(54) Inside Our AI Sales Stack - Kyle Norton | CRO @ Owner.com - YouTube https://www.youtube.com/watch?v=xHN3WmryKrQ

Transcript: (00:00) Welcome to the Revenue Leadership Podcast. And this is a special episode because this was a conversation >> [music] >> that happened behind closed doors at the Clay CRO Summit. >> [music] >> And to the best of my knowledge, this is the first time it's being shared with a broader audience. It's also been unusual because it's actually Kyle [music] Norton, who's the host of this show and the CRO at owner. (00:23) com, who's getting interviewed by Clay's head of GTM engineering, Everett Berry, who has some awesome questions for Kyle. They get into how Kyle's team is using AI tooling to generate more pipeline. They also get into owner.com's approach to AI experimentation, right down to the org structure and how head count is distributed. But also where they're intentionally focused on human involvement in order to get the best results. (00:47) If you've ever wondered what CROs are talking about behind closed doors, this is literally a recording of that from an exclusive CRO-only event. Enjoy. Momentum is a powerful tool for turning sales and customer conversations into go-to-market intelligence. Using GenAI, it extracts, analyzes, and automates customer intelligence across your GTM org. (01:10) The best part, it's not a new platform for your team to adopt. It integrates seamlessly with your existing stack, or it can straight-up replace your conversational intelligence tool. At Owner, we ripped out Gong and went all-in on Momentum. It writes back to our Salesforce directly, captures forecast and churn risk, auto-fills CRM fields, shares product signals, and tracks sentiment. (01:30) Companies like Cursor, Zscaler, Ramp, and Eleven Labs use it every day. Check it out with a free trial at momentum.io. Okay, next up we have uh part three of Kyle and I's podcast series on GTM engineering. For those of you that don't know Kyle, he is the CRO of owner.com, also the host of the Revenue Leadership Podcast. (01:52) And I consider you to be one of the sort of foremost experts in like AI-driven SMB sales. And so I wonder if you could talk us through, maybe actually what we were just talking about in the small group discussion, about what you're doing with your SDRs uh that is kind of resulting in the ability to sell to what I consider to be some of the toughest customers in the world, which are basically your mom-and-pop restaurants and so on. (02:13) Similar to Stevie, we pretty quickly figured out how much time is being wasted uh doing non-revenue-generating activities. We have this like slogan, RGAs over everything. And [clears throat] so I had first done a time-series study in like 10, 12 years ago now uh where I started to understand how many like bad leads my reps were calling, how much time they were spending on stuff. (02:38) And I had like super rudimentary solutions to that back in the day, like just Salesforce reports with like trigger words that would go into groups that I would like go screenshot and say, "Hey, don't call these." And so we've we've really applied most of our AI efforts to pipeline generation. Um it's my like uh opinion that if you can just generate way more pipeline, everything else gets better, like and it's much easier to generate pipeline than it is to move conversion rates. (03:04) And so we've largely had years of effort going after pipe gen. And so from top of the top of the funnel, we have a series of machine learning scores that power how big is this deal going to be, what's our e-win, estimated win rate. We've got this thing called e-connect now, how what is our estimation of whether or not that person will actually pick up the phone call, which led to like a 2. (03:28) 3x uh lift in decision-maker connects. And so from scoring to enrichment to AI pre-call research, just delivering on a platter what the sales rep should say on the cold call. We have a a pretty rigid cold call structure. >> And can you actually share what that structure is? I think it's quite good. >> It's Bukrios, the Greek goddess of cold calling, so we joke. (03:52) And so it's pattern interrupt, credibility, relevance, intrigue, offer of value steps. So some of those components have um some degree of personalization. So the pattern interrupt is what we call a local name drop. So basically, who we've taken every single prospect that we're going to call, we take their zip code, we find the customer that is closest to them, and it gets enriched into a Salesforce field. (04:17) And then in their Salesloft experience, we've we found a way to pump in like our own little window in the Salesloft experience that is AI PCR, AI AI pre-call research. >> I mean, Salesloft not easy, by the way. Yeah, like I would uh we're thinking about ripping out Salesloft and building natively on top of Salesforce. Um it seems like a little redundant now, but we've got this window that we've like injected into the Salesloft experience. (04:41) So the pattern interrupt is like, "Hey, Everett, it's Kyle Norton from Owner. Hey, wanted to give you a call because I actually work with Stevie's Pizza over on Fifth Street. Do you know Do you know Stevie?" And they don't know if it's a sales call yet, they just know that they're like this person they they very likely know, these communities are small, they probably know who that person is or them personally. (05:00) "Oh, we work with them because we work with them to to help them grow their online ordering, and I noticed" and that this is the second thing that's personalized out of the AI PCR, the I noticed is basically we hoover in everything from their um digital presence, and then this uh workflow ranks what is the thing that's most broken about their their experience today. (05:21) It's like, "I noticed that you use off-domain online ordering, and has whoever their current vendor is has probably never explained like what that does search rank? No, okay, cool." So like this was the same as Stevie, and then then that's basically the cold call structure. And we were creating intrigue by like this mysterious, "Do you know what that damage that does?" So that that is the structure of this AI pre-call research. (05:45) And then we've got, you know, like follow-up workflows and sell betweens, so like to to in between the call and the meeting happening, some automations to like send these videos, do this thing, AI cold email, like we're basically moving all emailing away from the BDR and the SDR, that's all done centrally, and then the BDRs are just told to call what leads based on the orchestration that we're doing centrally. (06:10) So my BDRs make like 150 to 200 calls a day now, and they talk to 20 to 30 decision-makers. And that is uh like probably 5x what what it was when we first started, where they were just like pressing the button. >> I was just going to say, you know, you didn't start with that level of productivity, which is which is an insane level of productivity. (06:31) What was the experimentation framework that that you ran to kind of arrive at this? Um early on there wasn't like much of a framework. We was We was just like me and the VP of BizOps and data like cooking up weird stuff, and then we're like, "All right, like let's give them this list today. Let's do this thing." And we would just YOLO it to the whole entire team and see what happened. (06:51) Now we have a more rigorous, but like still a work in progress, approach where we work as a team, so like my VP of RevOps, me, the VP of data, and our applied AI leader like brainstorming and basically scheming like, "All right, like what are the thing Where Where is the most uh lift that we can find in the stack?" >> And this is like a weekly meeting or a No, like once a month, maybe. (07:17) We know we like we know the next six things that we want to do, and and sometimes new things come up, and we're trying to think of a way to like actually bring more ideas from the team up cuz we're very centralized right now. And so now uh YG, the applied AI lead, will have his idea, he'll build a prototype for it, he'll fly He's in New York, and the sales team's in Toronto. (07:37) He'll fly in for like three days and sit beside the team with So like we'll pick two or three reps with one manager who this is their only non-hit-the-number project at a time. You're You're just doing the e-connect pilot. We're doing this like time-shift thing now, where we took four of the reps, and we have them start their day at 10:00 a.m. (07:57) instead of 8:00 a.m. because our data said that the best call window was actually 5:00 to 6:30, and like that gave us a big lift. And so so we do these things now these small batches. The data team pre-sets up what we're tracking, the manager is running the the team cadence to make sure that they're doing the thing, and then he writes the update at the end of every single day with the data that the team has given him. (08:24) And then if it's a winner and their stat sig, then we'll roll it out like then it goes it's an enablement ticket, the enablement takes it, and like we figure out how to roll it out more broadly. That's how that machine sort of hums now. Okay, and YG, your applied AI lead, reports into you. No, no, he applies into the VP of data and BizOps. (08:42) It was my head count. This was like inspired by Stevie, actually. She She was telling me about her applied AI team, I was like, "Man, I want that. That sounds awesome." Uh and so it's the first head count I put into the plan after we raised the Series C. And uh but we were hiring for the role, and it was clearly like this person was very technical, I wasn't going to be a particularly awesome manager for them to be a sounding board, and the VP of data is like quite technical and has enough business acumen to not make me worried about like, you (09:12) know, them getting too in the weeds. And so we I just gave him the head count with the with the rule that like he can only work on sales stuff for the first 12 months. Yeah. Uh and not forever, right? Yeah, yeah, so basically he's going to do sales uh ongoing until he builds a team under him, but we have another applied AI leader for growth marketing. (09:33) Or almost everybody in that team is data science and applied AI, or data engineer and applied AI. Every Everybody has to be massively AI-pilled to be in that org. Perfect. Okay, um so uh following Varun's uh lead with the heat here, I I've had some leaders recently, and I actually some folks in this room as well, tell me that they're actually planning on getting rid of their SMB sales team or their velocity sales team. (10:00) And I think OJ was kind of hinting at this. The idea is that basically highly transactional sales are very close to or it's actually possible to fully automate those. Considering you guys mostly do nothing but SMB sales, what what's kind of your reaction to that? If you can, you should. Uh we're already doing this. (10:19) Like there's a lot of things that we say, "Oh, AI is going to let us automate SMB." But like PLG is that. There there's massive companies that have a fully self-serve motion at the bottom of the market. So this just enables that experience to be better um and maybe creep farther up market. It's not as much a reality for us. (10:38) We sell to mom-and-pop restaurant owners who like didn't go to college, struggle to get on a Google Meet. Like the BDR has to help them get on the Google Meet cuz that's sort of their technical proficiency. So we're trying to to give an agentic experience for the bottom of the market with like uh mediocre success so far. Yeah. (10:59) The bar is just extremely high for how easy this experience needs to be for our customers to do it. I think uh many of you who sell to technical audiences, like way easier. They would rather talk to a robot than a human anyways. So it's it's more uh possible, but yeah, I I think it's certainly coming and and we're going to do a lot of it, but >> And are there like ACV thresholds that you think about for that where below a certain threshold it it has to be agentic fully or I ACV is one input to your unit economics. So (11:32) um yeah, you want really attractive LTV CAC and you want to have as much go-to-market efficiency as you can. But you know, if your ACV is 10K and you're closing at 80%, that's like way better than if you were at 12K closing at 40%. So the ACV is like one one input. But that's largely our lens. It's okay, if we can have uh like and I'm looking at LATAM reps uh to do the bottom of the market. (12:02) They're a quarter of the cost of our North American reps and we have people in my launch team who have enough sales like they're salesy enough that I think could do it. We tried to hire LATAM sales talent. It was really tough. There there's not very good uh sales craft there, unfortunately. Um so there's other options. Like I think people one-sided tangent here. (12:24) I think people too often it's a hammer looking for a nail and it's like what [clears throat] can I do with AI? And that's fundamentally sort of the wrong approach. It's like what are the most important problems for you to solve in your business? What are the different ways to do that? And some of those options might be with AI, but like you can have a deterministic flow for some of this stuff. (12:46) If it looks like this, then do these things. You don't need an agent at every single step of the way. Often times that leads to like sloppy outputs. Yeah. Um So it's more about like what's the business challenge? What are the ways that we can solve that that particular challenge? And like is AI the right solution? And that usually means like do I have to ingest a bunch of data and and make a reasoned decision that is not deterministic? Okay, like AI is awesome there. (13:14) Is it pattern recognition? Awesome. But but like jumping to it for everything is like not really what we're trying to do. Okay, perfect. So um one like thought that we're maybe vision that we kind of have is a lot of roles are collapsing and converging um especially within go-to-market. And um a version of this might be that basically the entire go-to-market team, at least at the IC level, converges into kind of two roles. (13:39) You have uh GTM engineers that build automations and scale the best ideas and that reps run direct human interventions like you just talked about and also feedback into the automations' understanding of the world. Is that kind of the end state feedback loop that we're that we're headed towards? I think it's a little reductive >> [snorts] >> because there's still like um so many other parts of go-to-market. (14:02) Like are your agents creating your core product positioning and creating campaigns? Like they're they're important parts of that flow and and ideating like all of us like we we ideate a lot with our with our LLMs. I think that those will be much bigger parts of a go-to-market org. We call it applied AI. Some people say GTME. (14:22) The the like tomato tomorrow, but I think you need way more of those people than you than most of us have today. Like we're trying to hire them as fast as we can. In terms of like oh, is there only a one other one other role? There's an equal argument to like have more specialization. Like the reason that we have this big SDR team when a lot of people are moving to a more full cycle is if you can strip if you can unbundle that job description into the discrete tasks, move all of those discrete tasks to whatever place, maybe (14:56) an agent or a workflow, or you centralize it, then like what is left? And and there is no way that any of my AEs could cold call the way my BDRs call now. Just like 200 dials to the face every day, like doing the grind, like just being so good at the knee-jerk and the the early interruptions. Um like you could argue that more specialization is where we'll where we'll end up because you strip all the other stuff out. (15:23) Then you have one person who is awesome at you know, demo close, whatever. And then a launch person, then see it like post-sales. I I see the job functions uh at a leadership level converging more than I do see the like the front-line role. Okay. So um in a post-AGI world, there's maybe a few things that are still protected. (15:48) One of these is personal brand. Um I think you of a lot of revenue leaders I I know have done an amazing job of building this uh for yourself and you're well-known on LinkedIn and and uh at different events. What uh maybe talk me through like how purposeful was that? Uh how much how many resources do you kind of dedicate to that? And then what are some of the outcomes that you're trying to drive for the business when it comes to, you know, having tons of LinkedIn followers or or or having a a strong point of view on where a lot of this is headed? (16:19) Semi-purposeful I think it was because of Naval Naval Ravikant talking about like compounding assets and like you know, everybody needs to be a brand and and sort of owning something that is uniquely you. It was like him and Seth Godin that originally like got me thinking I was like, "Man, yeah, I should really just like start to do this and like build it up over time. (16:39) " And it takes a long ass time to to like get the ball rolling. Um and the purpose was always recruiting to to start. Um I've typically done like, you know, big builds. Uh and so I wanted an easier way to attract talent. And so at at Owner, like we've never had outsourced recruiters. We've never really done outbound recruiting even in-house. (17:01) It's all been like Kyle writes a LinkedIn post and then 1,600 people apply and then we just wade through it. And that was awesome. Like we saved hundreds and hundreds of thousands of dollars because of it. And then like I got all these like other benefits. Founders reach out, want advice. (17:19) And I'm like it's an easy way to do advisory. And I get to do cool stuff like this and hang out with super smart people. And and that's like a great side benefit and more and more becomes the more important reason for me. But from a recruiting perspective, it's it's huge. I I don't monetize it for Owner cuz our customers aren't on LinkedIn at all, but Unlike Clay. (17:37) >> Yeah, exactly. And and the question is like, "Well, should I do that?" And like my my feedback is on is always like only do it if you're excited about it because you can solve recruiting through some other avenue. I like to write. I like to like get my ideas on paper. It helps me crystallize things. (17:56) I enjoy trading ideas. It's fun to have people like argue with me or email me back after a a newsletter goes out. And so I I enjoy it. So this is like a thing. Like going back to Naval, this uniquely works for me cuz it doesn't feel like work. I'm like excited to do it. And now that I've got all this all these AI workflows that help me, it's it's super easy to do. (18:16) But if if like that's not fun for you, then I like don't think it's a requirement. Maybe you could talk a little bit about those AI workflows or like what resources you have in the background behind this cuz you've got a newsletter, you've got a podcast, you've got a revenue org to run, and I think you actually are writing all these all these LinkedIn posts or maybe you have Not not much anymore. (18:36) So yeah, maybe you can tell us a little bit about the structure there. And and I do think um for folks in the room and we're actually focusing on this at Clay, leveraging your kind of exact presence to generate pipeline or do recruiting has been a major new unlock for us this year. (18:49) So so yeah, whatever you can share on the tooling or structure that you have behind this, I think would be great to hear. Yeah, so um the baseline is I have a bunch of writing skills. So I fed like I I I had 150 pages of my cuz I used to draft everything in one big Google Doc just to keep it all there cuz it gets lost in the LinkedIn abyss. (19:09) So I had this 150-page Google Doc of like all the LinkedIn posts I basically ever written. And I just dumped this into a model. I'm like, "Hey." So I it's a meta prompt. I was like, "Okay, write me a prompt that will be able to take my writing and turn it into like a skill." And then it wrote the prompt and then I took the prompt into another thing and gave it all the writing. (19:28) And it and it described in like excruciating detail how I write. The like tone of voice and the patterns and and all this stuff. And so I've got a newsletter writing skill, a LinkedIn writing skill. I've got an email one, which is hilarious. It's like, "Never say hello. Just like only write one line." Because I gave it a bunch of emails. (19:47) I was like, "Oh, I guess that is how I write." Um so that's sort of the baseline. And then uh I just have workflows for other stuff. So like I have a clay ta- like a big clay table for the podcast where when somebody's going to join me, I in my in my invite template in Superhuman. I've received these, yeah. >> Yeah. You go to the tally intake form and you fill in your information, your LinkedIn, and the topics that you think are interesting, and I ask for your contrarian views. (20:14) And so, that goes into the clay table, and then it does a bunch of research. And so, every column has like a big research prompt tell me everything that everybody has ever said online, basically. And then like then that then it goes into the next column, which is like break it down into some themes. (20:28) What is what is like different and unique about his views? And then it categorizes it into like how he how the guest thinks about leadership and then um systems building. And then it and then the last one is it pulls in your answers, all the research, and then usually I read it at that point, and then I'll pick the three topics that go into the last column. (20:50) And then then and when the last column is filled out, the very last one goes and then it writes me my like research my docket. And the docket is all the research, a summary, an introduction, a bunch of questions, and then basically my job is to spend 10 to 15 minutes just like editing the questions cuz usually as much as I freaking try, yeah, there's still like like really obvious stuff. (21:14) And I like really try with the podcast not to like talk about the things that are just like clichéd or obvious. I I really want to like dig into the dig in deep to things. Um and so then that just gives me the whole thing and that's what I go off with the podcast. And then then I get the transcript. Transcript goes into the newsletter writer. (21:32) I almost don't touch the newsletter output at all. >> Yeah. Cuz it's trained to not like write AI. Uh so, there's a Wikipedia page called like AI writing giveaways and there's like 15 things like the m dash, the the like it's not just this, it's that. Whatever sentence structure that is. And so I that whole I just copy pasted that into the into the writing style um file. (21:58) And so, it just kicks out the newsletter. I now just do a light edit and then it writes a LinkedIn post at the end. Uh so, it's pretty pretty seamless. >> Yeah. It's it is really interesting to hear about, you know, where the human intervention is in that process, you know, selecting the big three topics for example, cuz I think that actually is where I see a lot of people get things a little bit wrong is they they pick the wrong point in the flow to kind of Or no point. Right, or no point. (22:21) >> People are just like I put in the thing and then I like use the output. And so, even the newsletter one of the things I've done differently about the with the newsletter skill is um it will give me uh five different options for hooks because the hooks would be like sort of cringe. I'm like, oh, like I never for years I did this wrong. (22:43) And I was like I didn't actually do that wrong, but it's a cool like It and so, it gives me five different options and then I pick one and then it gives me 10 I wanted to write about the five key takeaways. And so, it'll give me 10 key takeaways stack ranked and then on sometimes I'll go like I don't like three use seven. And then it drafts. (23:01) So, now that I have the hook, the intro, the the key takeaways, now it basically one shots the output um because it's got the right like anchor theme from from the hook. And so, like that that having that intervention is really is is really helpful. Okay. A couple of minutes left here. Any questions for Kyle? Yeah, go ahead. Where do you find the time to do this? Like what is your personal operating system to dedicate focus time to those like learning all of these tools, but then also investing in all these other things you do outside of your core job? (23:30) I now have like a lot of workflows for anything that I do repeatedly. Like my the weekly update I write is like basically an AI workflow. And that's why I was asking OJ about the context thing. That's the thing that I'm still like figuring out. But the guest outreach, I just have templates for anything. (23:47) Guest outreach, the research, the writing. Like And so, you know, the I record the podcast for 90 minutes. I probably spend like another 30 to 45 on it every single week. So, it's like pretty seamless. The hard thing is like finding time to learn. And so, I just like, you know, after the kids go to sleep and my wife and I are sitting on the couch uh watching a show like I'm usually got an AirPod in and I'm like watching stuff on YouTube or on Twitter, to be honest. (24:16) And now that I've built up this base of knowledge, I find it's like much easier to like add the new thing to it and and uh you just like you know, I was lucky I just was so interested in it so early that I've like rode the wave. Uh and if you but if you were starting at a cold start now, yeah, you got to carve out a lot of time to like tinker and spend and basically waste like 20 hours building a vibe coded app that like never really works and it's sort of frustrating, but then you actually sort of start like building an intuition. I think with a lot of this (24:46) stuff you have to just build up like a I hate the word taste, but like you have to build like a taste and an intuition with it. Like what is the job that an AI is going to be like pretty good at good at? Like what's the thing that's going to like give you a slop slop output? What's the thing that's like overkill for it? And it's just like repetitions and and spending the time and finding like one simple thing to go use as the way you're going to learn that like skill. (25:13) So, like I okay, I wanted to learn I spent a bunch of time trying to learn make.com and like I don't use it at all anymore. And so, it's like sort of a waste, but it made me learn to like break things down. Now if I'm like using Claude code, I like know how to break things down into the little components and it'll spit out a PRD like it usually if I'm building something I just talk yap away, ask it for a PRD, then like in the user story I'll like break it I know to like break it down more discreetly. (25:44) You just have to like keep doing stuff and and develop the like feel for it. Yeah, I think I think that's kind of the lesson for GTM engineering teams as well is that it's sort of in the iteration and the and the experimentation that the taste making occurs. And then that's actually where you get some of the the GTM alpha as we say. (26:03) And I might be overkill. Like that's the thing. Like I I don't know. Uh I'm really interested in in all of it. I think you need to understand how the how these things work at a decent enough level that you can understand how to build your system, but like you probably don't need to be in Claude code terminal and like, you know, have a GitHub repo and like all this like weird other stuff. (26:28) Um that that's like not a prerequisite, but where the line is I like I'm not totally sure yet. Got time for one more question for Kyle if anybody has one here. Stevie, go ahead. >> If you could start over today, how would you build your stack and your team differently? This is going to seem like a really shameless plug because we've talked about this a couple times. (26:49) Like we built waterfall enrichment in late 2022 before Clay was a thing before I knew what it was. And we've talked a bunch like should we just rip it all out? And And so, I tell they reach out to me people reach out to me like, hey, where should I start? I'm like it's got to be first party and third party data. (27:04) Get those foundations right. And I'm like we do it like this, but I would just use Clay because right now the way we've done it because it's all like we in internally built it, only two people can really interact with these systems. And like RevOps doesn't really get good access. (27:23) My my BDR leader can't go in and do anything, but in Clay you could easily democratize access a lot more. I don't think down to the rep level. We have a different approach there, but um that would be one. Uh I pushed my managers really hard to learn a bunch of AI stuff and now I I'm like I don't actually need them to know that much. (27:42) They just need to know what problems they want to solve and then they applied AI team is going to solve them. So, I probably had them waste a bunch of cycles like I said everybody needs these two Andrej Karpathy videos. They're like six hours long. I was like, this is required watching by Monday. I wrote this big AI memo sort of like Toby did and the Duolingo guy did. (28:02) And that probably uh was overkill. It was my enthusiasm. So, like I wouldn't have pushed so hard there and taken them away from the craft that they need to be good at, which is like hiring, coaching, managing to the number. Like I I've like changed my my stance there. Actually, we we've seen that as well where you we used to actually test people in interviews for uh like knowing where to kind of click in Clay to some degree, knowing how to build Clay tables. (28:26) Of course now in Clay you can just chat with it and more or less build what you need and and it's more about having the sort of problems the right problems to to talk about. He- here's one good piece of design feedback or like if you're trying to learn any of this stuff, just ask Claude what to do. It's like, hey, I want to do this thing. (28:46) Like can you break it down? And so, when I'm building something, I ask it to explain it to me as I go. It's like, hey, I want to do this thing. Like I don't know how to do it. So, you guide me through the process of breaking down the problem set and coming up with a solution, but then I'm not technical, so explain what we're doing as we go. (29:04) And so, or else it'll be like, okay, now just like set up Vercel for the front end and your Supabase is going to be whatever. I'm like, whoa, hold on. Like why do I need a Supabase database? Why like what's what what is a front like explain like what is Vercel? Why like why can't you just do it? And And so, I've learned so much from like having the model walk me through the problems I'm trying to solve and why it's solving this problem in this way. (29:32) That arguably is where I learn more now than like like even YouTube. Cuz I'll like watch a little bit of a YouTube thing and I like get frustrated that I'm like [snorts] it's a bunch of like subscribe to my channel and later in this video I'm going to tell you all about and like skipping through I'm know, damn it. And but like you just ask the model, it's like, teach this to me. (29:50) Okay, amazing. Thank you so much, Kyle. >> [applause] [music] >> Thank you for listening to the Revenue Leadership podcast. If you enjoyed it, don't forget to subscribe and you can find a link in the show notes. And be sure to leave a five-star review, share it with your network, and please join me next Wednesday for another great conversation.