AI Insights - What's Working And What's Missing

So there are some things happening right now that I think are really interesting for all of us who have kind of, you know, love digital marketing, helped grow companies. With digital marketing, I think we're entering a real phase of disruption. And there's a couple of reasons for that. The first one is we just see this platform disruption, like raise your hands if people have pivoted their search engine use use case to AI vs Google. Anyone followed me? A couple. Not as many as I had thought. Maybe I lived too much in the AI bubble, but there's a lot of platform disruption. And I think one of the interesting things to actually think about is we have, you know, digital marketing has been built on the open web, right? We were able to create things of value, they rank in Google and you acquire traffic to your website. Actually, 64% of all traffic on the entire web comes from Google. Like Google are the kind of open web and that is being disrupted right now. AI is really a better use case for search because it actually does the search for you, provides you with an answer that's clear and concise. Anyone who has used perplexity ChatGPT, who's just launched their own search engine, all of these experiences are actually better than having to go through all of these blue links yourself. And we're starting to see people's use, starting to see people's usage pivot towards AI. And so we see this kind of disruption happening. And what's actually happening is Google are reacting to that. Google do not want to be left behind. And people have probably seen things like AI overviews get rolled out recently, they've just been rolled out to 100 more countries. So you're probably going to start to experience them a lot more. AI overviews are basically the AI modal in Google that gives you the answer. So you never even have to go into the website to get it yourself. And that is to me pretty interesting because it actually starts to move much more towards this closed web and what we've seen happening in social media. So in social media used to be a great way to promote content, used to be a great way to promote the things we wanted people to buy and do. And over the course of the last five years, social media has become a lot more of a closed web experience. And when that happens, you're not really able to promote your own stuff. And if you do try to do it, you kind of get penalized. And we see these with some of the results. You know, when you actually put a link in LinkedIn in your post and I do a lot of LinkedIn content, you actually get six times less impressions. We see the same thing happen on Twitter, we see the same thing happen on Instagram. That's why you see people on LinkedIn kind of go, hey, like DM me for the link, or I'll put the link in the comments. Because you get way less impressions. That is what's happening on social. And we're starting to see the same thing happen on search. The other thing I think a lot about is when I came into digital marketing, there was all of these new channels opening up. I came in through search. Google was still growing, search was really growing, paid marketing was growing, social media channels were coming online. We had all of these different ways to acquire customers. And what's happened over the last five years for the most part is that has really stagnated. In actual fact, it's probably started to decline because more of those platforms have moved to a kind of closed web experience. And so we see there's more B2B startups, right? Over the last five years, 30% more B2B startups. But over the last five years, we have not really got access to more marketing channels to market those startups. We say way more ad spend. That's why you start to see a lot more saturation in paid spend. Because there's just way more competitors, there's better tooling and the ability to actually optimize your ads to get any kind of leverage over anyone else has kind of gone down straight. You're really only competing on budget. And then we just have way more vendors, right? There's way more competition in any market you're trying to sell your product or service, there's just a lot more competition. So more competition, the same digital channels. And the last one is one of the ones that I have been fascinated by recently, which is AI's ability to take someone from something that they are finite, just making them really, really great. And in a very short amount of time. This kind of notion of time to mastery for me in digital marketing used to get a lot of leverage because you could have deep expertise in something, you could have leverage in something. You could be better at a bunch of people for a long period of time. And that actually was the way you were able to grow within these channels. What's really fascinating for me about AI is it actually helps people to accelerate their learning and to become very, very good in a short amount of time. And I'm going to give you some examples of that, which again increases the amount of competition you have in all of these channels. So to me, I think we are entering a period where the things we used to do will not work in the same way and we have to find a new playbook. I'll give you some of the ideas that we are thinking about internally about how this actually changes. Now, the first is really how we have traditionally published a lot of informational content. We've got that content to rank in search. We've managed to grow a lot of businesses through that means. And I think what we're going to have to do is figure out how we continue to do that. I don't think that is going to disappear, but actually do it in a way where we can AI proof it, we can make it much, much better, and we can actually still compete within the kind of verticals we're competing in. And so it's how do you really use these large language models to get 10x the value in the places where you can still differentiate? The very first thing we did was actually start to classify all of our content into these categories to try to figure out what is our risk because of AI, we categorize them into high risk. The high risk category, if you are dependent upon search, if you get a lot of search traffic, that's really all of the content that you can rank for. That is really just an informational query. And a large language model can answer it better than you're able to answer through a blog post or something like that. It's just very straightforward for a large language model to be able to answer that question. So that's your high risk. We had a bunch of things in high risk and we kind of marked them to say, okay, well bear case. If people do start to gravitate towards these AI assistance to get all of their search content that is kind of our high risk bucket, then we have this moderate risk, right? These are more complex questions, right? You have to have deep thought in how you answer these. Or much more research content. They're much more complex content. And I'm going to give you an example of how I think you can differentiate in that yellow bucket. But that yellow bucket, I think is still up for play, right? How can you answer something, provide something unique in a way where people cannot get through just an AI overview through just a ChatGPT query, something like that. And then the AI proof traffic, AI is not causing any kind of cannibalization for transactional keywords. So transactional keywords are you search for a product, you search for a service, you search to do an action that is still working really well. That's why freemium services and things like that are still performing really, really well. And from my network, in terms of speaking to people who have large content engines, get a ton of traffic from Google, we're seeing anywhere. I've seen anywhere between 10 to 20% reduction in traffic because of AI cannibalization. So it is a real thing. It is happening. And this is pre actually ChatGPT launching their search engine. They have 3.1 billion visitors in September. So they have a large distribution engine to get people to start using AI search. Obviously, people are, you know, still primarily using Google when people put up their hands here. But I do think that is gonna start to accelerate. I think AI assistants are just a better experience for a lot of search. Anytime something is easier to use, anytime something is faster to use, users you will usually gravitate towards that. And you see, that's the reason every single SaaS homepage is faster or easier, because that's actually what users want. And so one of the ways that we are doing this is we are going to start to integrate AI into every part of the content process and then try to figure out how we can get 10x on the places where we want to differentiate. And so traditionally, you had a content team that did all this work, right? They did a bunch of the research, they wrote the first drafts, they did a bunch of editing, and then they did performance analysis. And I want to talk a little about performance, a bit about performance analysis. And what we are working on is the ability for AI to do all of the research. One of the interesting things about where some of these tools are going, like Claude and OpenAI's enterprise, is they can actually oauth into your tools and start using the tools for you. That is where their B2B products are going. So you're going to be able to have AI OAuth into your tools, do a bunch of the content research for you. It can actually then build a first draft using style guides that you give it. I'm going to give you an example of how to build a style guide. The third for me is where you can really differentiate. It's in the editing, it's in the human nuance, it's in the competitive advantage that you add to that content. And then the performance analysis is really interesting, Right? The performance analysis to me is one of the more interesting parts about how AI is going to actually optimize a lot of the things we do. I was talking to a really incredible founder who was raising a Series A recently he has this paid tool. It's an AI driven paid tool and it's so cool, right? What it does is actually it can run hundreds and hundreds of tests, right? Prior to this tool you had to do all of these creative variations yourself, image variations yourself, all of the different video variations yourself. What that tool can do is basically do an infinite amount of creative variations because it builds the images, it builds the videos, it, it has a template to work for and then it can learn really rapidly what are the creative versions that actually work and then just use the same template to create more versions of that paid tool. And that is how you get this self referring loop from AI to make all of your stuff better. And that is coming online very, very soon as well. But I want to talk a little bit about the good from great, because when you look at this, you're like, oh God, like what am I even going to do? How am I even going to compete here? Is everything going to just look and feel the same? Because AI is going to do it all. And I don't think that's the takeaway from this. It's that if you can have AI optimize and accelerate a bunch of the steps, you can spend more time on how you can differentiate, differentiate your content from everyone else. One of my favorite examples is from Lenny. I'm sure people know Lenny's newsletters. What's really interesting about what Lenny actually did is he had like this one thing that was unique to him that worked really, really well. Lenny used to be a growth PM in Airbnb. He has an incredible network of people he can reach out to. And so what he really did was anytime he had a post, he would reach out to his network to get examples, example based content. And he would say, hey, can you tell me what your churn rates are? Can you tell me what your conversion rate to leads are? Can you tell me what this is, what that is? And so when he had a post, he had unique data points, he had unique examples, he had unique case studies that, that no one else had access to. And that's why his results accelerated so fast. And I think that example based content is a really interesting one for a lot of companies to think through, right? If you had a full time journalist to actually go and build a database of examples of case studies of things that are unique to you that AI cannot get because you have not, they don't have access to those things, that is how you can start to differentiate your content from everyone else. And so I think you have to be really thoughtful, right? I don't think we can live in a world anymore where we look at the keyword and we say, okay, let's just create the educational post to answer this keyword, the informational content to answer someone's query and think that that is going to be enough to sustain growth in the long term. It is not. You have to really start to spend the time to figure out, well, what is our content strategy? How do we differentiate? How do I make the stuff that I do 10x better than any of the competitors? And I think that is a very, very cool example. I actually know another person who is going all in on this Strategy, has hired three journalists for their B2B content marketing strategy just to do example based content. Something to be thoughtful of. The other one that I think is interesting is everyone asks me, we do this podcast called Marketing against the Grain. We had a guy called Ignit Smith on to talk about how, how you optimize content for large language models. That's what everyone really wants to know is, well, when you ask ChatGPT or when you ask these AI assistants, what is the best product for this or what is the best service for this? How do I optimize? How do I game the system to get my content in there? I spent many years as a SEO gray SEO hat and I figured figured out all the ways you could sneakily optimize content to get it into Google. The problem with LLMs is it's not going to work that way. And I've been really trying to figure out like how it might work, what are the leverages, what are the hacks. And I started to think that LLMs are much more of a brand tool, like a brand research tool than anything else right now. They're actually a really cool way to actually think, to actually see how the Internet perceives your brand or services. So we launched this tool called AI searchgrader. And when you go into AI searchgrader, you can actually put in your brand, put in your services and what it will do is it will show you two really cool things. It will show you your sentiment analysis and it will show you how aware you are within that training set. So when you think about what LLMs are, they are basically trained on the Internet. And so you can hypothesize that the way you can actually get more results and more visibility in is to just have more mentions, right? The more times your brand is mentioned for the product or services you want to be listed for, the more times you'll actually show up in the results when someone asks ChatGPT a question, Claude a question, Google Gemini a question, AI overviews a question. But right now you can even just go in and say, well, how aware is the Internet of me in comparison to my competitors? How does the Internet feel about me? And what it does a really good job of is this is an example where there's no negative, but when it does produce a negative, it will give you the actual reasons for that. Right. It will actually tell you why people think about your brand or your services in a positive way, why they think about them in a negative way. So if nothing more like I don't think there's really great advice to say, well, what should you do today to start to appear in these results? Oh, other than try to get more mentions in the Internet in a positive way, which is not like very tactical. But if nothing more, it's a really great way to start to think, well, how visible am I? Is my brand marketing working? Is all of these things I'm doing working in to make my brand more visible. So that's the content part of it, right? Content has been a staple of digital marketing. The next one I want to talk about is social. I talked a little bit about this. The historical way we would use social and some people still would like to use social like it was 10 years ago and promote things all over their posts. Does not work that well. Is we kind of used to create content in social to acquire people back to our website. What I think you actually have to do today is build an actual audience. You actually have to build an audience. You have to have a real social media presence and you have to kind of act more like a creator than a brand and that's a hard thing to do. And the other interesting thing is one of the things I think a lot about as well is when you think about the digital disruptions that happen. That's happening. The channels that are being disrupted are usually the brand centric channels like search paid, things like that. Channels that are still growing are actually much better for the individual YouTube, podcasts, these other kind of channels, some of the social platforms. And so you're much more, you're kind of forced to figure out like, well, how do I act more like a creator to get some traction in these channels. The thing that will definitely happen with AI is as much as you're being told that AI is going to optimize this and AI is like the way to integrate things across your go to market. And that's what I'm Going to tell you, humans will want more human connection within that buying process. Right? They still will want to interact with humans at some point across that buying process because they trust other people to make those recommendations. Seven, 10 people trust consumers over brands. But the thing that I think a lot about is just how unmeasurable the future of marketing may be and the things you actually may need to do to stand out. Right? One of the things I did at HubSpot was led the acquisition of this company called hustle in 2001. And people might be familiar with the podcast we got as part of that, which is my first million. And the reason we bought we acquired the Hustle wasn't because we were like, oh, they've got a ton of media assets. They had this really popular newsletter, they had a great podcast, they had some other things and we could just like monetize that into leads. That's really not the reason we wanted to do it. It was because we started to think, well, in the future, I think what you're going to need to have is this kind of media first approach to how you actually stand up real media assets across YouTube, across podcasts, across. All of these different places that I mentioned are still growing. And I thought this was a really great example. Not just for the fact that I could send to the HubSpot founders and say, see, you know, we did a good thing here and demonstrate that I was smart and that's part of the reason I did cut and paste it and send it, but also because it does show you how marketing kind of works, but never really gets discussed. Right. We bought the site in 2021 when we bought the Hustle. The next thing we did within the next six months is we stood up this kind of podcast network. We stood up a creator program, we stood up a bunch of YouTube channels, and this person talked about their buy in, how they actually bought HubSpot. Right? They were a fan of the Hustle. Then they started to see all of these new media assets that HubSpot stood up. And at some time in 2024, they decided, well, you know, I actually need a CRM now. They didn't need a CRM back then. I'm just going to go buy HubSpot because I've heard about HubSpot and all of these places. I love to consume my content. Now that is really hard to sell to a cfo, right? Hey, I'm going to do this thing. And in 2024, you're going to see some results. But that is how a lot of Marketing works is how do you influence your buyers prior to them ever needing your product? And I think that's part of what we're talking about here is that you can't just create content and hope someone goes to your website and they click something and buy something. You actually have to have a real audience, you have to have real influence. And I think AI is kind of really cool to actually start to help you do that. And I'm going to give you some of the personal things that I have been playing around with because what I hear a lot is, well, AI is just a copy and paste tool. It is not. It is an accelerated learning tool. It is a tool that helps you turbocharge iteration. If you think it's a copy and paste tool, you're just not using it right. You're just not using it right. So I'll give you three, like really cool examples. The first one is AI is really cool for building style guides, right? This seems really basic, but it was a huge unlock for me. So much so that actually I posted a ton of content about this. And Claude, the person who runs product for Claude and building Claude, reached out and they integrated some of this into their latest model because it works so good. And so what it can really do is can reverse engineer. What good is. To give you an example, there were. I really wanted to get much better at short form content, specifically LinkedIn. And I was like, okay, well a way I'll do that is I'll go find the best short form content creators and then I will just reverse engineer why they're so good. And there's a skill set to this. There's actually a well known framework called copywork. So copy work, if you're trying to get better at copywriting, is you take a piece of content and then you actually physically write that piece of content out. Every single day you start to learn that style, you can start to recall that style when you're writing your own content. This is really just copyright in an accelerated way. So what it does is I go and I take all of the creator's content and then I have a style guide template. And then I ask Claude, hey, can you recreate a style guide for this creator, for that creator, for another creator? And then I edit them to make them a little differentiated to play around with them. And then anytime I'm creating a piece of content on Claude, I just upload that style guide and so I can start to actually post different content based upon different style guides and start to see which ones perform best. And I cannot emphasize how this can be used in so many different ways. You can just quickly use Claude and I use Claude. You can use OpenAI's latest model as well. But I'm primarily, I love Claude. You can just reverse engineer what good looks like. You get a ton of like outbound emails sent to you all the time. Any ones you like, put in a folder and say, I like these outbound emails. And then just upload them all to call Claude and then ask it to create a style guide on that outreach email like you can think of. You can do the same thing for anything that you're trying to get better at. So when I built the style guide, I was like, oh, that's pretty cool. You know, the thing I should actually do is go build an agent. So Dharmesh has a platform called Agent AI. He was kind enough at the time to give me access to it before most other people had. And what I did was I went and started to create an agent to actually help me post content on LinkedIn. And there's a couple of things I did. The first thing was to teach it right? You actually have to have domain expertise to use AI really well. So this thing where people think it can just copy and you can copy and paste things and an average person could go from great to great overnight. You do need to have real domain expertise. And so I taught the agent what is good short form content. And I can tell you on LinkedIn, there's three posts that perform pretty well after doing a punch of analysis. There's the educational post, which is you can teach someone a single thing in a clear and concise way. Has to be a single thing. Cannot be two, three things. It's not enough. There's too just not perform very well. You have to have like one core takeaway, the spicy take. Unfortunately, this is the way the world works. For better or worse, if you say something that is counterintuitive to the way people think about something, it just guts a bunch of debate, a bunch of traction. And the third one is like the hardest one to do, but really is the one that works the best if you can do it. And that's actually really what your product positioning is and what your brand is. A lot of times it's like, how do you articulate something better than anyone else? Or how do you articulate something that most people believe, but they haven't been able to say it in that way? There was a really good example of this recently from Paul Graham. He wrote a post called Founder Mode. Founder Mode got a ton of traction. The reason it got a ton of traction is because it talked about the fact that founders should get back into the weeds of the work. And all founders felt like they got removed from the work during the COVID boom. And they felt that way, but they were not able to articulate it. So when Paul Graham wrote an essay on it, everyone was like, that's exactly how I felt. Thank you for saying that. They work really well, but you have to really know your audience to be able to do that really well. So these are obviously short form. I have a whole two to three page document that I put together to teach the agent how to do each one of these things. Then I included my writing style. So now I have a, hey, this is how you do short form content. These are all the different writing styles. I have 17 to date that I put into the agent and I combine those things anytime I want to create a post. And I went from like hundreds of thousands of impressions on LinkedIn to millions of impressions of LinkedIn in a very short amount of time, actually in about two months. And so I started to do this in a lot of old places. Like, wow, this actually works, right? This. How do I combine these style guides with these different templates? Really works for content. So the other place I did it was YouTube shorts. And so the other thing that's super interesting about AI is now all knowledge can be templatized. We had this incredible creator called Jenny Hoyas who came on the Marketing Against Grain podcast. She has over a billion views on our YouTube shorts. And we did this 45 minute interview with her and there was a four minute section where she went through the entire. She went through an entire short form clip, 60 seconds. And she talked about every five seconds and why she had put that five seconds in there. She talked about human psychology, she talked about different tests and how she learned that this works and that works. It was mind blowing. The expertise she had to explain every five seconds of a 60 second video. And in my head I'm like, well, I can just reverse engineer that. Now I actually know how to reverse engineer that. So I went to Claude, I said, from that transcript, can you condense the four minutes into a tutorial that I can teach an AI assistant how to replicate this. So again, the other unlock here is I use Claude to teach Claude how to do things. I always say, can you teach, can you create something that an AI assistant can replicate? You should always ask it, always tell it that it's for an AI assistant because it will Actually create the prompt with other. I'm going to get into a whole tangent. I'm going to leave that part because it gets complicated. The thing I would say is always say to Claude, when you're doing this, I want an AI assistant to be able to replicate this and you'll get different results. So I get the short form content, I create another template. So now it's created in a template and it goes through this again is a three page document. It goes through every singular part of how to put together a short form content, a short form video. And now anytime I want to create a short form video, I can basically just upload a long form YouTube piece of content and it will go through the long form YouTube content and pick out all of the talking points and create short form content based upon that. And that is starting to perform really well as well. So again, AI is really this amazing thing. It can reverse engineer good, it can templatize things. You can can iterate really, really fast. And that's the assistant you can actually. I actually was not going to make it available for free because it's the one that I use and then Dharmesh forced me to. So it is actually free on agents AI and I'm still kind of salty about it because it's now in the Claude model as well. So now everyone just has it. Even after like took me three, four months to learn all that. All right, I'm going to go through. I actually. It's not my timer, not my time. I think it is. All right, I'm going to go through the last one which is actually the one that like most people think about AI. It's, you know, we have, we've traditionally in B2B marketing we've done personalization at the segment level. You know, we categorize all these people, we say, hey, you have some things in common. We're going to just personalize our email and everything else for you. AI is obviously an incredible boom for personalization and how you integrate personalization across your go to market. And I want to talk a lot about the AI concierge because I think this is a boom for marketers. Sorry, for anyone who's not in marketing here, that is in sales and other, other departments. I do think marketing are gonna own much more of the customer journey. I think they're probably gonna own anywhere from like acquiring the contact all the way till that contact closes. And even some of the customer support success functions because you can start to automate a bunch of this stuff. And marketers really are the automators in companies. So there are three core things that we've been doing in HubSpot. You've heard a little bit about them today in the email, but I'll give you the practical tips chat and then BDR is the one I want to spend a little bit more time on. I was actually surprised so I, I before I came back to HubSpot, I was a GM over at Zapier for the self serve business and I used to hang out with Kip, who's the CMO for HubSpot because we do the podcast together and we would talk about like our different AI results. And he was like, hey, like we're getting these phenomenal results in email. I was like, no, you're not like email, you're not going to get a huge results no matter what you do in email. Right? Email has been out for so long, it's really hard to like move the, move the needle. And then he started showing me and I was like, wow, like we are. The results are pretty impressive. And then since then I came back, I run all the AI use cases across. Husband and the team have been doing some phenomenal work in email and it's been surprising how well Gen has actually worked in email. Conversion rates have been, we've increased conversion rates anywhere between a high 80 to 100% and this is all just integrating AI and personalizing that based upon the data we have. And there's a couple of tips I'll give. The last quick thing we did at Zapier, which I thought was pretty cool, you know that when we were trying to move the needle at Zapier for activation rates, we really had to figure out how to provide you with the right workflow you wanted to build. Zapier is a workflow tool. It has to automate a bunch of your kind of work. And there's over a million different workflows we could choose from. But we figured out when we used AI to capture someone's integrations and their role, we could actually build them a pretty great jobs to be done guide. And the kind of testament to that is the only time I've got someone write an email back to one of our outreach emails and say, this is the greatest email I've ever got, but I still don't want to buy your tool. But I just want to tell you this is an incredible email. Was using AI to do that playbook. But a couple of quick tips. If you're struggling to get any kind of traction with Gen in email, you want to separate out your prompts. You want to create separate prompts for your subject line, your body copy and your call to action. Right. That's. Most people don't do that, but you want to try to create separate prompts to those things and again try to connect. If you connect the things I've been talking about, you can actually go and take some of the. All of the emails that you think have the best subject lines and you can all put them into a data repository, upload them to quad and again build a template off that and actually replicate things that you think will work really well. You kind of don't see any initial results with email until you've iterated three plus times. So if you're kind of iterating away and you think, well, this isn't really working for me, you have to keep on going, like there's upside there. But it does take some amount of time to see. And then you actually want to have. I've said enough about the style guide, so I'll leave that. But I want to. You've heard about email, so I want to kind of quickly get onto chat. At HubSpot, over 80% of our chat now is automated through AI. And so it's, it's been a huge boom for us, right. And so I, everyone knows the customer support one, so I'm not going to spend a ton of time on customer support. Right. That is a well known AI use case we have today over. I think it's our AI does a work of equivalent of 100 plus people, right. In terms of its ability to answer support tickets and do all of that work. And we specialize that's going to drastically increase next next year as we fine tune the model, it can answer more complex tasks. The one that I'm really interested in is we generate a lot of our pipe through an inbound sales team that are on chat. So when you're a prospect, you're a customer, we have sales conversations on chat and we either convert you into a new customer or expand you into a different tier. And that is the one that I'm interested in AI being able to do is actually have a real sales conversation and actually close a customer. And we're starting to get there. We're starting to see actual real results of AI's ability to actually have a sales conversation and rotate a lead to a sales rep and book a meeting for them. There's some things we've definitely learned. AI is only going to be as good as your onboarding docs and that sounds really, you know, a boring tip. But there was a really great podcast recently from the Klarner CEO and they talked about this, right? Their initial they have their customer support is now ran a lot on AI. I think it's the equivalent of 750 people. That's how much coverage their AI has. And they said the first time they were trying to do this, it really sucked. Did not work. And the reason it didn't work is the CEO found that their onboarding docs for just a customer support rep were not good. Right. So if you were a new customer support rep, they didn't really have great onboarding docs to show you what good looks like. AI works best when you're able to show it what is a good version of this outcome. You know, how do you actually determine whether I've done a good job or not? So the better your onboarding docs, the easier it is to onboard AI to do that. Task annotations. So for people, annotations are you kind of provide feedback on AI, whether it's done a good job or a bad job. And annotations are, hey, like, these are how you actually can improve this answer, this output. You want to have your best people doing the annotations. Most people do not do that. They give it to like, you know, they think of it as a throwaway task. It's one of the most important things you can do. So actually pulling people off work just to do these annotations is really important. But you want to have your best skilled people. The people who know the role most deeply do those annotations. And then human nuance is hard. I'll skip past this. The human nuance is hard, which is basically, there's a lot of things that happen in a sales conversation that are really hard to capture. So there's a ton of fine tuning in an iteration to actually make it really good. And so what AI kind of tries to do is answer your question, give you the answer really quickly, and then say, my job is done. But anyone who does sales knows that's not what happens in sales. You kind of pull on a thread, you hear something, and you can kind of thread from one conversation to the other. There's a ton of fine tuning to help AI do that. The last one I'll just cover is bdr. So I think this is a super exciting one. So one of the things that we are starting to, or we've done a bunch of is like how AI can do prospecting. And we have seen a huge amount of results this year and have an AI prospect do business development to inbound leads, right it it generates thousands and thousands of incremental meetings a month and there's some interesting things that we found out about how to use AI for business development in different segments. SMB, mid market and corp. What I will say is mid market and corp are email is being really saturated by AI and so mid market and corp it is really hard to like prospect and book meetings in those segments if you are not also using call and social outreach. Email alone is not enough. You have to integrate AI into the actual BDR department and have it help them do their tax. But. But you still need humans to be able to call to do a bunch of the social outreach. We've also found that cold is really saturated. Like AI is starting to saturate things really, really fast. So when you're just trying to prospect into cold demand, right you have no interaction, no engagement with these companies. It is a kind of red ocean right now. There's just a ton of outreach being done by AI. It's really hard to get results. Inbound working great. Inbound is working really, really good for AI and business development. Unique data truly matters. The more unique data so the more you can craft a personal message for someone with differentiated data data that is just not off the shelf and everyone has access to your email is a little different. Your prospecting is a little different. The better. You need existing members to. This is the big thing. I'll kind of iterate. If you really want to make progress, you're going to have to integrate it into the teams and part of that team's job is to actually help you fine tune the model, help provide annotations. So for us, our business development team are going to look at the emails. Is that a good email? Is that a bad email? We're going to write them call scripts. They're going to be able to tell us this is how I would have made the call script better. This is how you could actually make that call script much, much better in the future. Right. You have to have it interwoven into the team and the team to provide feedback to improve the model. And then voice is complex. Due to legal this year, I thought we would start to be able to run kind of some pilot lightweight voice AI enabled agents to be able to do lightweight business calls. That is happening in Europe today. I know some founders that have technology with people using it to do lightweight business development calls, but primarily in B2C. Think of like someone selling a course or an education package. It works quite well, but there's like some legal complexities. I do think voice is going to be integrated across AI go to market next year much more than it is this year. But it's something I would start to look into. If voice is part of how you go to market, if it's in your go to market, the thing I'll end on and then I think, I think we have like five minutes of questions if anyone has any. But this is what I tell people. So like one, one of the things we're going to run out in the next month is a pilot with an avatar. Right. An AI avatar is going to help do a product walkthrough for a small segment of our users. And I look at the avatar and I walked around our conference in Boston and I was showing people the avatar and I got two pieces of feedback. I was. They were like, this is weird and I'm freaked out. And I was like, oh, it's probably not going to be a great test, but I still want to go ahead with it just to get some feedback and. Because the way I think about this is that's the worst version of that avatar there will ever be. Right. And that's why you want to try to get signal much, much earlier. We are really in the Ask Jeeves era for AI and go to market after Ask Jeeves there was Google. Google made things much, much better. We're still not there yet, but you want to try and experiment with these things really early. First mover advantage has never been more critical in this space. I can't emphasize that enough. I do think saturation is going to happen much faster than it's ever happened. And so those first movers are going to get much more upside before everything kind of equals out at some sort of rate. Yeah, that's me. That's my brain dump.

Two questions. If I can go quick. I'm in the middle here. So what do annotations practically look like and how do you onboard the people to do good annotations? And the second question, are you recording sales call data to turn into transcripts and getting data from that and using AI in any way to do anything with those recorded calls? So annotations and recorded calls?

Yeah, on the annotations we built an internal application to build do annotations. So for example, we'll bring in the chat and then we have the team can actually say was this a good answer, a bad answer? And how would you have made it better? And then we feed that back into the model. So we've done that in a bunch of different places. The way you onboard the team is you basically again, if you have an onboarding doc for that role and you ask them to make sure that it adheres to like what is best practice, that's kind of how you have them integrated into that. We use a bunch of unstructured data to actually train our models. And so I do think one of the booms for AI is its ability to take unstructured data and turn it into something valuable. I'll give you one really practical example that I think is one of the best we did, which is we have a, you know, we have customers return. A lot of customers actually never provide a reason. So you can't make that problem, you can't fix that problem for them. And what the team did is they took all the unstructured data from those accounts and through the email and the chat and the, and the calls, they were able to infer a churn Reason for 2 out of 3 customers who didn't leave a churn reason. And that way we were able to like start to fix their problems proactively. And so this is like one of the, one of the really cool ways you can use AI to take on structured stuff and create structured data. I think I have time for one more.

So just a quick one. It's about the, you mentioned training models and I'm just curious, what would you advise in terms of training a private model that your company has, you know, gone open source or using Chat, GPT or CLAUDE for. Yeah, for these purposes.

Yeah, that's a great question. That is a great question. I, so I'm a believer in open source. I actually, I do, I do a ton of like preceded C stage investing. I'm investing two companies who are going to try to help. This is not pitching those companies, but it's trying to help companies actually integrate AI open source models to solve singular use cases. But that technology is not there yet. So it's actually hard to get open source models, I think integrated into your backend systems. We primarily are still like using a lot of the primary large language models like ChatGPT, Claude and playing around with those. So I think that's a good place to start other than if you want to. If there's like a specific use case, there are some open source that are just better for specific use cases and you can do some research on that. Again, I don't want to promote one of my companies, so I won't say it, but there's like a bunch of places you can do research and open source models and will tell you the use cases that model is best for. But it is a bunch more work. All right. Thanks everyone. Enjoy the rest of your event.