Name: Signal Rich, Action Poor: What AI-Native Revenue Teams Do After the Insight
Uploaded: 2026-09-17T17:16:46.372Z
Duration: 1 h 1 min
Description: Signal Rich, Action Poor: What AI-Native Revenue Teams Do After the Insight

Transcript for "Signal Rich, Action Poor: What AI-Native Revenue Teams Do After the Insight": I'll commit to whatever you do. And we're live. Hey, everybody. Welcome to an incredible session that we have today. I'm Jared, cofounder of RevGenius. I'm joined with four absolute all stars. And, you know, the the key topic today, like, signal rich action for what AI native revenue teams do after that first insight is absolutely massive. Right? Like, every revenue organization has a system of record as a system of intelligence, but the work still often depends on someone manually taking action. The team's pulling ahead, like, the folks running them, joining me today, are building a third layer, a system of action where signals automatically become completed tasks. Before I go into the intros, let's get, like, a sign of life here. Everybody drop your LinkedIns in the chat. This is going to be an incredible question and answer session where we're gonna learn from literally the leaders in the space, what they're doing with their teams that's driving success. If you have questions throughout, please put them in the chat. We'll probably pick a few along the way. We want this to be a tremendous learning experience where you get a lot out of it. You learn what these folks are doing, but you could also ask questions to to help, better visualize that for yourself. I wanna thank our sponsors for today, Airspeed, an incredible, tool and, call intelligence. And and, actually, Adam, what what what are you all calling it now? It's not call intelligence. It's, It's, I mean, the space is moving fast. I think the if I were. to. kind of use just a single phrase, maybe AI chief of revenue AI chief of staff or revenue teams, it's really kind of building that one assistant for, for revenue teams that, you know, knows your goals, is linked to all all your systems, know the whole knows the whole context, how you win, how you lose, all that stuff. Hell, yeah. So that that's Adam Liska, the cofounder and CEO of Airspeed. Alex Lindell, the creator in residence, GTM engineering at Clay. We have Minta Soe, the senior director of Agenctic selling at HubSpot, and Tyler Phillips, the head of AI and director of product at Apollo. You all use some or all of these products. You've seen these teams and how they operate because they're they're loud about it, on social and stuff. The results are in the news. Let's learn. Alright. Let's let's jump in y'all. To get started, three layers why leaders can't tell them apart. Let's let's start with the map. System of record, system of intelligence, system of action, what actually belongs in each, and where do you see teams misfiling things? Yeah. Mitesh, No. do you wanna mean started or or Adam? oh, Minta Minta, go ahead. Sure. Oh, sure. Sure. Sure. Well, I'm a little biased on the system of record. I also think system of record feels, passive when in reality, it's not a passive. I think we think of it as, like, this warehouse, which that is happening. There's definitely, a a place for that. But I think your system of record needs to be an active participant. So I think CRMs, absolutely, obviously, there's the big two right now, HubSpot and Salesforce. And then as far as system of intelligence goes, I'm moving towards a world of thinking about being, agnostic, about providers. So I'm I'm a huge fan even personally about harnesses right now, which I think is something really cool that Airspeed does is they're thinking of being an agent harness, from what I know about your product. And then as far as system action, there's many, many. I I'm a huge fan right now of a playbook, for reps working a ton in co work. I've spent a lot of the summer, working on that and learning. So I think those are the big three that I'm thinking about right now, as far as, like, what's in all like, you know, what the starter pack. If I had to give someone a starter pack, it would be those three. In terms of, like, what companies I think are getting wrong, I think we haven't determined I use the word choreography often. It's probably not the best word. You can also call it runs. There's all kinds of things. I think we haven't determined how to use these things, and we're in a place where reps are trying to make sense of all these things, but we haven't determined the playbook yet. So it's less about the stack and more about what is the intended outcome and the intended behavior. So I'll leave that there. Yeah. To to add on to that, I think one thing that people don't realize or have thought about yet is that that system of intelligence is really, like, three layers in in and of itself. So you have, like, the intelligence, which is like AI, but you also have, like, a context layer that brings everything together. And then you also have a learning loop that takes feedback from the, layer of action, and that becomes kind of like a loop between the action layer and then the intelligence layer. And what we see in, like, the engineering world is, like, they've already, shown that loops can build things, and, AI will loop on something until it gets to that outcome. And I think now we're shifting into a world where that's gonna enter into the go to market space, and we'll eventually have these loops in almost everything that, everything that, like, the business does. And so I think when we think about these three layers, specifically intelligence, you know, think about the intel think about the AI, the context layer, and then the learning that happens in that intelligence layer. I don't think, most are thinking about that yet. What we see is, like, people start to go down the road of, like, hey. Let's let's add AI into, like, everything we do in go to market, and then they realize the data layer or context layer, isn't quite there yet. And then they're not getting the, quality outputs that they want from the, from their, like, AI. So that's that's where I think, like, where people get it wrong, but also where we're headed and and what we need to start, like, thinking about. Yeah. Plus think, one. one of the thing I'll build off that, Alex, is, so what you just described is, like, say you have, like, system records, system intelligence, system action. I actually would argue that you really just need one system. And and not in terms of, like, the product. I think that's a misconception that, like, you only need to use one product. I just think that, like, if you think about your system in three different layers, which they have layers, sure, but, like, they're technically need to be one system to be learning over time. because you do need to know, like, what, like, the past or things about the context or accounts or emails sent or whatever. And then you do need to have intelligence, let's say, to then understand based off of the data and signals that you have, what does this mean for your business and what you should do. And then from there, you need to act. But from the system of action, which I argue should just be, like, one go to market system or for all three, that like you said, that feedback loop, if that's not connected, let's say, to your system of intelligence, you're not actually gonna get that loop. So that doesn't necessarily mean you have one product per se. Of course, many companies, including Apollo, will probably say, hey. Yeah. Like, you know, use one. But I think in general, I just like people would like I think in general, like, you see a thing through, like, how you architect that system, whether you use, you know, a subset of tools or have it all hosted in cloud and connect all through there, whatever it may be, or APIs, whatever it is, a bespoke system. I just think that's an important, distinction, and I would argue that's what people miss files. So, like, you have these three systems that don't talk to each other. Yep. So yeah. Yeah. A good system of record technically should be able to have all of the things you need in it, but I don't think we're all there yet. I think that's I see that's where the industry is going to go, is that these things will merge. Because right now, it does feel like there's three to four parts you need, but I think that's the road maps. I think that is exactly where products are going that you won't need that in twelve months, because it will all start to blend together. Now a leader sitting on a stack, they didn't fully choose. Right? Like like, this is the perfect world. You guys are picking which tools for each. They're they're told to consolidate. How do they even work out what they have? Yeah. I mean, I really good question. I think kind of goes back to, you know, you wanting to really build just one system from all the tools you have. Ultimately, what you want your team is to be performing, you know, doing doing more and doing things better. And I think it kind of this should this is the end goal, I think, of all of us, and of all revenue teams. And you kind of need to work backwards, I think, from there. You know, look at all the tools, understand, you know, is it giving me any useful data that I can, you know, use within my single go to market system that I'm building within within my org? Could this be, you know, the layer that sits on top of everything, or can it actually feed and give give that data to that one interface that I'm that I'm building building for my team? I think that's kind of where where I would start just diagnosing what what I'm what I'm sitting on, and then how, you know, how useful it is. But I think another way that I would do as a as a leader is just to look at my my top performers and understand what they use, what they do with our data. They most probably you know, your top performer is most probably doing something interesting that, that you should you should scale and kind of replicate across the team. You can potentially replicate that within that one system. You just, you know, set set it up better, or you then go and look for a solution that kind of does that for you. I think that's kind of how how I would approach it. Yeah. Two things that have worked well, for me is one, doing a process map, a good old fashioned one. I've actually think and but tagging which, tools our rep is touching. If you're in a big company, it will be shocking when you actually, I've also done, like, a click study. So, like, measuring how many clicks it takes to do some of the revenue generating activities. And then, actually, I think because in general, a lot of the things we're doing, time savings is the KPI, which I'm a huge hater of time savings as a KPI because unless you know how you're reinvesting it, you're in the business of making dollars, not just but more time. But I think it's you can definitely infer, like, the performance if you've improved the process with, like, how many clicks it takes to do something that's revenue generating. And then that group can perform better because you're gonna then do the second part that you mentioned, Adam, which is I would a 100% do a study of what the best people are doing. So, actually, over the summer, I have been really fixated on improving on call sales behaviors, and we did a study of, like, what do the best people say, and how do they have control over the call, and how do we make sure that we actually do that? Because in my personal opinion, I I like to have very draconian perspectives. That's just my specialty. I think, the call is where the selling really happens. You can't outdo it's kinda you can't outrun of that diet, can't outdo a bad call. So I think those are the two most important places that you should start in my personal opinion is, like, what are the best people doing on calls, and then what are the systems and clicks and inefficiencies it took to get there and after? Because That's how you make money. The rest of it doesn't matter as much. It does matter. So, again, with the draconian perspectives, but it doesn't matter as much. I'm gonna I'm gonna take us in a different direction, and think about this in the world of, like, coming from a world where we have a small number of people building to now in the AI world where your whole team could be building different things. And so as you think about now that you have all these builders, I don't like the the time savings metric either. But we have a saying at Clay, which is like more is more. Ultimately, you save time. You're just you you know, you're gonna fill it with something else. So our goal is to produce more or help reps manage, like, a larger territory, so you get to better, like, revenue per headcount. But going back to the building thing, I'll take a step back and look at all the primitives that the team has to build. So you think about it, you need, like, first party data, you need third party data, you need your, like, AI, you need your CRM. You need all of these different, things, and then you assemble these primitives in different ways to build, like, the workflows or the GTM systems that connects everything. So I would think take a step back and, like, you need to do everything else. Like, what are the primitives that you need moving forward, to, like, build the next generation of, like, a go to market like system? And, you know, that that may be that may be consolidating. It may may not be as long as it, like, all talks together. Alright. Let's jump into where a handoff breaks. Okay? So the premise of the title is that the work still lands on a person at 6PM on a Thursday or Friday. Where does the handoff actually break? What kind of handoff are you talking about? Customer, task? I think that I could go into each one, but I think that really is what the difference is because I think task handoff is a different conversation than customer handoff. I think one needs to be more, conservatively managed than the other. I think customer handoff is the last kind of handoff that can take a risk. So when you're thinking about how all of this works, I think, the hierarchy of conservatism of how you, you know, agentify your go to market, the the one you're most conservative about is a hand off that a customer has to experience or your end user, and that's where your highest risk actions are. As far as where those break, I think the the biggest and most key area is when, there is a lack of control on any type of customer facing interaction, that goes between a human is guiding some of it and then AI's a part or drafting or in any part, I think that hand off can be a little tougher. Yeah. And I'll give some examples. One is obviously outreach. You gotta be able to back up whatever you sent. Step one. And sometimes it's easy. Sometimes it's not. The second one is, when you're preparing for a call, like, sometimes the knowledge we we think is available never hands off to the human who needs to then say it. Like, sometimes the hand off is less obvious than we're thinking. I actually think the biggest hand off that the entire industry is dealing with right now is, like, think about how much you get back in AI. Right? It's usually like, do you actually read every word your prompt gives you back? No. You skim for the answer and you move along because there's a lot of words there. So guess what? You had a hand off drop off of the knowledge that was in there to your brain. You probably absorbed 40% if you're lucky. So in my opinion, when you're using AI, agentic, anything to drive human behavior, you have to capture the drop off between steps, between what you think should be happening, what your systems are telling a person, and what needs to happen that then needs them to do something with that information. So those are my big two that I think about that keep me up a little bit at night. So? I think I Okay. good? oh, sorry. Go for it. Go ahead. No. Please. Please. Alright. So so we obviously created, like, a lot of workflows at Clay that support the reps, and those could be, signals from, like, a customer, a lead, you know, follow-up to approve, like, all sorts of things. And we started to pipe all those things into, like, their own, like, Slack channel. And then there are so many alerts or notifications, but then it just became, you know, signal fatigue. So then we stopped paying attention to, like, almost all of them. So I think there's all the part of the reason there's a drop off in some of these areas is, like, you're just giving someone, like, way too much, like, information, and you gotta figure out, like, what is absolutely the most critical, like, next best thing that person can do. And it's gotta be, like, a small number. Otherwise, like, nothing gets done. But it's, like, really hard to get that, like, balance right. And, going with that is, like, the trust. Do your reps trust, like, the data? Do your reps trust the what's being recommended? And then do they have a say in that process to help, like, improve it? So if they don't have a say in the process to improve it and then people aren't helping to, like, improve it and iterate on it, then they're just gonna, like, ignore the system. So that was, like, a key learning from, a lot of our workflows was getting that balance right and thinking about how we build trust with the reps so that they actually want to use it, and take action on the next best step, because they understand, like, everything that went into, like, getting there. Yeah. Okay. Hey, Alex. I would say one thing, though, really quickly about that is, is that, like okay. So I agree with, like, notification fatigue. I think the first part of the handoff, that Mintus brought up was, like, I think waiting for reps to also, like, interact specifically with the agent to human, like, hand off is, like, if we're waiting like, if you build a systems where reps have to do the work to interact with whatever AI you want them to interact with, I think that's the first pattern that necessarily doesn't really make sense because you want agents to be work on your behalf. And, like, we've also seen that at Apollo when you, like, build products where you expect people to come into a product and start prompting it. The interaction is one is they feel like there's not really saving any time where you're also not even producing, like, better outcomes because they still need to do good amount of work to, like, massage, like, the AI in that case. I would say the second thing, though, is, which I think is even more critical, today is, where, the handoff occurs. So a lot of times people assume that, hey. Like, they'll go and, you know, obviously, no one's going to their Centimeters and a slate CRM. So then they'll also say Slack. I think everyone has a different preference on where they wanna work, and I think that's something kinda that's been brought to the forefront with, like, building headlessly. Some people wanna have their own micro app. Some people wanna work in cloud. Some people wanna work in Slack. And I think that's, like, one really critical point is, like, going to where people are, not where people where you want them to be. And as you build systems, like, considering that in the, like you know, so everyone can interact in the way they want to, which leads. to a much higher doctrine because you lower the friction of usage. So that's one thing that at least I've noticed, like, for Apollo, per se. Yeah. Plus a thousand. They'll. also go. wherever they have the most freedom of the data, as in, like, the data is as most connected and proliferated. So even if that's not the best system, I won't name the system names. Connect with me if you wanna know more examples of what I'm saying, but we had so many examples where, like, something was more connected to other data. Therefore, like, the answers were theoretically smarter, but the brain itself wasn't. We, like, had a tough time getting people out of that system and into the one that, like, actually had the context we gave it, but the context was was, like, more, like, gated, if that makes sense. And but people want they want it to span everything. They like if you allow me to access things, I'll take crappier, insight, but in with more control that I feel. So that's that was a very interesting learning early on. Well, Linda, now I gotta set up time for you because you you really piqued my curiosity there. But two quick things here as well. You know, 100% agree with, Alex's points around, you know, the the amount of data for, like, you know, over the past, you know, decade, maybe two decades, we've been going from, you know, having not enough data to right now just having access to so much data that it's, that it's overwhelming. And I think that kind of the power of AI now is actually maybe just reducing the noise and giving you only the things that you that you need. And I think that's that should be kind of one of its jobs as well. Otherwise, yeah, you're not gonna you're not gonna be paying attention. You're not gonna be acting on on anything. Second point, kind of the way I understood the handoff as well is around, you know, where do we like, at which point do I hand off my tasks or delegate my tasks to to AI? And I think here, we kind of as a as a field or as humans in general, I think we're just a bit too much, conservative. We think about that a lot at at our speed where, you know, you as a mid market rep, you, you know, you're managing your top deals, and you you will be just really focused on your top five deals and then overlooking your 80% of the bottom deals and only do the task there. You know, at the end of the day, okay. I need to follow-up. I need to send this information. I need to reengage. And, and very often, these things just don't get, well, don't get done at all. And I feel like we should just be a bit more comfortable delegating these tasks to, to AI. It's gonna send an okay email, you know, surprise surprise maybe for everyone. If it makes a mistake, it's fine, but we humans and, you know, our reps and ourselves, we're making so many mistakes as well. And so I think it's fine if an AI makes a mistake here and there. But if it actually manages to reengage one of my, you know, deal that I thought was was dead or that I'm just kind of dropping the ball on, I think it's worth it. And so I think we should just be a bit more comfortable handing over tasks to AI, even even customer facing ones. I'll be honest. Every mistake I can think of in the last six to nine months that was actually a problem actually started with a human's mistake using AI, being honest. Like, it was like the the criteria was wrong. Something, like, went wrong in the, instructions or, you know, something that was the setup. It it wasn't actually so I do think it is very obedient. So if it goes wrong, I think it the the funny part about it is it's very often between the chair and the desk that caused the wrong part even though the the the lack of trust is something that we have on the technology. It's just following directions. So, typically, when the directions are not right, I think that's the the part that we haven't gotten perfect yet. I think we're obviously all like, you know, we made a ton of progress in just this year alone. It's crazy. But, that's that everything that I can think of this year that was like, oh, wow. We made a mistake. It was literally all because of the person, that gave me instructions. that that's a phenomenal point coming from a massive organization. For for six to nine months is a crazy amount of time in 2026. Right? Like, how we're evolving so fast. What is HubSpot done that you can share to enable the human to better operationalize, like, the AI. Like, it's great to put the blame on it, but you guys are a world class organization. You've done something. Yeah. I mean I mean, we started with proliferation, to be honest. I don't you know what's funny? We actually haven't spent a lot of time like, we don't do AI training specifically. I think we focused on, like, proliferation in the form of getting everyone access, and then allowing the people that really wanna run fast and be leaders to be leaders, and to push the envelope, and then making an example out of those people. So I think we really create a culture around, almost like a coopetition, if that makes sense. Now I'll tell you where this has bitten me specifically in the behind, but which is in the sales org. But I think in general, we have a culture of, like, you know, I don't think in AI you can focus on giving people parameters and then, like, controlling how they use it. I think if you wanna be ahead, you. have to let people have it, and we did that. So I actually started the year off in an operational role in marketing, and we started, I'd say first. I think that's and then I would say that we kinda like the rest of the org, kind of followed across, like, operations and sales. And I think we really started because the use cases were there a lot more, for, like, you know, general consumption. And you do have to be mindful at an organization as large as HubSpot because of so many, pieces of, you know, governance liability. There's just so many more things when you're talking at the enterprise level that we're still working through of, like, how do we do this properly? But I if I had to give one answer, it really would be, like, proliferation and giving people the room to move and to do things and to try. I had an amazing boss and a huge shout out in to Kieran, who's the agent go to market leader. I've been reporting to him for a couple years, and he's never told me no when I had a crazy idea. And, whether it's, like, just building something, and I that's the culture that wins. Like, I'm queen of crazy ideas that people will tell me. Recently, I had someone tell me, he's like, you don't work well under constraints. I'm like, you're right. I don't. I don't wanna be anywhere that I have constraints. I'm reality. I like reality, but I don't like constraints because I will figure out how to get out of them. That's the whole point. That's what you pay me for. Give me a pay cut if I like to operate in in constraints. So you need people like that as a setup start. And then those people need the people who are like, okay. So there are some constraints. Here's how we're gonna do it. So I personally think there's like a persona starter pack you need in every or of every size. You need a little crazy person like me, and then you need some really amazing technical people around that person. And if you have that, you'll go really, really far in my personal opinion. So that was a really long winded answer, but I feel like that's been a summary of the year. yeah. I I totally agree with that. You you gotta give people time to, like, tinker because it's through the tinkering and, like, experimentation that you actually discover, like, what's possible, what works well, what doesn't work well. And then you you start to, you know, build a mental model of, like, how to use AI, like, productively. But you kinda go through this, like, unproductive phase where you're tinkering and experimenting, and then finally, you come out of that phase, and you and then you're, like, more productive. But you don't you don't kinda, like, immediately become more productive. You gotta go through this, like, you know, this trough of, you know, actually investing some time into it. Yeah. Now we're we're at a time when reps, are spending more time in AI assistance than in the CRM. You know, two years ago, it was wishful thinking that we could have a quicker way to enter call notes. Remember? Like, that that wasn't long ago. But now that's been solved and then some. Right? They don't need to do anything. But, like, what does that mean for where the work actually lives? Yeah. I I mean, this is this resonates so deeply. I I think I posted on LinkedIn a couple weeks ago that I had a rep rebuild lovable. Like, that was really, like, a moment for me. I was, like, literally laughing on the call there. Like, what's so funny? I'm like, because you didn't build with lovable. You rebuilt lovable. So I have, I feel like every other week, I'm talking to a sales rep who's, like, just so excited to do things. They're, like, rebuilding tool stacks of the literal things. But as far as, where it needs to live, I mean, it does need to live in the CRM, ultimately. Like, I think we really will have an issue if it doesn't. Because I think, in general, you everything needs to go back there. I think some execution can happen. Some interesting areas I've seen that work really well that are not in the CRM is, like, deal rooms. That has been really interesting, like, interactive point that I've seen, reps really want some, like, innovation around. And, like, again, message me if you wanna know more about, like, what vendors, but we're using some vendors. And the reps still wanna see, like they're, like, making their own version that's even better than what's out on the market because they really want to see, like, some of their connection points with the reps with, customers get more, like, familiar, and they've even created messaging platforms. They can talk to people at any point. So, anyways, to answer your question, the CRM. But I'm noticing that they they want to get closer to prospects as well in ways that are not in the CRM also. The the messaging platforms they're building, they're making sure to API and it's HubSpot, or. are they running. Renegade with. it? I'm just talking or or this or or there is all these so this this goes to the original question. Right? So now they're building something all the way out here that's helping them here. Who's who's tapping them on the shoulder saying it needs to come back to the CRM That would they're me. going that renegade? That would be. me. However, I again, with that whole culture of, like, you know, letting you run, I'm always that polite person who's like, okay. So let's not do that. But I like the I like the spirit, but can we figure out how to put it back in the CRM? But I'm just I let some of those smart brains run wild because there's a lot to learn from them because they're frontline, you know. Yeah. Tyler, what are you seeing? I'm just really perplexed by the case of it must live in CRM. Yeah. I the CRM was not built for AI native world, like, flat out. Like, it was built for a world where people can input information into it, and it was meant for managers to see what their reps are doing. I think in a world where, like, data is so prevalent, AI is everywhere. I just like, even in, like, Apollo, like, we have a like, wait. No. We have, like, over 20,000 DAUs now on Apollo, MCP. That's our fastest growing product in Apollo. We built our AI product for assistance at Apollo for, like, two years now, and it's already eclipsed that. I think that the reality is is that people are gravitating towards working in assistance because it's a place where they feel almost a sense of, like, anything is possible. And I think that mindset is really important, as you're building with AI versus, like, working in a CRM, even, you know, product. And I think that, like, in terms of, like, where work should live, I don't know if I agree with the question because I think work should live where people want. to work. And. I think that the second point to that is is that, like, there should be a store of data or record. That is true. Whether that'd be a CRM or not, I'm not sure if that's the case because I've primarily seen that, like, databases or even, like, Billy on Top of Snowflake, or something like that is potentially even a better system because it's so flexible where all the data is available for anyone to work wherever they are. But that's just my own, like, pontification there. Just that, like, people want to work where their agents are, and that's not in the CRM. So agent should be. in the CRM. That's the key like. the word should is the hard part I get with. Yeah. Yeah. Alex, do you just so happen to have? we we yeah. Yeah. The another, system of action person. Yeah. Yeah. I'm I'm, like, looking at this room, and I was like, there's a clear Yeah. path here. But yeah. Yeah. I don't I don't think the future, for reps is working in the CRM. They don't need to. They don't want to. Correct. But the challenge. is the business needs the data to get in in there. Yeah. So that needs to happen, but I think that could all be, like, automated. And that interface that reps use will become will become something else. Yeah. It it just it just can't be the future. So here's what I will tell you. I don't disagree with you, but I do disagree with you for big companies. Because, when you have like, you're right. Reps don't wanna work inside of there. But I think one of the biggest pieces of feedback that I hear from I used to be in sales at HubSpot, and all of my friends have most of them have. gone on to new roles. And the biggest they work at some of the biggest startups now, and none of those companies are caring about their rep experience like a big company would because they're why why would they? And they're like, oh my god. There's so many systems for me to deal with. That's the number one thing I hear. And some of them, it actually impacts their performance because they're, like, in so many places and they're not If they're not organized at doing that, it's like it's a new role. So some of your most tenured sellers who are really good at selling, they are not used to this. They're just not. So I think the answer is different in enterprise versus, like, an AI native company that has a smaller org that this is the DNA. Like, Clay's DNA of a seller is, like, a much more technical person, who's, like, you know, building and just as technical as they are a salesperson. That's not the case for all companies. So I think that's an important thing that you have to figure out is, like, what's the persona of your seller with the tenure? Are they AI native sellers? Some most of these enterprise sellers aren't. They've grown up in a world where they had one system and that it's been more curated and they didn't have to go in and talk to their agent somewhere else and do this and that. They're just not used to it. So if you have that kind of sales order, you have to keep that in mind and you have to do different kinds of trainings, enablement. You gotta think about your stack differently because you might have the best stack in the world, but if your people can't, like, operate in it, then what? You know? So I think I don't disagree. I think the answer is different depending on what you actually need to achieve. Do you have 2,500 sellers. that have to do this? You can't have them talking to five agents. It won't. Totally. But I would say yeah. Alex, go ahead. I think those reps will, like, eventually learn or have to learn. Otherwise, people are gonna, like, outwork them, and they're they're gonna get left behind. Yeah. Possible. The way I see it is that, you know, the company the company does need those structured data somewhere. You know, it can be it can be in a CRM. It can be, you know, in a in a data lake. That that kind of flexibility is a it's it's a double edged sword. And so if you wanna manage it, manage it. If you wanna use something that already has so many integrations, you know, kind of outside of the GTM system, I don't know, into into into financial reporting and all that stuff. I would say stick with the CRM, but make sure that all the other systems are are talking with the with the CRM. But I 100% agree that kind of people will be working working wherever they they want to work. And we've kind of looking at, you know, software engineering. I think software engineering is really ahead of the curve in terms of, how AI is changing changing the work, and we've seen a massive change of how where where rep where where software engineers work. They're they're not kind of in their IDE coding at one letter after another. They're really kind of doing their high level actions in cursor, in cloud, in whatever system they're using. Ultimately, it needs to land in the code base, and maybe we can consider the code base to be the CRM and kind of things land there. And that's kinda where you have where the whole company collaborates around, but, everyone's using something that makes them makes them work much faster. And I think that's kind of definitely where we're going. It's a great quest it's it's a great answer and a great point. I I wanna, go into call recording, as the nucleus and the closed ecosystem problem. So there's an argument that call recording is the nucleus of the agent to go to market. Everything upstream exists to get you on the call, and nothing downstream works without what happened on it. Is that right, or does it overstate the case? Well, there was a there was a startup that just launched, I think yesterday that actually blocks your recorder from getting the other side, of the conversation. It, like, disrupts the the agent. So if that becomes more, popular, then that's gonna that's gonna be a real big problem. What do you think, Adam? You've you've lived in this space. Well, I'm I'm well, I think, you know, I agree with Mintus' point from from earlier where that actually the selling happens on the call, and that's kinda where, you know, where you connect with the prospect and when you kind of align and everything. And it it is the central point, I think, of of the selling experience. And so calls will be there. We will either you know, we will we will be able to record them. If not, we will be able to maybe, you know, transcribe them or send voice notes afterwards. But kind of what happens on a call is central to your to your kind of go to market journey efficiency, your motion. You need that data to figure out kind of which playbooks work, which don't work, and all that stuff. But so we call as a conversation is central, but it's not not enough. You need all the other data, Yeah. in order to build a proper proper system. Like, you you can't build any any any, you know, commercial brain of your company or any kind of that kind of central nervous system of your company if you don't have access to everything that a rep has access to. Otherwise, you're always getting, just a limited picture of what's happening. other the other interesting thing too is, like, the the show happens on, like, the call, but the deal in reality gets done in person over Slack, over text messages, and that's, like, where the real, like, juicy or critical moments happen, not necessarily always on the on the call. And those things are, like, really hard to, like, capture. Yeah. I'll say one thing about conversation. So, I did look at this pretty deeply, like, how you use conversations to, like, improve your go to market motion. I think there's, like, two things I wanna say here. So one is, like, for early stage companies, it doesn't matter. The reality is that, like, you're just trying to get people on a call. So saying that you're gonna mind conversations doesn't make any sense, unless you're doing it for customer interviews. Because you just you need to focus much more on how do you, like, get people to respond to the message you send, whether it's email, LinkedIn, other channels, and then learn through, like, mid to get message market fit. Although for more established companies, I 100% agree, that conversations, matter, quite a bit. It's not just one, like, caveat I wanted to to bring up. It's like it really depends on where you are, like, in the life cycle. Hard listen. I'm gonna challenge that though. Because, like, let's say you get a whole bunch of conversations and you close none of them because you had a terrible pitch, and at the end of the day, that's agent prospecting, not agent excelling or, like, signal. Signals are. often we're thinking a lot about the top of the process, which I could understand because I think a lot of people on this call are thinking about getting to the call. Clay, Apollo, they're all about getting you to the call because that's where their ROI shine is pipeline creation. But But if you're talking about how to get to the dollars, you have to think about the whole process. And I used to own the prospecting motion at HubSpot, and I would have given that exact answer when I did. But now that I have owned the full cycle, I have been forced to care about everything. And I think you have to care about the call after that, because I probably would have had a different answer a year ago. I'm cared about all about because, like, I don't think you need the call to get to figure out what gets meetings and creates pipeline. You don't. But to figure out what gets to dollars, you do. And that is, at the end of the day, what companies care about right now because you could have a massive conversion problem. Then what? Yeah. And and, you know, bringing to the next question, like, teams are paying 6 figure contracts. Some of the bigger ones, probably 7 for call capture, Who's that company? put. Just curious. Just kidding. The 7 figures, No. I won't say the actual company. don't have, I'm saying the product. any of them. I think, we could, guess what, me to negotiate. don't don't trigger me. Don't trigger me. yeah. I I'm right on the. edge right now. Sorry. They're for it. also paying, they're they're also paying 27,000,000 a year for Salesforce, AIforce. But hope you have five we'll 5,000,000 sales reps. we'll we'll we'll. we'll we'll put that somewhere else because I I just read that yesterday. I I had to look at the number. I'm like, why is 27,000,000 coming out? Oh, that's the annual contract. Like, anyway, like, in 2023, there there were different criteria for for ROI and success with call recording tools than there are today. What should people or what companies, leaders, expect to return in 2026, that they weren't expecting in 2023. I think when, you know, back in the days, I think we were really interested in visibility because that was nonexistent. I think we've got the visibility now, but you should really be asking about kind of how it drives productivity in in my organization. I think that should be the tough question where, that that's the ultimate goal, and that should always be the main question now. I think now that we've got the visibility, we can we can kind of understand what's going on if we if we spend the time. But if it's not, you know, if it's not moving the needle on productivity, I think you need to ask hard questions. Excellent. Now, does anybody else have one I have one thing to? say about this. I. would ask them, tell me a customer where they've used it for a year and the product got better. If it has not gotten better, that means it's not learning over time and ask for explicit examples. Because a lot of people will say their product improves over time, but it does not. That's think. that's. really important because everyone needs to be considering what no matter what product you use, like, everyone needs the self learning revenue system. That's, like, the only way that we can all create edges for our own products. Distribution is the last edge within that self learning, revenue, is the kind of, like, in my opinion, one of the last edges that we can have versus just raw intelligence. So I do think that's really important for any product that you use because that will create a competitive advantage for you versus other sales teams. They're using products or systems that don't improve over time specific to your business. Yeah. I love that. So, you know, now, you know, going to agents in general, what does an agent need to access before you trust it to act on a live deal? Not just, like, draft, but but act. What inputs? I It should I think it should have kinda as much access as a a rep on a normal day would have. And so you would want it to be hooked up, you know, to your to your email, to the to, you know, all all the CRM data, potentially even to support tickets if there's, you know, if there if there's a live POC, things like this. I think that's kind of one one way to look at it, and I would I would say kind of make sure that you you add as much data into it as as possible. AI is kind of actually quite good at making sense of, of data. You can you can you can throw more data than you would throw at your normal human, and it's gonna make sense out of it. But then at the same time, I the fact that it, you know, it might not have access to this one thing, the other thing. I would just not consider that a blocker for it to act. But then just figure out, okay, in which situation can it act? What what kind of deals can it act on my behalf and things like this? And then be open and start experimenting because you'll see that you might you'll actually get a ROI even if if the performance is not a 100%. Yeah. I think something that's worked very well. So we had a deal assistant that we built out, a an agent. But, something we put inside of it would be basically all of the frameworks in which we want the sellers to sell. I think that one actually was very helpful. Like, it was pretty basic, but all of the ways that we want sellers to sell because I think it's a style, that it will always answer in context to, like is I think about scale, obviously. I think I've talked about that a lot of time on this conversation. And I think when you are not talking about five to seven reps anymore, and the minute you have double digits, you have to be thinking about, like, what's actually the best answer, and, like, how does that align with, you know, like, what works and what doesn't. Because when you have more people, there's more risk that something is a unproductive behavior. So I'd say putting in all the context libraries of, like, how you want your reps to sell, frankly, can get you better answers and responses to deals than if they wrote them out themselves. Because, again, the human hand off, they probably didn't absorb that in the first place. So if you just give your deal agents, you know, like, how you want them to sell, like, what does a good good examples of outreach from winning deals so that to the self learning point, it can do an imp pretty impeccable job at doing all of the, like, parts between steps. I would say, funny. I use, a sorry. Go for it. Please. Please was. gonna say I use I I I use a product that I'm fairly certain some people at Clay use, Town, and it just has my, email context automatically. And I remember it beating me out in responding both in time and depth, like, really fast to emails. I I had this 42 email chain, and, the person I was talking to said, okay. What do we agree to? And I'm like, fuck. I gotta pardon my French, everybody. I gotta go through all of these emails. And I saw an email drafted automatically that said point a, point b, point b. And. I'm like, do I trust it to send that? And I'm like, I don't have time at the moment. I'm gonna just hit send and pray. And I gotta thank you so fast, like, for how fast I was, and it was spot on and everything. I'm like, I probably should have done a little more more check-in before I trusted it, but it was interesting. Anyway, Tyler. So I believe there is no trust with AI. And so what I. mean by that is that, like and maybe that was, like, a this is a take. But, like, what I mean by that is, like, if you think about why coding has succeeded in AI is because there's a way to verify the work, because you can test and run the code and confirm it works. And the AI can iterate because there's always an answer. So I actually think, like, naturally AI is nondeterministic, of course. So if you want to trust it for especially even a live deal, which is obviously the most important thing, for you to close, you know, your like, hit your number. Like, it needs to have a right answer. Therefore, AI should only be really used in automatic fashion or Thomas fashion when it can check its work. So, like, one example is, like, for example, for Apollo, we have, like, sequence optimization, which is not necessarily a live deal. But when you're trying to, let's say, book meetings, it can basically look at what are the opens, clicks, replies, etcetera, and then use that information, and then it's optimizing for apply rate. So as a metric to keep optimizing against, so it can check its work. So each time it runs an experiment, it improves over time against a metric. If you don't have a way to check the work, it's very difficult to have it be something you can use without even looking at it. And we've seen that other products we have, for example, like messaging. Like, no one's ever gonna I mean, maybe some people will let it go. Most people we know are not willing to let that go. So that's my my opinion there. Well, this is why the inbox right now is so powerful because if you have a rep that has been there for years and you did give it access, it does have something to check against because there's all the deals that an interaction. And then if it can also see the outcome. of the same deals in those basically, look at the emails that were sent, triangulate between those people and what deal outcome was, it can it does have that closed loop. Because it says, Is, that kind know, in of follow-up you mean the connect? inbox in HubSpot or email? Like, you mean your Gmail, or how does that work? No. Your actual Gmail. Because, like, think, okay. about it. If I've been here two years and I've been selling and I've been emailing people back and doing follow-up emails, and then it can give me what calls were had, what deals closed, and what did I say to those, Yeah. it can look at how similar did was their answer to the outcome. So that's why I'm saying when you're looking at the whole deal, it's a little different, but I agree on the prospecting side. That's the best way to have a closed loop. I'm I'm curious, with your orgs, like, some of the rules you have, like, the line between what runs autonomously and what waits for a human. And and, like, what moves that line? Like, do do you have some, like, schools of thought around that, like, for your organizations? Definitely. We're for for outbound and the learning stuff, like, we're, we don't wanna automate it yet because we want more learning to happen from the reps having, like, a human in the loop, looking at the messages and and so on. So, yeah, you don't you don't necessarily wanna go there, like, too quick. It's like full automation. I might I, might I might disagree have I got the that one, I think. oh, Adam. I'm Go might. I might I'm a dis all. yeah. Well, I I I was gonna say I mean, even even Jared's example with, you know, that that long chain and and email. I think if you didn't have that answer prepared, you would probably you wouldn't have sent that email that day. Maybe you would have waited, you know, to the next day. It would have taken me forty five minutes. Yeah. Yeah. And the whole the end of the whole chain would just get delayed. And I feel like, we should just, I think, embrace it a bit more and, be a bit less risk averse and just, you know, actions that don't don't happen, you know, it's that's, you know, that it it carries a cost as well, I think, if you don't do some actions. And so if you're, you know, if if AI can do those actions that you wouldn't you wouldn't do yourself and it does them at 98%, it's probably better than you doing them at 6PM or 7PM. So that's kind of my philosophy. And so I'm, I'm just trying to push my team to do more and more. I'm not saying, you know, we're gonna automate the whole, you you know, customer journey and everything, but I feel like we should be always thinking, you know, what do I delegate, and what do I delegate to AI, and pushing ourselves to do it a bit more. I would just argue that, like, was that? is, like, earned. Oh, yeah. Like, so, like, I think, like, for example, what what Alex is bringing up with, like, reps reviewing the output before it gets sent out, that one part of that course is, like, QA, but I think the second part is, like, that feedback that's being hopefully entered in for, like, before, like, the. email sent out is a way for us to basically align the AI to what Clay wants to do for outbound for the specific rep. So I think, like, assuming that someone from day one is just gonna hop into, like, anything into a product, whatever, and start sending, you know, AI emails. I think they'll quickly learn that, like, everyone is so opinionated about what they write. Like, it's crazy. Every single person has different opinions. So, like, it has to be very specific to you, and that's not something you can usually do on day one. Maybe that'll change. Maybe it'll be some model that comes out that's like, oh my god. This is, like, you know, cold emails finally actually works, so they hide. And that would be cool. But I think in general, like, where we're at today is, like, for things that are nondeterministic that have no right answer, it's a taste call. You still need that loop at least enough times to confirm that it's something that you like. And then from there, of course, automate to. to. I'm not. I'm not I agree. Yeah. Yeah. Yeah. Hey. I, have. I think email is the easiest thing, if I'm being honest. So I've we found like, there's there's so many ways that you can have constraints over it to make sure. that it can not have a human in the loop, but not go off the rails. I'd say, like, we're pretty bullish on using email in lots of places, and not only for outbound, but for also for, like, using it for, like, you know, like, more personalized inbound for conversion. That's been a massive lever on the prospecting side. And not, like, just, like, for example, QL follow-up. But, like so, like, Clay is used actually for that, and that is given massive ROIs here. Well, last year, actually, and into this year. But, like, that was the unlock last year. So, Yeah. like, there's places that are not just outbounding for this that you can a 100% use it to elevate and per like, and, differentiate. I'd say, like, I think the original question was, like, where are you more cautious? I think it's it's all about proving the use case out, one. But two, I think it depends. I hate to keep going back to it depends, but I think in general, like, I'm the big company here on this call, from, like, a headcount perspective. And, also, it depends on your customer base. I don't, like, I don't know Yeah. how much do I pay to the attention to LinkedIn over the summer, but, like, you know, we had a a little bit of a a snap through this summer on data and how it made our customers feel. And, I think it depends on what kind of customer base you have. If you are, like, a company that's just coming up right now and for example, there someone mentioned a call a company that, blocks the call recording. Okay. So you now know that every person that buys from you or comes on your website cares about not being recorded. Okay. So do you think certain things that they might experience for your brand going forward that are AI forward, do you think they'll react well to that? What is what else goes along with that persona? So I think you actually need. to know who you're selling to and know the risk Yeah. that your customer is okay with taking. If I'm anthropic, like, my yeah. I'm I I get there's certain things they can and can't do too. So I really think the answer is, it's a brand conversation, to be honest. And your this is where your go to market people need to be beyond just people on keys. They need to be thinking about your strategy for, how does your brand wanna be perceived? Because it's very easy to become the next viral LinkedIn, Yep. story because you did something and it's a bad look. And now you who cares about whatever little AI gains you got because you have a bigger problem on your hands? You have a perception problem. Which perception. is that thing right now? So that's what I will say. It doesn't mean to not do it. It just has to be in line with the strategy that your company has and who your ICP actually is and where they are in the process of this, like, of this entire like, how do they feel about automating things? Like and are they coming to you because they want more or less of that? Are they just getting their feet wet? You kinda have to match them in certain public facing ways. It doesn't mean behind closed doors, you can't lose more, but you need to match them when it comes down to the interactions they have with you. That's how I feel. Now, I I wanna be conscious of time. I'm curious, from you all, and and maybe we could start, with Adam. Like, what's one thing you rolled out that held up better than expected and one you would not do again? Like, let's let's hear what failed first. Like, let's get into it. One thing that I wouldn't do again. Now I mean, we discussed one thing where, you know, at one point, I was it's a bit of a boring one, but, kind of just giving you all the context that you need that we have access to. And I think that's just, it was just a bit too much, and people just don't pay attention to that. I think that was maybe one one big learning, kind of also building the systems internally for us. You really just want to have the smallest amount of information that you surface to your reps, that they need, for the conversation. One thing that we rolled out I mean, that that I rolled out personally, I think computer control and just, like, browser control, it's, it feels magical with AI. And so I'm I'm an I'm an engineer by background, and so I do have a Mac mini that I control from my phone. And it does things like browse web for me and does some of these small things for me, and it's just, I feel like that's that's the power of AI that feels really, really magical. And so I I rely on that a lot now. Tyler, what about you? So sorry. Is the question, like, something we rolled out with, AI? like, what's what yeah. What that that that held up better than you expected, and and another thing that you wouldn't do again and start with, you know, I got the. failure. Okay. That's easy then for me. One is I would've so Apollo was, like, in the original release for MCP, but we didn't do anything with it for, like, a year and a half. Huge regret. I think that would be one is, like, something that, like, I would have done sooner is, like, admit that, like, building four agents is honestly better than building AI for humans. Because it's just, like, their usage patterns are much more predictable as well as, like, much higher volume. So I think that's, like, one thing that, like, no regrets, like, ship headlessly, everyone. Like, there's no regrets there. I think in terms of one thing I probably wouldn't do again, is this like, is, like sorry. Going back to the AI messaging front. We spent a lot of iterations on AI messaging. I don't know if we, like, we ever really got to where we wanted to get to with the data that we have inside of Apollo, which is, you know, very limited versus something like a CRM or, you know, you know, other integrations that the legit products have. So I probably wouldn't have spent as many iterations on, like, more of a taste call workflow. It would have focused more on things that, like, have a metric to optimize for. But, guys, it's simply I feel like it's always a losing game. There's always, like, an nth thing that you need to do to improve it. And we kinda seen that play out with, like, different product lines at Apollo. Yeah. What about you, Alex? Yeah. One thing that surprised me was, you know, our reps were originally, you know, creating a lot of assets, and, they looked off from the brand. And then the design team put our brand system into, or integrated into, like, Claude. And, it was surprising how good that got in terms of the, how on brand it was. However, then we kinda went wild, and then the brand team and design team were like, wait a second. There's actually small things that are off here. So let's, like, reel this in a little bit, and, let's have some, like, checks and balances. So, like, before doing something or before putting out something, like, big, you know, check with the design team, to make sure it's, like, not, you know, not off brand. And then we do have some certain rules. Like, you can you can never create, like, claymation type style, like, images. So there's some, like, very specific things. But, yeah, that's something, like, we rolled out. Did better than we better than expected, but also something we had to, like, pull back a little on, because we almost, like, gave too much too much freedom. But it worked really well. Yeah. There's new claymation characters that you didn't know you approved out of nowhere. Yeah. That could be. a problem. Yeah. Mintus. Oh, tough one. Okay. This is gonna be super basic, but probably one of the biggest, things that stressed me out in the last few years was rolling out scores to sellers. I actually have a hard and fast rule that I'll never do it again, in general. So, like like, TAM scores, all of that. As far as something that, you know, went way better than expected, like, AI summaries for, like, how to prospect reaching out, summarizing, like, the next best thing. People, actually really liked it. They actually don't like, they want those types of things. we we're we're at time. This has been a tremendously engaging conversation, and and thank you everybody for coming through. Thank thank to the incredible panelists, for sharing, like, what you're doing and what you're learning. Thank you. thank you to Airspeed for being a tremendous partner, and an incredible product building in this new AIH. Thank you. This has been awesome. Thank you, everybody. Thanks, everyone. Bye bye. bye.