a pretty sophisticated agent. We built a multi-layer sales intelligence stack. This diagram is probably a little overwhelming, but I'm going to walk you through it one layer at a time. First up is the ingestion layer. The moment a Gong call ends, Gong actually posts a webhook to a Vercel function That kicks off a Vercel workflow, which is a durable 10-step pipeline that reliably finishes start to finish every time because of Vercel workflows. That workflow pulls the transcript and runs it through Claude via our AI gateway to produce a structured data summary. That summary, including topics, objections, deal stage, sentiment, stakeholders, and more gets embedded into a hybrid vector and keyword index. Then a central— this is a central knowledge base that also includes other call summaries, atomic objections and blockers, cross-deal patterns that we see, and more. This whole index is exposed via a secure MCP as a secure MCP server, so that any agent, whether it's Deal One, Claude Code, or something we build next, gets the same context through a single authentication boundary. And then we have our actual agent layer. This is where reps feel it. They can @mention Deal One in a deal channel on Slack and ask anything. What were the top objections from this customer? Customer? Why did this deal slip? Can you coach me in advance of my next call with this customer? And they get back a cited answer that links to specific moments in Gong call transcripts or Slack messages that they have access to. Agent execution all runs inside of the Vercel sandbox, and after each turn, the sandbox is snapshotted and resumed for following follow-ups in the same thread. Durable conversation state, zero infrastructure to manage. And the best part is that reps get to interact with this agent exactly where they already are. They're already working in Slack, so that's where they have real-time deal assistance. They're already in Notion, so that's where the structured, uh, outputs will be shared with them, and then they can share as needed. And then we have a Next.js app with guarded by Okta SSO for the dashboard. The intelligence comes to them and this is one platform end to end, all built on top of Vercel's primitives. So this slide shows how it all really connects. Data at the bottom, intelligence at the top. In every single sales, sales call a Vercel rep takes automatically turns structured data into intelligence and it surfaces it to the reps right where they already work. All right, so over the past 2 years we've been building agents like Deal One and also v0. For those of you who are familiar, have any of you used v0? Okay, a few of you. v0 is a web app for building web apps with AI, and it actually under the hood is an— a web development agent. Um, and we've been using a series a series of technologies and creating a series of tools, uh, to fill the gaps that we needed in order to build tools like D1, Deal One, and V0. We call this collection of tools the Agent Stack, and I'm going to walk through each of these tools. There's more that aren't listed here, but these are the big ones that I wanted to highlight today. The Vercel AI SDK lets you easily build streaming and interactive AI apps any model without complex setup and without worrying about vendor lock-in. It simplifies integrating pre-trained models or custom models with just a few lines of code and dramatically reduces boilerplate that you need in your apps that leverage AI. It also unlocks developer freedom to choose and switch models seamlessly without changing your application code. This is really important because if you want to experiment with the latest model that came out, you shouldn't need to push a new version of your code base or make invasive changes. You should be able to just try it out. And every customer I talk to is spending tons on inferences— inference. Many customers are spending millions of dollars a year on inference, and AI Gateway gives developers a single unified way to access over 100 AI models without needing to juggle API keys and provider accounts or rate limits. It includes features like fallback routing, which prevents literally millions of requests a day from just silently failing. And this is how Deal One accesses Claude to generate structured call summaries. If our access to that model is limited, rate limited, or if there's some problem with the upstream provider, AI Gateway seamlessly fails over to another provider so we don't miss a beat. Next up, we have the Chat SDK. The Chat SDK is a unified TypeScript SDK for building chatbots across all of the apps you can imagine, like Slack, Teams, Discord, even WhatsApp. It enables agent builders to connect with users wherever they are by abstracting away the complexities associated with integrating with those chat apps. If any of you have ever tried to integrate with Slack and Microsoft Teams, you know you have to build all of the same stuff twice. And it shares none of the code. But now with a slightly higher level of abstraction, the Chat SDK makes it so you can target any app you want to wherever your customers are with a single codebase. All of our internal agents at Vercel, including Deal One, use this so we can interact via Slack and all other means. Another big innovation from Vercel is fluid compute, and it's been actually critical to this new wave of software. Fluid Compute combines the scalability of serverless with the flexibility of servers, enabling real-time efficient workloads like APIs, streaming, and AI. Now, for traditional apps, Fluid Compute was a nice optimization. It was actually a, a price-performance optimization. Um, but it's actually absolutely critical for AI apps because they're much more I/O bound generally. You're waiting for responses to stream back from the model labs, and you don't want to be burning compute just waiting. Vercel Sandbox provides a secure ephemeral compute environment ideal for executing untrusted or experimental code without exposing any of your production systems to risk. Each sandbox runs completely in isolation with strong security guarantees, enabling your teams to test, validate, and audit models or code artifacts safely. Sandbox runs on Fluid Compute, and agents like Deal One leverage Sandbox for safe code execution. And lastly, another important piece of the puzzle that makes agents like Deal One possible is the Workflow SDK. Workflow brings durable orchestration and state management to complex tasks with built-in retries, state persistence, and event observability, Workflows enables resilient, auditable automation across long-running systems. Again, this is another nice-to-have for traditional software. If any of you have spent a bunch of time building microservices and orchestrating between those things and manually configuring retries, this would just dramatically simplify the entirety of that system. But it's, while it's a nice-to-have for traditional software, it's absolutely paramount for agentic development because agents need to be able to pause, resume, wait for a human in the loop, automatically configure and wait for retries. Teams can build adaptive processes that respond to market events or compliance triggers without needing any additional infrastructure. Combined with Vercel's managed infrastructure, we think this collection of tools and a few others not listed here provide you with everything everything you need to build agentic experiences, AI apps, and agents. And this is important because 2026 is emerging as the year of agents. I can't even count how many times I've said agent so far, but I'm sure somebody will do a, a, a clock. It's, it's been a lot. The companies that move, uh, agents into production first will compound that advantage for years. And the window to be early is sort of closing, but it's still open, and you're all in this room, so I think you're already thinking about it and you're probably in the right place. If you've never built an agent and you don't really know where to start, my suggestion is actually the following: use a coding agent to help you build your first agent. Here's where I would start. Identify some low-value, highly repetitive work, um, that someone on your team is doing or that you're doing yourself. Support triage, data pulls, lead qualification, anything you're doing dozens of times a week. That's where agents can deliver the fastest ROI and where you'll learn the most about how they'll work in your environment. Second, document how that work should be done. This is kind of counterintuitive. You're like, I can already do it. Why would I document it? An agent is only as good as the process that we give it. So if you can't write down what a human should do or what a human should be doing for that process, an agent won't be able to do it either. So just document in gory detail exactly how to accomplish that task. And once you have it documented, you have something that a coding agent can actually consume. And then you can work with Claude Code v0 or whatever tools you prefer to give it that documented plan and iterate on it with it all the way into production. Start really small, something repetitive, something you do every day, um, but learn as you go. At Vercel, we're building the agentic future by solving problems for ourself first and then sharing our work with everyone. Uh, I hope you'll leave inspired today at SaaS or Deploy, um, to figure out what's possible for your organization and to how you can leverage these tools and this infrastructure structure. If you want to learn more, catch us at one of our upcoming Vercel Ship events, uh, kicking off in London in June. Um, but really excited to be here with you all. Appreciate you spending some time with me as the first talk. Uh, have a great time. Thank you so much.

Please welcome the CEO and founder of Artisan, Jasper Carmichael Jack.

Thank you for coming along to this demo. Uh, I'm Jasper. I'm the founder and CEO of Artisan. We make AI employees that we call Artisans, and we create the software for them and human teams to collaborate in. Um, our first is Ava, an AI BDR. You might have seen her face haunting your dreams or in the billboards in the city. Uh, but today we're going to go over what Ava actually does, how we generate responses for our customers, and how you can actually deploy AI SDRs successfully. So I'm sure we've all seen the shift that we're currently undergoing from these fragmented manual stacks to more consolidated automated outbound software. The days of having a sequencer, a data tool, an enrichment tool, a copywriting tool, ironically also called Jasper, and a bunch of other stacks taped together into one cohesive end-to-end solution, it doesn't make sense. If everything's fragmented, you don't know what's driving ROI, You're not able to track how each thing is impacting your response rate, and no single tool has accountability. We've built Ava and we've built Artisan to be accountability first. In our UI, you can literally see the cost per lead that you're paying, the cost per meeting generated, and all of the underlying credit costs to achieve an outcome. With some other products, you don't really know what the ROI you're getting is. So the contract sits there unused and not generating real value for you. So one of our customers is Saastr, as you might know. They've sent thousands of emails in one 6-week period. They sent 7,000 emails, probably to some of you as well. They achieved a 3.6% positive response rate. They have automated a lot of their outbound and the majority of their outbound motion with Artisan and with Aeva, and this was all with Aeva 1.0. Aeva 2.0 has a lot more capabilities and functionalities that we have literally been launching over the past few weeks. But yeah, they've generated hundreds of thousands of dollars of revenue all by using Ava to automate their outbound. And the difference is really in the quality of the messaging. When you're running a sequence, even if you enrich a list with clay and run a sequence through Outreach or Instantly or whatever, you're getting these templated, relatively generic messages. And even if you use AI snippets for personalisation, they're never really that good. We've built everything around getting as much data as possible on the prospects collating it all together into a comprehensive profile view of that lead, and then generating messaging that's actually coherent, cohesive, and convinces them to buy our products. And I'm gonna give a live demo in a second of how all of that works. We have a video that we're gonna go to now, which is the launch video. We're launching officially Ava 2.0 next week. We did a marketing launch last week. We're gonna do a preview of that video for you all now. Yeah. Fernie, we're gonna play that.

Don't hire humans.

Why do I need you when I can get a robot?

And then we hired the Wolf of Wall Street.

Then our billboards broke the internet.

Stop hiring humans.

Stop hiring humans. I started my first business at 7 years old selling candy out of my bedroom. 17 years later, we've raised $36 million, and today we're launching Ava 2.0, the first autonomous AI VDR. Over the past few weeks, thousands have joined the waitlist, and today we're dropping it. Anyone can sign up today, no credit card required, $300 free credits. This is your average BDR. Let's call him Sam. Sam is drowning in manual work, juggling 8 different tools, some requiring a PhD to operate, taped together by a GTM engineer. And that's why we built Ava 2.0. It's 4 AM, you're asleep, but Ava is watching for intent signals. A CRO just posted about pipeline problems, and Ava picks up on it instantly. She analyzes the prospect's website, notices 15 new hires this quarter, reads every post they've ever written, and sends hyper-personalized multi-channel outreach. Hours of human work done in seconds. When a CRO replies, Ava handles every question an objection, and books the meeting on your calendar. No human in the loop. Now imagine thousands of these conversations running in parallel, booking meetings on autopilot, with Ava continuously running multivariate tests at scale to improve conversion rates. This is your entire sales stack replaced by one AI employee. And remember Sam? Ava booked too many meetings, so they had to make him an AE. Finance are not going to be happy when they find out about this, so claim your $300 of free credits today while you still can. No, we're taking that out. No, keep that in. Get rid of him. But that's too much money. Listen, I know— You've burned out a company. You've never built a company. Claim your $300 of free credits today at artisan.co. Yeah, so as you can see, finance were very angry about the offer that we're doing next week, so you got to get ahead of that. Now we're going to jump into a demo and show you guys what Ava 2.0 and Artisan 2.0 actually looks like and what you're able to do. And the, the real drivers of outcomes when you're doing outbound are 3 things. It's who, what, and when. Who you're reaching out to, the quality of the data, what you're saying to them, and when you reach out and if it's timely. For the who, we've built two different databases. One of 280 million B2B contacts, probably all of you are on it. Waterfall email and phone enrichment from a couple dozen different data providers. A database of every local business globally, anyone who has a Google Maps profile, any email address or phone numbers associated, we bring together. Those two databases perform pretty well, but if you're using your own data from your CRM or from data warehouse with a webhook, that tends to perform much better because we can use CRM fields, any previous activity, any relationship you have with them to personalize the messaging. Here's an example of what our database looks like. We have data on basically everyone. We work with all of the best data vendors in the industry to bring together as much data as possible. Then it comes down to what you're saying. So we have dozens of different data enrichment sources built into Artisan. We kind of built a mini version of Klaviyo for dummies, Klaviyo for me, within Artisan, so you can pull any data in and enrich it. When you're doing it through a campaign, this is fully automated, so you don't need to use these individual fields. You just use this personalisation view. So when you're setting up a campaign targeting a group of people, you have these personalisation settings where you can choose what data sources Ava should check to personalise messaging. So you can see I can use if we have mutual previous employers, mutual investors, their work history, any career signals, company-level signals. All of this data is collected by Ava by brought into an aggregate view, and then Ava chooses the best personalization to use for each message. And you can even create your own research agents, so you can tell Ava to look for anything. She'll deploy a swarm of web research agents to find any data on the prospect and pull it in. Um, in 2.0, we've gone beyond just doing the outbound messaging, and we now handle responses fully autonomously as well. So you can turn on response handling and meeting booking, and Ava will go back and forth with the lead, book in the meeting, and get it on your rep's calendar without them having any involvement. And you can set escalation rules for Ava to escalate it to a human if you'd like to, but you don't need to. Ava can run fully autonomously. The one thing Ava doesn't do is cold call because it's illegal and humans are better at it. So we queue up calls for your reps to make within our dialer, our native dialer in the platform. So they can just come in, start dialing, and it gives them talking points for each individual person they call. So that's what you reach out to them with, and when you reach out to them is sometimes just as important. So we built person and company level website visitor de-anonymisation so you can find people on your website when they're actively showing buying intent. So this is for our website live. You can see we're tracking all the people and companies who come to our site. Maybe some of you are even on here, and we have loads of intent signals that we're building as well. So things like funding data, hiring data, we're launching social engagement signals, custom AI signals all over the next month. So stay tuned. There's a lot more coming here. And we've launched all of this self-serve. So as of later on this week, you'll be able to sign up, get $300 of free credits, get value before we ask you for your credit card. I know there's some skepticism in the outbound space, so we really wanted to put value first before we ask anyone for money. But the results don't necessarily rely on when you reach out. This is one campaign that I ran reaching out to YC founders in San Francisco, and you can see I coached the messaging to make it sound really human. So having lowercase, using two exclamation marks, you obviously— this isn't the standard messaging because most people don't want it, but by coaching Ava, you can get her to replicate any writing style that you would have your reps use. So this was just a mass cold outbound campaign, but it looks very human and it's pulling personalisations from their social media posts, from our mutual connections, from a bunch of things like that. And then we can see that this campaign wasn't targeted on the when, but because the who and the what were good, we achieved a 4% response rate. Response rate, pulled in a bunch of leads, got a lot of leads coming through to the platform from that. So we've consolidated the entire outbound stack into this one end-to-end flow with everything centred around campaigns, email outreach, social outreach, dialer analytics, monitoring the cost per lead for each of your campaigns on your analytics so you can really know what you should and shouldn't be investing in. Everything's consolidated together into this one easy-to-manage view with a fully database list enrichment, everything just consolidated end to end. We're going to switch to questions in a second. There's, I believe, two mics over here. If anyone has anything they want to know about Ava or Artisan, please don't be shy. You can walk to the middle and ask anything.

Hi, quick question. So if I use your product, I don't need ZoomInfo?

You don't need ZoomInfo, no. No, we have a database of basically— there's a lot of trashy data in ZoomInfo. We had a database of 450 million, we trimmed it down to 272 million to get rid of the lower quality contacts, but they all have verified information. We waterfall enrich emails and phone numbers from a couple dozen different data providers. And we bounce test the emails before we present them to you, so every time you're getting fresh, accurate data. Yeah. Hi.

Hey man, I'm just wondering, so I'm someone who pieces together Apollo, Instantly, you name it, right? I'm trying to figure out, I see a lot of companies here consolidating, but what is the vision beyond, or what's the endgame beyond just consolidation for you guys?

Yeah, so the biggest thing for us is automating as much as we can so you don't have to manage a stack. So for example, with campaigns, Ava comes up with campaign ideas for you and Ava suggests audiences for you. With the messaging, you don't have to come up with the messaging, you just put in your website and Ava generates the messaging for you. With the campaign settings, you don't have to respond to leads, Ava responds to them fully autonomously, Ava books in the meeting. So wherever we can, we'll take you out of the loop so you can just let it run on autopilot and not have to manage the product. So the main thing for us, even over consolidation, is automating as many of the workflows as we can. And we're going to release a full self-driving mode for Ava where you can basically just turn Ava on and then it'll run fully autonomously in the next couple of months as well.

Thank you.

Thank you. Ada, you mentioned earlier that you think humans are better at cold calling at the moment. Yeah. In the hypothetical scenario that it wasn't illegal to do it. Yeah. How long do you think until you might compete in that space, or people might try and do that from an automated perspective? So it's not illegal for inbound, it's just illegal for outbound. A lot of our competitors do the inbound calling. We haven't dealt with it yet because we just really want to focus on, on outbound. And you can tell when you're speaking to an AI, like, you— if you just ask a question and it like takes— the latency's a little bit longer it gets the answer slightly wrong, it doesn't quite get the nuance and context of the conversation. I think it will probably be a good couple of years before it's really as good as a human. We, we don't want to build anything in, in a space until we're confident that we can do it super well and flawlessly. And without outbound, you can never do that just because it's illegal. So yeah, we focus on building a dialer. In the dialer, we generate talking points for each person and we guide the rep on what to say. We draft the follow-up email, we handle everything except for actually speaking on the phone.

And with respect to inbound, how do you think?

I don't know. So we've purposely avoided inbound for now just because I don't think it's good enough yet. Um, the live chat on the website, you can get very strong, but when you get to like the video interaction, we have a partnership with Tavis, the AI replica company, to build humanoids. And we have one part of our onboarding journey next this week when we launch Ava 2.0 fully publicly, you can see you can talk to Ava and it's a humanoid video that talks back to you, but she doesn't really take any actions yet and it's very early. I think we're years away from getting something really, really strong in the humanoid back-and-forth interaction, but it's early and promising.

Thanks.

Thank you. Yeah, y'all.

What about LinkedIn?

Do you think you're going to LinkedIn at all? I plead the fifth. They don't like us. Yeah, you have to find out for yourself. Yeah.

Hey, we are in the voice agent space as well for healthcare AI.

Yeah, non-clinical use cases mainly, mostly for the reasons that you talked about.

We do both outbound and inbound on behalf of providers.

Yeah, and payers also more recently. Recently.

Um, do you have your own models for voice AI, or you are using state-of-the-art models?

We focus on, on text outbound. So we, we analyze calls and we draw follow-up emails and we do the outbound over email fully autonomously, but we don't actually do any calling yet.

Okay.

Uh, yeah, we work with Anthropic for all the email stuff.

Yeah.

Okay, sounds good.

You said, uh, that you are able to detect, uh, humans.

We are not close to humans, so that's why I was wondering that.

That's where this question comes from.

Yeah, I mean, we use like Cartesia, 11Labs, pretty good, but then there are—

Yeah, there's like edge cases where like if you have a truly high-value lead, you probably would just want to pay someone to deal with the call.

All right, I'll meet you offline.

Yeah, nice. Hello.

Hey Jasper, I'm Josh from Claim Your Aid. Great demo, sorry about that. Makes perfect sense. Yeah, €300 credit, why not try it? Curious, how do you charge after that? Is it outcome-based? Is it per user?

We switched recently, so we now charge with a credit system because our product now does so much and we cover lead discovery enrichment. We cover basically every component of the GTM stack and you could do each one individually. We charge a fixed number of credits. One credit is roughly 2 cents for different actions. Like enriching an email costs 2 credits for Waterfall enrichment, and we package everything together into the cost per lead enrolled in a campaign. So usually you're paying like roughly 30 to 60 cents per lead enrolled into a campaign, and then on the analytics page we track the cost per lead, which is the cost per positive outcome, so you can see the amount you're paying in credits to get each positive response or meeting booked. So we try and, to the user, present everything in terms of the ROI you're getting from from a specific campaign.

Can you focus it down that you don't want it to do certain things, just book meetings and don't enrich, don't do the other stuff?

Or you kind of have to do the other stuff to get the meeting. So like as part of a campaign, Ava will find the leads for you. We don't charge for that. We do that for free. But enriching the email addresses you need to get the meeting, enriching the data for personalisation, like finding social media posts, finding signals, all those different things will be done by default as part of a campaign. But you can get the cost for you. You can get it down to like $0.20 per lead you're contacting if you do like a more bare bones, like 3 email campaign. Yeah. Thank you.

Yeah.

Sorry.

Hey, my name is Ken. I think Jason M. Lemkin talked about how it takes, what, 1, 2, 3 months sort of dial this in with training. What is best practice these days?

Who's—

how many people are dedicated to this? What's the profile of that person?

That's something.

Yeah.

Yeah.

So we've had in some cases where people come in on day one, they're doing super well. Other cases where they come in, like CookUnity is one of our customers, in their first 2 months they got horrible results, like terrible, terrible results, but they stuck it out and they kept on iterating and they now achieve like under $50 cost per meeting booked. So sometimes you'll see really strong immediate results. We're trying to kind of liken it to outbound market fit. So like you don't necessarily, even if you have product market fit and you have good customers, people still might not reply to your emails because they don't, they don't like your products, in which case you probably have to do warm and CRM-driven outbound rather than net new cold. But if you just iterate on the messaging, iterate on different targeting, iterate on the who, what, and when, you get to a recipe 90% of the time that works well. Some companies just, unfortunately, people don't want their products, but that's not very often.

Hi Jasper, it's Michael Bertoni Jason M.

Lemkin from SaaS Talent. I had a question about LinkedIn specifically in the businesses.

Yeah.

Are you— is Ava sending messages through LinkedIn and does she send sequences like through email and through LinkedIn? So my specific question was around LinkedIn.

Yeah.

And then does she have the ability to send sequences like on email and can we— can Can you just dive into maybe that, the LinkedIn and then the sequences?

Yeah, so the emails are fully autonomous. We handle the entire outbound from your primary domain, or we also have secondary domain infrastructure built into the platform so you can scale to very high volume. LinkedIn, we draw off the messages, we do the research for you, we pull everything into the platform, but yeah, we don't fully automate the LinkedIn component yet. We're speaking to them right now.

We have to send that. So you populate and then we have to send it.

Send it?

Yeah, we have a system for it built in. Yeah, that makes it super easy.

Thank you.

Yeah.

Hi Jasper, Jay Henderson from LeaseWeb. So I have a couple of questions. One is in and around how do you derive the messaging for like different CTOs and the different type of people? For me, I'm outbound, so yeah, I want to better understand how I can use Ava to do that. And then on the other side is when we're thinking about LinkedIn, when we're out outreaching on LinkedIn just person to person?

Yeah.

Recognizing when we're, you know, looking at their, their stuff. Is it going through Sales Navigator in comparison so that way we don't get so recognized when we're, we're creeping on people's LinkedIn profiles?

Yeah, yeah, yeah, yeah. So we don't pull data directly from LinkedIn. We have our own. We work with a bunch of data aggregators to pull things in. So it's not like your profile is going on a bunch of others and creepily viewing them and looking get data from them. With personalisation, we pull data from more sources than any other platform, so we have dozens of different data providers that are all pulled together. Any intent signals, any social posts, anything we can pull in that's publicly available, we will. We even have research agents that will go on the web and find data for you, and the personalisation that's used is also used based on the context of who you're reaching out to and what the message is about and if it's relevant. So for us, if they're hiring for a bunch of BDRs, Ava will usually choose that as the personalization, but for you that obviously wouldn't make sense. So it's not chosen in isolation, it's chosen with all the context of everything that's happening around the personalization.

So we can personalize it based on who our ICP is?

Exactly. And you can also put in specific personalization research agents. So if you are reaching out to CFOs at public companies, you could say, what do they say in their recent 10-K report? So, or anything like that. And then Ava can go and find that information on the fly each, and it can be whatever you want. That was just a— yeah, natural language.

Like Apollo. Apollo did something where you could do automatic, like, outreach where it would bombard people with messaging, right? Is that something that Ava would be doing, or is that kind of banned?

Do what?

Sorry.

So it would be like an automatic email that would go out rapidly to different, you know, uh, contacts within your LinkedIn, as an example. So you'll be prospecting, right? You're going through that, and you could send based on the people that you choose.

Yeah.

An outreach message that would be pre-populated as an example.

Exactly. And then we prioritise based on who's the most likely to respond and we have a backend algorithm for that and we do everything we can to maximise positive responses.

Okay.

Yeah.

Thank you.

Thank you.

Hey Jasper, this is more of like an Artisan question, but I feel like I see Ava, Artisan everywhere.

It's kind of like the marketing strategy behind you guys, the company.

I know it's a broad question, but I think, I guess the strategy for us is get the eyes and then convert the attention into pipeline. Step 1, get attention. Step 2, push it further down the funnel. And we try, we've tried a bunch of different things. Like we did a bunch between our two big more controversial campaigns that were less controversial. It just doesn't work as well. I'm sure we'll evolve over time into more friendly brand-safe marketing, but yeah, it works really well for us to just get get the eyeballs, get people on our website, and then build CTAs in and product education and social proof to push down funnel. Especially because we're in a pretty crowded space. There's a lot of small players doing different components, so we just try and be noisy.

Hi, you mentioned that the lead discovery part is free. So is your model essentially usage-based?

If we get good leads, we pay for enrichment and sequencing?

So the actual finding of leads and Building a list is free. We charge for the enrichment data and for the personalisation data and for executing the sequence, but in the product, the analytics are centred around how much you pay per lead that you generate. So everything's centred around that.

Got it, because we have tried different tools in the same area and what happens is if it fails at lead discovery, the rest of it doesn't matter.

Everything else, yeah.

So that's the most critical component and we end up paying for something when the leads are not so great. So that's what I was checking. So it's after the lead generation.

Yeah, you don't pay for the actual building of the list and the lead discovery.

Thanks.

Awesome. Thank you. Thank you everyone. Appreciate it. Chat soon. Thank you. See you guys.

Let's give a warm welcome to the CEO and founder of Light Field, Keith Paris.

Hi everyone, thank you for coming. I work at Lightfield and we're building a natural language CRM. We've got about 3,000 customers and today I wanted to do a live demo of assembling your CRM in real time, you know, exploring a stalled deal, and building an automation based off of what works, and then finding new customers based off of what we learn from a single customer. So we're going to do this all in real time with real data using Claude. So, so bear with me. And after that, we will go through a bunch of questions for anyone that's curious. So like I said before, we're, we're going to have a 4-step demo today, starting with setup. Unblocking a stalled deal, automate what works, and use the learnings to go find more customers. So on that note, let's get started. And just so everyone knows, we're going to be one of our own customers here. You know, one of our own customers is building an AI process automation company that helps large manufacturers, you know, save time and money. So we actually built a synthetic dataset off of this company, and I'm going to be sort of roleplaying this as I work through different parts of their funnel. So on that note, let me open up Lightfield. So show of hands, how many of you have used Lightfield before or have seen it? Okay, we've got a couple here. I'll start by saying everything that's in this demo instance, uh, was put in automatically, you know. And what I mean by that is I connected my mail, calendar, uh, data warehouse, and, um, and, um, and call recorder, and the rest Lightfield took care of itself, you know. So I've got up-to-date accounts that are enriched, um, you know, using a couple of different providers. I have opportunities that were automatically created off of my emails and my calls. You know, I've got a contact book that's been filled out by about 10 different vendors. And then, you know, I sort of have tasks and meetings here. Well, no meetings just yet. And, and this is sort of the like base of my customer context engine that I'm going to use to, to sort of build a go-to-market on top. So, you know, on that note, let's try to work a deal. We are selling process automation to actually this very large company, Johnson Controls, and I'll pop this open, and you can see that a lot of things have happened in this deal. I'm sort of scrolling through our timeline of information. We've done some lead scoring, we've updated the opportunity, there's been some email back and forth, but if I read the summary here in Lightfield, You know, it looks like the deal is stalled. You know, we've been working with the VP of Innovation and it looks like we've been trying to scope a 30-day POC and we just haven't been able to, to, to, to, to move the deal forward. So before I build any automations, before I build my go-to-market, I'm going to ask Lightfield what to do here. So I'll just pop this open and ask, you know, "why is the deal stalled?" And here Lightfield's gonna do a couple of interesting things. One, it's going to look at every interaction we've had inside of this opportunity, but it's also going to compare what's happening here to our close-won, close-lost opportunities, and even be able to look forward into QBRs. And you can see it's sort of like running code in its sandbox, being like, what are the similar deals here? Are there any patterns we can learn from? And, you know, sort of through this analysis, we'll, we'll learn what Lightfield's best take is on how we can unstuck this deal. And you can see that we, we've got some interesting analysis here. So we sell process automation and it looks like we haven't built a relationship with IT You know, more specifically it says that we haven't met the CIO. So now Lightfield is starting to build proof for this, right? It's looking through my deals, it's seeing every deal that we've won before where we typically got, you know, a head of IT or a director of IT in early, and then some lost deals where we didn't get IT approval early. So now we're stuck. And Lightfield has an action. You know, the action is try to get a meeting with the CIO. You know, it looks like our champion has mentioned him a couple of times, but we just haven't been able to get the meeting. So you know what I'm going to say? I say find Vijay and add them to the opportunity. And here we're going to run through a couple of web searches, run through about 20 different account and contact enrichment tools and find this contact, add him to the opportunity. You can see we're doing a LinkedIn search, and we're already, you know, creating the contact and associating him. There he is, the, the CIO on our enterprise manufacturing optimization. So I'm gonna hit create. Great. And now, now that he's in the CRM, I'm I'm just going to send him an email. I'm going to say, "Draft an email to loop Vijay into the POC." And here we're going to go look at all of our interactions with the champion, with the facilitators, with other heads of IT. It's going to do some research and draft an email in real time. You can see there's very little tool calling. It's written in my tone as the sort of founding AE at this company, and, you know, it's sort of brought in very specific details about our Milwaukee POC, some context, and a little bit more to be able to unstick this deal. So I'll hit send here, and, you know, now I've hopefully unstalled this one deal. And, uh, it's an interesting reminder now for all of the other reps at my company. We should make sure to loop in IT earlier before we get to this POC process, right? So, um, I'm gonna turn what I just did into an automation. Um, and, uh, and this is the really exciting thing about working in Lightfield. You, um, you can build automations and you can build workflows as you learn from, from one deal at a time. So I'm going to say, run this process every time a deal reaches the POC stage without an IT contact. And now we're building a natural language automation. You can think about this much like Apex, except that you get to write them and iterate on them with natural language. We actually run this in Python in the background, so it's a far more powerful platform than a traditional CRM. And, you know, now we've got our automation. It's sort of pulling out what parts of the CRM data it's going to use. It's pulling out the logic. You know, it shows how it's going to find the new IT contact using our web search tool. And then at the end, it's going to draft this email to try to get time on the calendar. When we move from, from one op to the other. So that seems like a pretty good automation. I'm going to hit Activate. And, you know, as I hit Activate, Lightfield is going to recognize that you need a bunch of different automa— different permissions to run this automation well. Right. Which is probably something you're running into. If you're the RevOps leader, if you're the CEO, you might have access to every part of the CRM. But if you're a rep and, you know, you don't have access, we're going to ask you for permissions. We obviously need to access everything in the account. We need to be able to create notes. We need to be able to use our enrichment service. So I'll hit Approve and Activate.

Great.

So now we've unstuck this deal. We've built a process automation to make sure that reps in the future will run this process. And now I think we've got an interesting hypothesis here, right? We actually have a good value prop for heads of IT, which is that we, you know, we help them with uptime, you know, and more. So now let's try to use what we've learned to prospect. So let's ask Lightfield, you know, show me— or what patterns— oh, caps lock, my bad. What patterns in our closed won deals can we use to generate new pipeline? I'll close this. And now we're obviously gonna go look at the context of everything that we've spoken about with Johnson Controls, but we're also going to go look at all of our closed won deals and see, you know, is there a there there with prospecting directly into IT, or do we need to to find another approach here. And here are the findings. One, much like this very large deal we're working on, if you're a big industrial hardware manufacturer, we can really sell you on this downtime figure. That's where the real pain is. The other thing is Lightfield agrees with me that IT leaders are actually very receptive to this value prop based off of our QBRs, our existing sales engagements, so it looks like there might be something there. And, you know, it's also telling me to scope the POCs more tightly, which I should go back and tell my sales team. So now let's turn this into a prospecting campaign. I'm going to say, actually, one more thing. Let me check, find contacts at 10 companies with this profile. And this is sort of the beginning of an automated sequence where we found this criteria for contacts, we found this criteria for companies, and now we're going to go and find new companies to sell to. And you can see that we're running our account discovery skill over here, where it's reading through all of the accounts in the CRM we haven't sold to. It's sort of doing like an ICP search in the background using all of the different data sources And it's also reading through a lot of unstructured data, you know, of every deal that we've tried to get that we didn't get, you know, every deal that stalled. And now you can see it's like found a list of companies and it's looking for the IT leader, you know, in each company. So we'll come back with some results. You can see that they're cross-checked, you know, across 2 or 3 different sources. We're obviously looking into LinkedIn and now we've got some matches. So it's going to create the contacts and it sort of follows good CRM hygiene, you know, which is that if you prospect to someone, they should be in your CRM and they should be connected to ops. So now we're going to create these contacts and you can see it's sort of writing code to do so at the moment. And as soon as the contacts are in the CRM, it's going to be able to build a custom sequence using everything we know about similar companies. So there we go. We've got 10 different heads of IT at these like very large manufacturing companies who might be interested in our product or service.

Great.

Here are the 10. I'm going to create all these 10 contacts. And it'll create the sort of accounts associated with them, so I don't need to think about it. And then we'll do one more filter, and based off of everything we've, we've learned, we'll be able to create a custom sequence based off of everyone we know. So I'll do one more filter, say show me which of these companies are already using legacy factory factory floor software. And why am I saying this? I think there's a lot of pain if you're using legacy factory floor software. We might be able to find your pain with downtime, and then we'll just have a better path in. So now I'm asking Light Field to run this huge web search across these 10 different contacts to better filter them before I write a sequence. Candidly, I'm also doing is to shrink the time a little bit so we don't have to, you don't have to watch me write 250 sequences at the same time. So you can see that we're doing a search for SCADA, we're doing a search for PLC, we're doing a search for some names of some legacy software here to see how many of them are good matches. You can see that we actually have a range of signals here. It's brought up this aerospace company where we have really high signal. They're sort of using SCADA. We've pulled the signal from job postings. Job postings are always a great source of information for prospecting. You can see in this other one, someone made a LinkedIn post. Mentioning this. And then there are a few others where, you know, we've got lower signal based off of some signals and heuristics. So I'm going to, I'm going to add these companies, even the low signal ones, to a sequence. And now when I, when I say new sequence, Lightfield is going to write a create a custom sequence for everyone that fits this criteria. It'll learn from our previous sequences, it'll learn from the research we just did, you know, and it'll also learn from my sort of writing style. And all right, let's just say just the high signal ones for now, and we'll sort of see what it comes up with. Interesting. It's coming back with, "Who's writing this? What's the tone or the angle?" You know what? I think let's go with IT engineering time. That seems like a pretty good angle, and include the proof point from KONE. KONE is a giant elevator company. We work with. It was able to sort of find a proof point from the QBR, and I guess I didn't give it direction on tone, so it should follow the tone that I've embedded into the system using my knowledge. So now it's gonna read our sequence skill that's— that we sort of built into the platform and build a custom sequence, right? So the goal is to book a 15-minute meeting with an IT leader to discuss their architecture. It's chosen, you know, the highest priority leads we found from our research, and it's built a 3-step sequence. So the first one is to sort of lead with legacy burden. The second is sort of a bump on opportunity cost. And then the third is just leaving the door open. So let's, let's run with it. And you'll see Lightfield in real time build the sequence automation. So that's, you know, we can effectively outbound using all of our CRM data. And here's the sort of sequence map, and you'll see that it's a little bit different than a traditional sequence, right? Which is to say that it's, it's not just variables and parameters. There's actually an approach that goes into it, research that it sort of does in the background, and then it can sort of re-engineer every part of your outbound to be more effective. You know, it says to use Cone by name. You can obviously modify how long you wait. And now we've got a preview. You know, let's, let's actually run a full preview for— show the full preview for Tano before we activate it. It'll be a little bit different for everyone in the list, but let's just take a look. All right, Hannah, I'll be brief. It brought up some feedback we got from the research, and it gets shorter and shorter. It all seems pretty good, and you know, it also caveats with these are disposable illustrations, so let's activate the sequence. Great. And we're good. So the reason I wanted to show all of this live was that we connected our CRM to email, we looked into why a deal wasn't— was stalled, we learned something through that process, and then we we turned that into an automation and ultimately used what we learned to better inform our outbound campaign. So this is just a little bit of a vignette of what's possible when you have complete customer context in a single spot like Lightfield, and when you have an automation framework sitting on top. So I'll pause there and see if you have any questions. Please, yes.

Hi Keith, thank you for the interesting presentation. I'm interested in AI-native CRM and wondering that, you know, the structure used to be very important for CRM, but seems like everything is like automatic and free. So I'm wondering with the data governance, yeah, like I, I don't, I'm I'm not sure if this happens, but for example, some people overwrite some important information unintentionally. And can you trace back or can you secure the previous information?

Yes.

So our approach to data structure is one, we store everything at a foundational level, you know, emails, events, and more. And then we have a CRM schema that sits on top, which we have fields for, and we actually track the edits and the version history for every field, every attribute, every object. So we can absolutely see if an agent or a human overwrote something, changed something, and sort of fall backwards and sort of rewrite. The other thing is that we have role-based access control on every piece of data in the product, so you can always lock something out and, you know, have it follow your sort of enterprise's permission structure.

Thank you very much.

Of course. Yes.

Um, okay, hi, I'm Sindhu Srivastava. I am the CEO of WeCrush Events. So it's an event planning company, 8 figures in revenue, Inc. 5000. I have about 13 reps or so using our CRM. We use Zoho right now, and when I look at this, I'm like, okay, this is night and day. And my, my biggest question is, how do I get 13 not technically motivated people to adopt something new? Like, what has that looked like for you in your customers' adoption?

Yeah, that's a good question. First, I would say we're really fast with migration, so we tend to be able to migrate someone off of a Zoho or a HubSpot in 2 hours. So we can have them playing in the new tool very quickly. And then I would say the thing that reps really love is that they don't have any work to do anymore. All of the CRM fields are updated, all of the forecasting works, all of the reporting works. That's sort of the carrot, that they don't have any CRM work to do anymore. And the last thing I would say is we do a little bit of training. It's usually 30 to 45 minutes. 5 minutes, and anyone that's used ChatGPT can very quickly, um, sort of figure it out. So, uh, you know, I'd love to chat with you later to, to figure out something for your team.

Yeah, one more question, right? So right now, um, our Zoho uses our traditional email domain, and then obviously we were sending out a lot of email and that didn't go so well. So now we have like a cold outbound where I have 50 inboxes, 3 emails per inbox, that sort of a design. As reps are sending out, say, mass emails, like I think we have about 200,000 contacts in our CRM, so as reps are sending mass email, like how are you controlling for email quality and deliverability?

Yeah, it's a good question. We do our in-house email warming and we can distribute across different emails. So just sort of like any, any frontier outbound company. And then separately, we have very intelligent sync rules for what you want to bring into the CRM. Like I would say for most, most reps that use Lightfield, their outbound only syncs to Lightfield when someone responds. So you don't end up with, you know, 500,000 contacts in the CRM. Them, it's only when a contact is real and you've got a meeting booked. So we've got a bunch of intelligent ways to sort of keep your accounts and ops clean while connected to all of your outbound accounts. Yes. Hey, first of all, I really thought, I thought the Agentic workflow is really robust and great job on that. My question is around metrics, How, you know, I guess we didn't get a chance to see that part, but how do I know, you know, if I'm testing 20 different combinations of ICPs and angles, um, how does metrics look in the dashboard? Oh yeah, um, I could show it to you after, but we, uh, we have a whole code sandbox in Lightfield, so you can generate any custom report you want on demand, um, and then save it. So you can build an interactive chart for each each of your campaigns and look at the same one across to sort of find performance.

Awesome, thank you. Hi, I have a question about security.

Yeah.

In the demo, I believe the AI was generating and executing code directly in the chat panel, and since the agent is freely generating code, how are you ensuring security? Yeah.

Do you limit the actions that agent is allowed to perform? Yeah, it's a good question. So the agent, external systems, and the UI are all running through the same Light Field API, and we're able to sort of control, uh, rate limits and data access through that API. So if I was, you know, executing code as an AE I would be bound to the information access and rate limits as an AE in the system. That's absolutely enforceable at the admin level.

Thank you.

Of course. Great. If that's all, thank you all for coming. I wanted to share a couple of things. One, we don't hide our product. Behind a sales team. So you can go to lightfield.app and get started via self-serve. There's a 14-day free trial, and then if you need help, you know, our sales team is happy to jump in. And then separately, we've got a, uh, the big white booth over there, um, and, uh, you know, we, we like to show off how powerful our migration is. So if you are open and willing to migrate your data from a legacy CRM in a couple of hours We've got some, some rewards for you and some incentives so that you can ultimately build, build the workflows of your dreams. Thank you so much.

Give a warm welcome to the CEO and founder of Attention, Anis Benasour.

Hi everyone, I'm Anis. I'm the co-founder and CEO of a company called attention.com, which also happens to be the coolest domain name that you've seen out there. You've probably seen planes with our name outside. What do we do? We're the platform that automates increasingly intelligent work for your sales team to help you drive more revenue. I'm a repeat founder obsessed with go-to-market and AI. Um, we're a Series B startup around $15 million in ARR, growing 4 to 5x year over year, and we work with companies like Scale AI, Lovable, Abridge, engine.com, Netlify, and many other amazing logos. Today, I will share with you some secrets with you on how to grow your pipeline. Today, the problem is simple. Growth is stalling, and you probably want to have some viral loops, right? But the thing is is that your customers, your prospects, when you speak to them, they're actually the ones handing you all of the secrets to grow future pipeline. And so this is what I'm gonna talk to you about today. I'm gonna go through 3 examples, 2 of which I'll present on the slides, and then the 3rd one will be a live demo. The truth is, if you actually are not using this, you're losing most of the intelligence that you're capturing from your conversations, and therefore you're not able to compound your growth. Let's start with the first one. Today, when you think about it, your markets are moving way faster than ever, right? You have competition all over the place. You have, you know, things that are becoming obsolete in the next few few months, there are so many things that you need to be on top of, and these things are getting captured in your conversations. So the first thing that you want to do is always refresh your ICP, right? You might have a wrong definition of your ICP, and even if you had the right one, this might be outdated in the next month or quarter. So what do you want to do? You want to build this agent that very simply either every month or every quarter, goes and gets your most recent closed won deals in whatever CRM you're using or call recording. Then you analyze, you get it to analyze all of the conversations. You ask it all sorts of questions about the buyers, right? What is their function? How long have they been at the company? How big is their company? Where is it geographically located? Basically all of the firmographics, technographics, intent data, anything about the buyer persona as well. Then you use a top-of-funnel tool like Clay or DeepLime or any of these that goes and deeply looks at all these personas, gets as many characteristics about them as possible, and then gives you hyper-detailed profiles, right? So if you look on the right, Obviously I anonymize a lot of things, but what it will tell me is, hey, this is a persona which is the frustrated veteran that's been using an incumbent. This, these are all the things about them. These are all the things, you know, how long they've been in the role, etc. And you start now from going from like a broad definition of your ICP to all these different pockets of ICPs with exactly everything about them. Who they are, how long they've been in the role, so that you can, when you go out and reach out to them, you're actually targeting them with a very, very, very relevant message. Let's go to the second thing and zoom out. This is over at Attention, we built our own go-to-market machine, right? Here, I'll move so you guys can take photos. This actually compounds and the, the, what, what actually feeds it is your prospect conversations, right? And so what I showed you just before here was the lookalike targeting and the exact verbiage that you're capturing from your prospect conversations, but the really fascinating things now is that you can create synthetic personas based on all of these that will actually go out and be used later to predict the reply rate in your campaign.

Gains.

You mix that with a probabilistic model and it actually gets really, really good. And so what we built internally is that for each person that you're going to actually reach out to, you're not only going to get everything from the conversations that you might have had with them 9 months ago, 12 months ago, or ever, but also you're going to do deep research online with your agent. You're going to just rank all the information, get the AI agent to write your email and then you'll predict the reply rate. If it is actually, if you had to send 100 similar emails, it will actually tell you, all right, are we over 5% response rate or below? If it's, if the model tells you it's below, then it goes back and it iterates and iterates until it gets it back to 5% or more. And then it shows you the message in Slack and you just validate or invalidate it and then it sends it, right? How did we get to do that? We actually find all the alpha in the prospect conversations that we've captured before. The third one now, I'll move to a quick demo. It's really simple. We actually use attention here. Attention with a proactive agent has the ability to do a few things for you. A, obviously better execute on your deals. But B, this is by the way, I filled it with other, with random data, but it will better execute on your deals, re-engage the prospects that are ghosting you, and also re-engage people who, for example, you closed loss 9 months ago, right? It gives you proactively all the things that you should know about your deals at risk, your your prospects that you need to actually re-engage and so on. It even builds your own decks. We use it for everything, by the way. We use it for sales, pipeline generation, account management, customer success, and so on. Just to get everything done, reps just log in, know exactly what they have to do, and it takes all the actions for them. Oh, can you go to the next slide? So yeah, if you look at today, what all these things that actually get done for you are getting captured out of your conversations. Your conversations help you best execute on your deals and your pipeline. They help you also optimize your outbound, and also they help you consistently refine your ICP. I think we're right on time here. Thank you. All right, any questions? Please don't be shy. Yes. Oh, do you want to come here? I can't hear you. Sorry.

Yeah.

Yeah, Gong is a call repository platform with some insights. Attention will actually get the actions done for you. Right, so give you proactive things that you have to do to better either win your deals or re-engage your pipeline. Yes? Yes, it is. Well, this is actually our newest feature, proactive. We'll set it up for you. Any last questions? All right, thank you everyone. Oh yes, sorry, I can't hear you.

Do you have any template you recommend to build in order to take advantage of every conversation you have and later on improve your, yeah, your ICP or whatever?

Yeah, of course you should use us, but also you can connect with me, add me on LinkedIn. I'm happy to also show you how you can build it. There are also some great RevOps people out there that are sharing a lot of great content, like Brendan Short, for example. Who are writing really, really good stuff. Brendan Short, like the word short. Awesome, thank you everyone!

I'm not sure it's gonna work, but I'm gonna take off the GoPro for now.

Good afternoon.

Check one, check two.

Sorry. We can also test the audio like we did on the booth. Okay. Hey Piper, can you tell me about what you do? I don't hear anything.

That should be—

oh, he hears it. Okay. Can you tell me one more time, Piper?

Sure thing.

I help you explore how Asana brings people—

I'm good.

Okay. Awesome. Yes. Mm-hmm. Yep. Okay. Ah! Let's give a warm welcome to the Vice President of Demand Gen from Qualified, Sarah McConnell. Okay, everyone, thank you so much for joining me for this session where we're going to talk about driving inbound pipeline, 3 extra inbound meetings by building and deploying an AI agent. I promise the majority of this is going to be demo behind the gear of Qualified, but I'm going to set up some context first. My name is Sarah McConnell. I'm the VP of Demand Generation at Qualified from Salesforce as of a couple of days ago or a couple of weeks ago. Okay, so before I jump into any sort of demo, I wanna set some context about foundations that need to be in place. It's really easy to see really cool new tech and think like, yes, I need that. But there are some things you're going to want to do first before you even evaluate these tools. And then even when you do evaluate them, there are gonna be things you're gonna do beforehand before you launch this AI SDR. The first is where do you need coverage and do you even need coverage? So do you need an AI SDR to cover after hours? Like, do you have a predominantly US-based SDR team, but you're getting leads in from other geos, other languages? You want to determine where you need that coverage on your SDR team. Or alternatively, are you building an SDR team from scratch and you want your AI SDR to handle all inbound? Set that ahead of time and understand what your priorities are before you launch anything. Second, and the most important, and I cannot overstate this, is onboard and train them like a human SDR. If you just put in an hour and you set it loose on your website and then you can't understand why it's not responding in brand, it's not answering the questions the way that you'd want them to. You have to onboard them. You have to give them context. You have to give them access to the materials that your human SDR would have. The beauty of an AI SDR is there's no turnover. Once you give them that context, once you train them, they can do this infinitely at scale. And you're not worried about rehiring and retraining over and over again because SDRs tend to be a role that has fairly high turnover. The third is set appropriate targets based on your use case. And what I mean by this is if in step 1 you determine, I really want an AI SDR to work after hours, I don't have anyone covering my inbound leads after hours. That's where I want coverage. Look at the amount of volume that you're getting after hours and determine your expectations based on that. And what I mean is, when I first rolled this out at Qualified, we did an analysis of how many meetings do we book with our SDRs, and we're like, we're going to hold our ASTR accountable. They booked this many meetings. If it can't meet or beat that, what is the point? But if they're not getting access to the same type of leads, they're not getting access to the same type of traffic, then you have to set those expectations accordingly. So what I did is I looked at how many leads are we getting after hours, how many of those are actually converting into pipeline, and then set my expectations from there. So based on step 1, you want to make sure that you're holding your AI SDR to goals, but set appropriate goals. 4 is determine SLAs. How do you want your AI SDR to work with your human SDRs or with your human AEs? Where are those handoffs happening? Where do you not want them working accounts? For an example, at Qualified, we have a pretty healthy sales-led motion. We have assigned accounts, we have human SDRs working those, those accounts for their AEs. We want to set SLAs ahead of time so we're not hitting them with an AI SDR message and a human SDR message and an AE message and a marketing email. You have to set that all up ahead of time and determine what the expectation is and the SLAs are before launching anything. And then fifth, evaluate their knowledge and provide coaching. I'm going to show the backend so you can see what this looks like, but you want to continually go in there and see how your AI SDRs responding? Are they saying the things that you want them to? And if they're not, provide that coaching and make sure there is a constant feedback loop happening. So with that said, I want to show you guys the process of building an AI SDR in the backend. This is Qualified. So this is the backend of our product. This is where you would build your AI SDR agent. So the first thing I'm gonna show you here is our themes. So this is where as a marketer, I would come in and determine what is my brand, how How do I want this to look on my website? So I'm gonna use Asana as our example. They are a customer of Qualified. They use our AI SDR agent. The Asana marketing team would come in here and say, okay, based on the positioning of my chat, do I, do I want my agent to be in opened mode, in pinned mode? Do I want it minimized to the right-hand side? You can set all of that ahead of time. What will this look like in different messengers? This is where you're gonna set all of your branding. Branding. Now, the main thing is done in what we call Agent Studio. So I'm going to go through a couple different tabs here and show you how easy it is to set this up from a setup perspective. But again, not to be skimped on the coaching and onboarding, which I'll show you. So right here in Agent Profile, this is where you give the personification to your agent. You're going to name them, you're going to give them a title. There's avatars that you can choose from. You can see here, this is Piper, our AI SDR agent, but there's others that you can choose here. There's also very— coming very soon is custom avatars. So if you have one that you want to customize, you want it to look like a specific person, that is going to be GA here very shortly. And then there's a lot of other things you can set here about like tones and how, how you want your AI SDR to talk. Now going to the onboarding part here. So this is where you're going to give your AI SDR access to the, the content that it needs to be effective at its job. So the first thing your AI SDR can do is it can crawl your whole website. Now, this is a cursing and a blessing in that it will pull in every piece of content on your website. And as a marketer, I've learned that there's a lot of content on your website that is old and outdated. And actually, I mean, AI-SDR is great because it shows you where that content is. And we realized we had to remove some content from our AI-SDR. But Piper is going to come in here. She's going to scan all of this content. But there is a lot of content that also lives not on your website. Things like PDFs, things like sales slides. And I'm gonna show this in an example here in a minute, but you can actually upload offline content for your AI SDR to use in sales cycles. So after you give your AI SDR all of this content and access, there's snippets here and other things. So if things don't exist, whether it's like a lot of times it's like competitive material where you wouldn't have that on, on your website, but you wanna give that access to your AI SDR agent, you can create them through these snippets. The next thing you're gonna do is guides. So guides are essentially the guardrails that you're giving to your AI SDR and how you want them to interact, both from a branding perspective and from your unique business's segmentation perspective. So like right here, Asana would come in, they would give a description about their brand. They could come in and give guardrails. So think about when you're interacting with your SDR team, there are things you would tell your human SDRs, do this, do not do that. Good examples are do not give out discounts, do not give out pricing information. They need to talk to a sales rep for that. This is where you're going to come in and give all of those guardrails. Never give a visitor exact pricing information. Always when pricing is discussed, make sure to mention that pricing is highly dependent on each customer's unique— this is all customizable to your brand and how you want your AI SDR to interact with your leads. The next thing in here is emails. Everyone's going to have different styles of how they want their emails to be and how they want their AI SDR to send those emails. So you can come in here and give in natural language context to your AI SDR for this is what I want you to do for your email sends. And then the exact same thing for conversations. When she's having conversations on the website, what do you want her to do? When do you want her to offer up certain things? When do you want her to try to book meetings? This is all set here and it's in natural language. And when I say natural language, it's actually a really interesting thing we did. It is best if you come in here and you talk to it as a person. You would— are describing this like you would describe to a human. If you use another LLM to produce this content, well, it saves you time. It actually doesn't. Your AI does best off that natural human language context. So come in here, talk to it like you would any human SDR, type it out and say, this is what I want you to do for your emails. This is what you do for your conversations. Here's your guardrails. Now, the next part of this that's really important is these, we call these guides. There's accounts and there's leads. And I'm not going to dig into each one of them, but I'll show it here. But as a marketer, you have specific segments and it's how you want to interact with different businesses. Think enterprise accounts versus SMB accounts. Think geos versus different verticals. Usually that's all built in your data. But how do you then give that to your AI SDR? How do you let them know how you want them to talk? And I know from personal experience, when you're working with human SDRs, It's hard for them to context switch when someone's on the, the website, they're having a conversation, but they're from a specific segment. It's hard for them to know how to, to context switch between those different segments. Within guides is where you're gonna set that. So based on your data, you're gonna say any employee count greater than 5,000. Here's who these accounts are. Here's at a high level, like, here's who they are. Here's the messaging that I want you to share with them. Here's what I want you to focus on, and here's the tone and style. Because if you're talking to an enterprise account, it's gonna it's gonna be a vastly different message that we wanna share, and it might be a different tone and style that you want your AI SDR using. So anytime they recognize someone on the website, or anytime we use reverse IP and we think it's a known account, your AI SDR in real time can change the way that they're communicating. They can change what they're sharing and how they're interacting with that visitor. So the same is for accounts as it is for leads. And then last but not least are your goals. So you're gonna come in here and you're gonna set goals for your agent. Which a lot of times is book a meeting. That is the goal of your AI SDR is you wanna book meetings and you're gonna come in and set all that up. And then lastly, you have your scorecard. I don't have this set up, so this doesn't look the way that it should, but you're gonna come in here. This is where you're gonna give your AI SDR guidance. So at a high level, this is how you would set up your AI SDR. Now you would then go in there and you would do some testing, ask it questions. I tend to find there are about 10 to 15 questions that your SDRs get asked ask over and over and over again? Those are the questions you're going to want to go in and test with this AI-SDR agent and make sure it's answering it the way that you want them to. So do that work ahead of time to find out like, what are those questions? So I want to show you guys a little bit about what this looks like in action. So here I am on the Asana website. I have built my AI-SDR agent Piper. She is in pinned mode here on this right-hand side. Pinned mode means she is this always-on, ever-present agent on your website. If I was to minimize this screen, she's not covering up content, she's not hidden in a chatbot down at the bottom. She is always there and always ready to guide buyers. But I want to show you guys one of those slides that I uploaded. What does that look like? Hi, how can I help you? Hey Piper, I'm curious what a creative process workflow looks like at Asana. Do you have anything that you can share with me about a creative process workflow? I've pulled up a slide showing a creative project request workflow. It walks through each stage from request to final approval. What specific part are you most interested in? So a few things here. One, she's able to pull up those slides. She knows exactly when to use them, just like you want a human SDR to do. And two, I've onboarded her. She knows exactly how to talk about Asana, but Most importantly, she asked that question at the end. Like any good SDR should do, she is doing discovery. She's trying to figure out more about this buyer so she can give them a more relevant and personalized experience. I wanna show one more thing here though. Hey Piper, I'm from Disney, which is a large enterprise media organization. Do you have any customer case studies that you could share with me from large media organizations? Here's a case study featuring Gannett, a large media organization. They scaled their campaign production using Asana. How does that compare to what Disney's looking for?

Awesome.

So again, she has access to everything on my website. When I ask her a question about a case study, she's not just giving me a hyperlink, she is actually popping it up. She's making it more visual, and she knows that website content that she can share. So I could go on and on and test this. If you If you guys want to see more in action, we actually have a booth over in the other hangar. Come over, we'll show you like a full-blown demo. She can talk in a bunch of different languages, she can share content. It's very cool, but I am limited on time, so I want to dig into a few more things with you all. So a couple different use cases. If you're looking at this, you're like, okay, cool, I have a need for this. I have gaps on my SDR team, but how do I use this? What are my, my different use cases? The first one is high-intent buyer, no form fill. You're going to get a lot of buyers coming to your site. They're going to go to all the pages you want them to go to, and they're not going to fill out a form because people hate forms. If that person is cookied in your system, if they've come and had a conversation with you at some other point and they have hit a high-intent page, or heck, if they even had a conversation with your AI SDR and they did not book a meeting, you can set up follow-up campaigns, which I'm going to show you here in a minute, to follow up with that buyer and try to push that meeting booking. So you don't have to have them fill out a form fill out a form in order to follow up with them. You can use these high-intent activities, whether it's page visits or conversations or campaigns in your Salesforce instance to trigger a follow-up from your AI SDR. The second one is maximizing ad spend. If you are a startup in this room, if you're a mid-tier company, whatever you are, you're running ads. That's typically the first place marketers go to try to drive more traffic to their website and drive brand awareness. Ads tend to have pretty poor conversion rates. On your website. And as a marketer, I spend a lot of time trying to prove my cost per ad click. That is all I do. And I'm trying to prove that I need to spend more money and that they're actually fruitful. Well, if you send someone to your website from an ad click and the only thing they can do is convert on a high-intent form that says book a demo, you're going to have low conversions. Piper, your AI SEO agent, she can offer up more content. So if I click an ad, I come to the website, she can offer up content versus a meeting booker and make sure we're capturing that information. Everything is done in context from that buyer where they're at in their journey, and it's going to be personalized and bespoke to them. So they're not going to push a meeting booker if that person is not showing the intent for that. So she can offer up content instead, and you're going to capture more leads from your ads and convert those into pipeline later down the funnel and make sure you're proving value from that ad spend. And the last one I'm going to use as my example is pre and post-event outreach. We're obviously at an event. This is the best use case for an AI SDR agent for follow-up. We actually use this as our first example when we launched our email proponent of our AI SDR agent, and we saw a 70% increase in pipeline from events within one quarter. And it's because when you come to events like this, you get a ton of leads. Some of them are high intent, some of them are low intent, some are ICP, some are not ICP. SDRs do not like to work these leads because they don't convert at a high rate. And I've worked with SDR teams for a very long time. And I've begged them to work these leads because I've spent hundreds of thousands of dollars to get them and they don't want to work them. If you have an AI as your agent, they don't care. They will work those at scale infinitely and bespoke to that person based on the context that she has about that individual. So when you're working all those leads, you find the diamonds in the rough, you find the people that were going to convert that your humans weren't able to work because they were at capacity and didn't have the time to work those leads. So I kind of want to— I want to show you what that's going to look like when you go to set up one of these post-event email campaigns. So back behind Qualified here, we call this behind the gear. I'm gonna go into my email section of our AI-SCR agent setup. And again, I'm Asana. Asana has a huge conference every fall. It's called the Work Innovation Summit. Now let's pretend that post-event they have all these leads, they wanna work these leads. What are they gonna do? They set this up and you can trigger your campaigns on any number of of things I mentioned. It can be things like they had a conversation, but they didn't book a meeting. It can be things like pages viewed, or it can be things like campaign statuses. So here I've set it up based on a Salesforce campaign. They were at a summit in London, they were attended. That's going to trigger this campaign. And then you can start to set things like your audience filter. Do you want it to send it to everyone? Probably not. Do you want to filter down by titles and people in persona? Probably. You could come in here and set only to specific people. It's all based on your data in your CRM system. Then you have things like exit criteria, goals achieved, all of these things. So you're giving that AI SDR content or context of who to send this to and when. Now, there are things like sender replies. Who do you want this to come from? Do you want it to be from your AI SDR agent? Do you want it to be from Piper, or do you want it to be from a human sales rep? Your AI SDR agent can send it on behalf of an AE or on behalf of an SDR. They can do that. They can send it from a specific user. That's all very customizable. How do you want it to handle replies? Is there a two-way conversation? If someone replies to that email, do you want your, your agent responding or do you want it to hand it off to a human? You can set up either way here. And then delivery. How frequently? When do you want drop off? You can set like how persistent your agent is. Again, it's all very customizable to your company. Now I want to dig into the campaign guide here because this is where you are giving your AI, your Agent Piper, all of the context about the event. So in this case, you would come in and type one of these up. The great thing is these are— you can duplicate these. So once you set up one and you get it done really well, you can duplicate these campaigns and you can just swap out content in them based on the event or based on the campaign. But here I'd come in and say like, hey, we had this summit, here's what happened, here's the things we focused on, here's why it's important, here's things you must do and must not do in this email campaign. Tone of voice, messaging focus. I'll be honest, this is probably overkill because I've already set that in that guides and context and she'll always use that. But I put it in there for fun anyways because I already built this a long time ago, honestly, before we even had those. So you can come in here and tell Piper exactly how you want her to treat this campaign as opposed to building a really rigid rules-based email builder. And it's going to be personalized, it's going to be bespoke, and it's going to be triggered off actions of that buyer. So once I set all of this up and I give Piper all that context, I can come over here into preview and you can choose different people within that campaign and see what is this going to look like. So you can see here one of these contacts, Piper, would say, hi there. One of the clearest themes from the Work Innovation Summit was this. She's going to give that bespoke email. Very cool. Before you launch campaigns, you can see based on different personas, like, what is Piper going to say? It's never going to be canned, though. It's never going to be just this email. This is giving you an idea, but she's going to personalize these. And change these based on that buyer. Now, you can also come down and say, what would happen if the account started surging? What would happen if they replied? What would happen if they didn't reply? What would happen if they filled out a form? And Piper's gonna go through all of this data and she's gonna look and give you another example of this is the type of email that I would send. So you can kind of come through here and you can see like different experiences for your buyer based on what is happening. And then as always, like I showed in the very beginning, there is an analytics on the other end that you can see and you can dig into. Sorry, this is very slow to load. Of course, always with conferences. Give her a second, but we're short on time. Okay, well, normally she should pull up an email here and we would see Piper's thinking and she would show me an email. And then once that is done, you're gonna hit save, you're gonna hit send, and she's gonna start emailing those prospects. You would come back into analytics here. Here, and you would see how it's performing. So you're gonna see if people are responding, are they opening? You have all this analytics at your fingertips, and she is going to work every lead from your event. So at a high level, that's the demo. Really quick results. If you have been wandering around the campsite here and you've seen Amelia AI on any of these screens, that's actually powered by Piper, the AI SDR agent. So this is an example from SaaS. They booked 130 meetings, $2.5 million in pipeline generated and $1 million in closed one revenue. Amelia set it up. We actually did a custom avatar for Amelia, the chief AI officer at Saastr, and she— we actually have a podcast with her where she talks about it, but she was like, it's incredible. I'm never missing leads. There's all these conversations happening. I'm actually generating— when sponsors come to the website, we never miss these opportunities and we can follow up with them at scale. There's a ton of other examples of this. If you go to Qualified's website, we have a customers page with a ton of great examples. But I wanted to highlight SaaSr in case you guys see Amelia AI throughout the event. So I've got a minute and a half left. If there are any questions, I want to make sure I leave time for those questions. And whilst we wait, I'll remind again, we do have a booth over in the other hangar. If you guys, like I said, we can give you a more detailed demo, show you all the power of what Piper can do on the website. Yeah.

Do you have partners who help implement this, or is this up to the company?

Super good question. He asked, are there partners helping implement this? We don't work with third-party providers technically, but we have a very robust in-house customer success team. We call them QSAs, or Qualified Success Architects. They're super knowledgeable in CRM and marketing automation and in marketing strategy. So they help all of our customers set these up and they have a really rigorous process and they go through what we call crawl, walk, run. Where they're like, hey, based on your business, you can't just throw the whole kit and caboodle at one time based on your needs. Like, here's where we would start with a crawl. Here's your walk, here's your run, and really help like move people into what's going to be most effective. They're there every step of the way with our customers. They make sure that Piper or their AI agent is working. All of our customers, they personify their AI agent. So like we have Bella at Box and they're all different. But we don't just leave you to do it on your own. It's going to be like hand-holding from our our customer success team. And if you ever go to G2 and you look at reviews of ours, every single one mentions our customer success team because they're that involved. Yeah, I think they're working on it. Sorry. Do you want to try this mic here?

Yeah.

Okay.

Yeah.

All right.

Went from nothing to a lot.

Yeah, really loud.

Sorry. Okay, I have a question. So we're actually using Qualified.

Amazing.

But we're kind of like immature in the setup.

Yep.

So one thing I would like to know, and I haven't looked into this, so sorry for the ignorant question.

Yeah, totally fine.

But do you have built-in evals in the system, for example, to like help us refine the prompts and content that we have for different things? Like email campaigns and chatbot feedback, or is that something that we would build separately and then do a bunch of backtesting against with Qualified and then refine our stuff that way?

So two different things. And a lot of the natural language, there's like prompts and like it gives you almost like suggestions within to help you like make decisions. But also we have a whole university page where they have a bunch of different examples and things to get you started with those. So you can always use that to like get started. But again, my biggest recommendation, ask your CSM like they want to help you and they want to help you get started. And if you come to them with that question, you're like, hey, like, we're not using the most. Can you help us? But between the suggestions, like within that natural language box, it'll usually give you like some like grayed out text of suggestion and our university. You should be able to get started there.

But do you have like a way, I mean, to like specify questions and then see the responses over time?

Oh my gosh. Yeah, sorry, I misunderstood.

Yes.

I want to create like a grading rubric.

Rubric for—

Yeah, yeah, yeah. You don't need to create anything. So it's actually called our AI scorecard in that backend, which if you come by the booth, I'll show you like a full example of it. But you can go in there and ask questions and like do essentially like back and forth with Piper without being on your website. So you come in, you'd ask questions in the scorecard, you give her like a thumbs up, thumbs down, you give her like, this was good, this was bad, here's how I would change it. So all of that, like grading rubric is like in that AI scorecard.

Okay, understood.

And you can do— obviously if you want to build something yourself, you can, but that is built into the backend in the AI studio for you to ask those questions and grade her and then give her that feedback.

Okay, cool.

And then I have one other question.

Sorry.

Is there also a way to use like LinkedIn messages as a medium of communication? Like we get a lot of like inbounds from there, you know?

Yeah, you have email, you have the website chatbot thing, which is great, but like it'd be cool if we could like delegate or hand There is not because LinkedIn has like an ironclad lockdown on their API and to stay in compliance, we just don't touch it.

Got it.

Yeah.

Okay. All right. Thank you.

Awesome. If there are any other questions, I am over time. She's waving at me back there. Our booth is over there. Please come chat with me. I will be there for the rest of the day. Thank you all so much. Joining us now from RSL AI, the founding account executive, Carol Lee Robards, and the CEO and co-founder, Jason Eubanks. How's that going?

Mic check. We good? All right. Well, I have 3 surprises for you today. One, if you stick around to the end of the session, the first 5 people that ask a question, we're going to throw swag out to. Two, the rest of you don't ask questions and you still stick around, you can go back to our friends standing up at the back of the room back there with Oricel shirts on, scan a badge, and still walk away from this session with swag. The third thing we're gonna do today is we're gonna show you live product. So, I figure you've seen a lot of presentations, a lot of chalk talks. We're gonna wrap up the last half of this with a live product demo. Sound good? All right, let's get going. So, you can see in the slide here a little bit about our background. I think the most important thing that you can take away as for context around my background is that I've been a CRO and a go-to-market leader of sales marketing, CS, at 4 other startups prior to starting this one. And the way I like to introduce that is to get the clicker to work. Maybe you'll take over. The way I like to introduce that is to say, if you're sitting in this audience and you're a go-to-market operator, I am you. And when I was building and running sales, marketing, and CS at the last company that I was at, which was Harness before starting Oricel, this was my go-to-market stack. It may look like yours. It may look like one that you've started the journey on building. Well, I'll tell you a few things about this go-to-market stack. There were 22 products in it. We were spending over $3 million a year in SaaS fees. It had an incredible, incredible amount of complexity. 22 tools. I had 11 ops team members. Running this stack, stitching together the integrations, breaking the fragile workflow layers that were all stepping on top of each other, augmenting the integrations between multiple databases and data stores, and finally, trying to chase that ever problem, that ever big problem that we have when we put together a stack like this, which is to get a single view of the customer journey. So, If we think about some of the complexities with the legacy approach to building go-to-market tools, you can think about it like this. And this was our reality every day. And maybe it's yours today, but it doesn't have to be. And we're going to show you how you can get out of this and into an AI-native go-to-market platform. And like Kara Lee, as you're a new rep, close $3 million in your first 41 days. So let's talk about some of the drawbacks of this current state. Well, it's complex. It has multiple data stores. You buy niche products for every part of the sales process. Eventually, you end up with legacy tool sprawl and all these legacy vendors that are now trying to push agents at you, multiple agents. And that's gonna drive up your costs further. It's gonna drive up your complexity as you now you start to introduce agentic thrash. Why would I say that you start to introduce agentic thrash? Is because as you add every one of these niche products, they bring a siloed database in. And that siloed database might integrate with your existing CRM at a data level, at a field level, and that's fine. But stuck in that siloed database is the context of all the conversations, all the activities, That context in the era of AI is what you need in order to drive high-quality intelligent automation. So if you have agents that are operating with a fraction of the metadata, lacking context, and don't see the entire contextual layer across your knowledge graphs, you end up with the potential for low-quality automation, or, or worse yet, agents that fight autonomously amongst themselves instead up over each other. Go to the next one. So what are some of the negative side effects of a legacy architecture like this for your sellers, for your ops teams, and overall for the productivity of your business? Well, it's that your sellers aren't selling. So I would challenge you, if you're sitting here today and you don't know exactly what the percentage of time is that each one of your sellers spends in front of prospects and customers face to face or over Zoom, then you need to go measure that KPI among everything else that you consider to be the top KPIs of your business. I went through this exercise at Harness. It was middle of COVID We were being asked by our board to do more with less and go from a really high growth, high burn profile to accepting half of that growth, but driving more efficiency. I walked out of that boardroom and went on an internal journey, the project that we called X-ray. We went to every major department of the business. We drilled down, logged every single activity every single person was doing, what tools they were dependent upon, where were there overlap in those tools, did we need to buy a different set of tools or build a different set of tools to drive more automation to make them more efficient? And one of the key findings that came out of that exercise was a graph that looked like this one, a chart that looked like this one, when I realized that my reps were spending time in 10 to 12 products a day to do their job. They were losing that time to contact switching and manual activities, and they were also spending an enormous amount of their what should be productive time on just precursor activities to actually driving customer interactions. Manual account research, prep, follow-up, even prep for internal meetings. Deal reviews, QBRs, et cetera. So this is probably what your team looks like, and you may be sitting there wondering, "Ah, I don't think so. You're wrong. You ran a really inefficient org, Jason." You might be right, but I will tell you that the industry benchmark says that the average B2B seller spends 24 to 30% of their time in front of customers and prospects. It's impossible to maximize productivity for your business if that's the world you're living in?

Let's go on.

So what we did is I got together with the SVP of product engineering of that company, that Harness, that I was at at the time, and the two of us went off and we started Oricel. We started Oricel to solve this problem that we're talking about, to deliver maximum productivity of the best, of the of the business to maximize the productivity of every go-to-market employee in your business with an Agentic platform built in an AI-native air, providing you a single intelligent automation platform from contact to contract across the entire sales process for you and your customer's journey. Things that are required to deliver a great platform like this in our opinion, is you need to have a unified data architecture. That means structured and unstructured data alike sitting in a unified data platform architecture. We deliver 900 million contacts, 85 million accounts. We auto-enrich all of those, and we give you the ability to extend it with custom enrichment. You need to have a conversational context layer that provides rich context across all conversations. All channels, all signals, all the time. And you need to leverage that with an automation layer, some of which we build for you and deliver in, in the way of autonomous agents, others that we extend to you to build custom agents with natural language. When you move to an AI-native go-to-market platform like Oricel, you can do it in a single platform approach where we auto-migrate you off your existing tool landscape. We are your AI-native CRM in addition to your entire Agentik operating system platform. Or you can choose to allow us to meet you where you are with that AI-native go-to-market Agentik platform right on top of your existing CRM like Salesforce or HubSpot with a bidirectional integration, allowing you to mature into go-to-market with AI-native workloads at your pace of where you are. And the pace of where you want to go. Alright, now I'm gonna hand it over to Kara Lee so you can take a look at a real-life journey for Kara Lee's first 41 days as a seller at OraCell.

Hey y'all, hope you can hear me okay. Seems like mic's good. So essentially the way that I was able to find and close a $2.7 million deal deal in my first 41 days at Oracle, other than a killer pair of purple stilettos, was that unified contextualized data layer. So sitting between all conversations, all channels, all signals, that intelligent automation enabled me to have those superhuman operators start to take action on my behalf. And that's where Oracle really becomes becomes a system of action. I log in and the territory agent has built an agent— has built a territory for me. My SDR agent is ready to rock, getting outbound off of those high ICP accounts moving. My sales agent is coaching me through the opportunity, giving me insights on MedPick, helping me to quote intelligently. And then finally, the operations agent allows me to build custom agents to automate different things across my business. So if I dive into a demo and show you what this looked like when I logged in on day one, here's where you're seeing those accounts already prioritized by ICP. Oricel built this territory for me. There were no spreadsheets, no other tools used, And right away, in addition to prioritization, I'm able to do custom research at scale by using these AI agents. These AI columns allow me to look at things like who are the new CROs across my territory and have they actually hired a new CRO in the past year? These are both indicators for us that this would be a great account to prioritize. Oricel's agents look at structured and unstructured data and intelligently understand what are the reputable sources where I'm gonna find that, whereas all I did was prompt, give me the details about the newly hired CRO and their background, and it already did this research for for me. I can also pull all of this research across anything else in the account, including the ability to prompt into Ask Oricel and give me lead lists, prioritize accordingly. And here you're seeing examples of where I can start to filter some of this data to say this is the number of go-to-market employee counts, so on and so forth. And then when I'm ready to dive in to reach into the contacts that, that intelligently pulls into the CRM, the contacts are automatically pulled from Oracle's 870 million database records and prioritized by propensity to engage. Remember, we're looking at external signal as well. How likely are these contacts to be receptive to my message? What buyer persona are they? And give me the background off of where they've been before, who they're connected to, and the full timeline of the engagement with that person or that account, and then auto-enrich their LinkedIn, their email, their phone number so that I can auto-dial straight from Oricel. And then now that I have an account list that I'm ready to reach out to, I'm able to create sequences that, for starters, I can prompt build all of my sequences in Oricel so that it's quick and scalable for my entire team. And then when I actually go to reach out to those contacts and put them into a sequence, even though this is for a mass outreach of many contacts, every single account or persona is going to get a unique message off of that custom research by simply prompting to pull in from other custom, um, uh, sites that I've pulled into helping me to determine that that's a good account fit. I can also set up cold calls with a pre-built script so that whether it's me or my SDR that's reaching out to that contact, they're going to get something custom to them. I can also action LinkedIn straight from Oricel because of our Sales Navigator integration and our LinkedIn integration that allows LinkedIn connections and messages to show up across that contextual data lake. When I actually go to action these cold calls, I'm able to see this as a task that surfaces so that I can do a cold call block, or like I said, my SDR or my SDR agent can do a cold call block. And yet I have key contacts from the account. Have they been to a recent marketing event like SaaStr? What discovery questions should I ask? What are the objections I might get? Maybe a report core notice and then dial them straight from Auracell with an either Auracell-generated phone number or your own personal number that's uploaded into Auracell. And once I have an opportunity, thanks to Auracell giving me this great outbound motion, when I actually go to manage that opportunity, I'm also going to get help closing the deal. What you see here is called an AI block, and this allows me to surface insights from all of the context across this opportunity and start to de-risk with my leadership, which is something that Jason M. Lemkin and I spent a lot of time doing on that deal. Who are the key players in the engagements? What are the next steps? And critically, how well am I executing? Am I executing against our sales playbook, our frameworks? Do I have a champion? We give you both a completion and a quality score and allow you to get coaching on each of those sales methodology frameworks in line with me trying to close this opportunity. So that if I ask a question like, what's your inbound lead meeting conversion rate? What's your conversion rate? And they tell me, well, right now it's 15, but we have a target value of 60%. Oricel is saying, that's a great question. You now need to go and tie that metric to revenue to make your business case stronger in helping you to actually close this opportunity. I can also then generate a quote, and all of this is pulling from the timeline, allowing me to see see every activity across the account, whether it be SDR team, marketing team, text messages, which I can easily drop into the timeline and add to that context graph, is now feeding every single touchpoint. And this is all separate from the very many Carely agents that are running in the background because we also ship with with an agent builder that allows you to build custom agents that take action on your behalf. And this is a great example of that, that's routing leads directly to whatever I want to happen, whether that be an automated outbound sequence or into my inbox on my behalf.

All right. We didn't get to show you AI forecasting, the voice dialer, or analytics. There's so many things. That's why we'd love to invite you, if you go back to the slides please, that's why we'd love to invite you back to our booth where you can see a full product demo of everything that Kara Lee was just talking about. And maybe more exciting for you than the cool demo of our, of Oricel's platform is the opportunity for you to register. There we go, one more, there we go. If you swing by, you'll get swag today, let's scan you for that in the back of the room. If you swing by the booth tomorrow, we're giving away $15,000 in cash cash prizes, but you have to be present to win. Thank you for your time and attention today. We appreciate all of you. We're out of time.

We no questions?

Please give a warm welcome to the co-CEO and co-founder of Relevance AI, Daniel Vasilev.

Hello everyone, I'm Daniel. I'm co-founder and co-CEO of Relevance, and I'm very excited to be here because in this room I'm assuming a lot of you are already at the frontier of AI and trying to find ways to go even further. Some of you have probably deployed Copilots in your business or are leveraging them yourself, but there might be a nagging question going on in the back of your mind. Why Why do we have lots of copilots, but why don't we have a lot of autonomous agents yet? Why aren't we there? And so today I want to cover a little bit about the journeys we've seen of customers going from copilots to coworkers, how they've achieved it. We've done over 100 enterprise implementations, and I want to share a few tips on what we've seen work well. So first, a little bit about Relevance. We are a Series B startup. Based out of San Francisco and Sydney, Australia, and we've raised over $42 million to date. We work with some of the best enterprise go-to-market teams in the business. And the way I like to position us to a lot of folks is, yes, we're an enterprise agentic platform. Yes, we help automate a lot of work, but a helpful distinction is to think about us not like the point solutions you buy to outsource, but to think of us like hiring. The process of defining the role, setting the scope, giving access to your tools, managing its performance. Building successful autonomous agents is a lot more like building successful teams than it is like deploying software. And let's try this. And effectively where we want to get everyone is to this promised land, this promised land where your teams are able to build and deploy autonomous agents based on what your top performers are doing today. And we can say this confidently because we've seen this work extremely effectively. These are just some of the examples of customers we've worked with where they have seen enormous amounts of impact in terms of FTE equivalents with agents that do things the exact way they do them. Agents that follow the recipes and the processes of their top performers, not agents that just pick up on some basic process or some basic amount of configs that help them deliver some part of the journey. No, we are focused on helping them deliver entire swaths of their journey. And in our product, we really help make this possible from start to end. We make it as easy as we possibly can to help your team actually achieve an entire AI workforce that helps impact specific metrics and outcomes of your business. But there's a little bit of a mismatch that we're seeing widely in the industry. A lot of teams are buying a copilot, but they're actually expecting a coworker. A lot of boards, a lot of CEOs, they're asking their teams, do work autonomously, do more with less. Let's use AI to power parts of the business we could never do before. But the approach has largely been, let's give them a copilot that helps improve their individual productivity. Now, that's very impactful. That's very helpful, but it fundamentally doesn't address the expectation that perhaps your team has. And so that mismatch actually isn't a fault of AI, it's typically down to the approach. And so we found it very helpful to not paint AI with a single brush stroke, but to actually break it down into 4 levels of autonomy. 4 levels that are actually very useful depending on the different use case you have. Level 1 is what we call assist. This is what a lot of you are doing today, specific discrete tasks that you can delegate to AI and it can solve it for you. Level 2 is Copilot. Now we're combining playbooks and routines, sometimes called skills, that can do more than one little task. It can actually do multiple combined for us. Level 3 is where it gets very exciting. Now we're starting to get to a place where we have Autopilot agents running autonomously for us, triggered on events and signals, doing work without without a human telling it to do it every single time. And then finally, we approach Level 4, self-driving. Think of self-evolving agents that can be targeted to improve a specific metric in your business and every single day trying variants to achieve that, pruning the best wins, the best ones, or the worst ones, selecting the ones that can give you the best performance autonomously for you every single day. So what does it look like to deploy L3? What does it actually look like to be in a place that has coworkers? Well, you need 3 things to be successful. Number 1, you actually need a documented process. L3 is not for experimentation, at least not today. You want to give it something that's a clear, repeatable workflow that you're able to teach people. If humans can't do it, you probably can't teach an AI to do it today. It's much easier to give it processes at work. Number 2, you need a clear owner. Someone needs to be accountable for outcomes. Agents are not accountable to anybody. You need someone who's gonna manage this like any other team. And number 3, start with a narrow scope. Do one process with a single trigger with a single outcome. If you wanna see success, you wanna prove it very, very early so the team and the business has buy-in into achieving your goals. And once you have those 3 things, you have to think about what makes for a good use case. This is just as important as actually delivering on the use case. You want use cases that can prove to your business fast ROI. So what are those? It's typically the ones with high volume. They're very repeatable. It's ones that have very clear inputs and defined outputs. They need to already be documented, or you have to be able to translate that knowledge into some documentation. And it sounds obvious, but it needs to be measurable. It needs to actually have an impactful outcome because someone should care about you achieving this. It shouldn't be some side quest. It should be a top 3 priority of the business. And I know I've spent a lot of time so far talking about process, but the fundamental thing that a lot of folks miss when they try to build coworkers is they think it's all about technology. And we're a vendor, we build technology, we build a platform, we make this easier than anybody else, but without a process, you're not going to succeed. The only reason we've managed to deliver so much impact to the organizations we work with is because they've had fundamentally fantastic process in delivering autonomous agents. Let's talk about the technology required. So number one, your agents need tools. If we're hiring coworkers, they need to be as flexible as humans. If you have a solution that's limited by the tools it can use, well, it's not going to operate like your top performers. It needs to be completely vendor and model agnostic and needs to be able to plug into all of the solutions that your team leverages today. And it can't have limitations. It can't be like, oh, sorry, it doesn't write here, or it doesn't actually read this field. It has to be able to leverage everything if you want it to be able to operate like a human. Number 2, it needs excellent instructions. So your job description is really critical. It's more critical almost than hiring someone. You want to be able to document the SOP that's going to deliver results for you. You want to be able to turn this autonomous agent into a predictable engine of revenue for using it for go-to-market use cases. And the only way to do that is if it's following a process you know works and is actually going to deliver results. It's very nice to make a functional agent, but who cares if at the end of that it's doing things that don't impact anything, right? You're going to fire the agent. Focus on delivering. Instructions and processes that deliver value to your business. And then finally, governance. Here we have a workforce of agents that work together. This initial agent that does the demo qualification doesn't know how to handle certain prospects. So we've given it a friend. We've given it a sales strategist, a sales strategist that it can now work with to identify how to best sell to a prospect that it might not have clarity on. And once you have a workforce like this running for you, once you're in a place where you're delivering millions of tasks per year, hundreds of FTE equivalents of output, what gets really important is governance. And now governance is obviously important for security. It's important for IT to give you permission to access things. It's important for you to monitor your impact. But the reason why it's actually really critical, and if you don't have great governance, it's going to be a massive problem for you in 12 months' time from now, is it's how you get to level 4. If you want your agents to be able to self-evolve and self-improve, they need to be able to learn on the results that they've done. And if you don't have excellent governance, you don't have the data to be able to identify what works. And what I really implore all of you as you're thinking about your AI strategies over the next 12 to 18 months is think about those levels of autonomy. Think about where you sit on the curve of autonomy today, what use cases you're tackling, and think about and ask yourself and your teams whether you have an L3, L4 strategy, a true L3, L4 strategy. Do you have a pathway in the next year or two to have a team of agents running autonomously for you, delivering outcomes that previously only teams of people could do? Because that's where the real leverage comes for a lot of customers we're seeing. And the good thing is it's very early days. A lot of folks are still not there yet. Most are at L1. Some at L2, very few at L3, and there's tons of leverage there. So the message I want to leave all of you with is that teams that have AI coworkers really compound because they're not bottlenecked by the capacity and those same constraints they were in the past. And if you need assistance on going on this journey, come find us. We've got a booth in the other hall, Relevance AI. We really care about process and the systems required to make you successful in addition to delivering you the best possible platform for those results. We're one of the very few who can confidently say and demonstrate with amazing enterprise logos that we have agents running millions of tasks a year fully autonomously, flattening headcount curves, changing priorities in the business, and we'd love to show you how we're doing that and help you achieve those same results. So thank you very much. I'm Daniel from Elements AI and hope we can see you soon.