E20: Building Data-Driven Sales Processes in the AI Era with Jordan Crawford

And I'm very excited because today's guest is Jordan Crawford, who is the CEO, founder, and go-to-market engineer extraordinaire of Blueprint Go-To-Market. Prior to Blueprint, Jordan founded Scout, which was a postcard marketing company, and he advises a bunch of awesome startups, including Clay, who just raised a monster round. And Jordan is a Clay savant of sorts. Uh, Comscore and Tenor as well. Um, and his expertise is really in outbound sales strategy, data-driven go-to-market approaches, and, uh, trying to figure out how to leverage AI for business growth. And, and, uh, he's more colloquially known as my smartest AI friend. Like, this is the guy that I go to when I've got AI questions and I'm thinking about stuff. And, uh, Jordan has been extremely helpful to me as I've wrapped my head around not only how I use AI day to day for personal productivity, but in the business. And he's also one of the few people who has a louder laugh and is more frequently teased for his laugh than me. So I appreciate that. You know, you're like my— you're like my cannon fodder when we're out for dinner.

But, but we both are glad that we are not Seth Rogen. Like, he's the only person that that, that really sets the new high bar.

Hello and welcome to the Revenue Leadership Podcast by Topline. I'm your host, Kyle Norton, CRO at Owner.com, and every Wednesday we dive deep into the strategies and tactics being used to drive success by the best revenue leaders in the world. So join me as I sit down with revenue operators to discuss actionable frameworks that you can implement in your business today with no fluff, no sales pitches, and no platitudes.

This episode of the Revenue Leadership Podcast is brought to you by Superhuman. Superhuman is an AI-powered email app that helps sellers respond, collaborate, and close faster with features like inbox triage, with storage, automatic reminders, and collaborative email drafts, reps can filter out distractions to focus on what they do best: selling. Get the email experience that keeps deals moving. Try Superhuman free for 1 month at superhuman.com/pavilion.

Yeah, we have that going for us. So yeah, excited to have you on. Thanks for joining me.

Thanks for having me, Kyle Norton.

Um, okay, so we're gonna talk about 2 things today. One is where to start with GT, where to start with AI for your go-to-market team and a bunch of work that you're doing in advisory and consulting in this space and just giving people a landscape of how they should be thinking about this technology. Because I think many of us have AI-induced anxiety, you know, this, this relentless pressure if you're on Twitter to like not get left behind and seeing all the cool things. And so, so really talking through how people should think about prioritizing these investments and the strategy associated with it. Then we're going to also spend some time on personal AI applications. What are you using AI for day to day? How should people get started? You and I text about this all the time, and it's been very useful for me personally. So we'll get into these two very closely related topics. But we are one week out from the Deepseek earthquake. And so I want to start with a warm-up question. What's really going on with Deepseek? I want your take.

Well, okay, so I should say I'm not an engineer. Like, you know, you put a gun to my head and you ask me to train a large language model, and I'm not gonna be able to tell you how to do that. I have read a good amount about this, and my thought is that the place that these models are going, and specifically models like Deepseek, and actually, do you want to define it?

Just Maybe like not everybody is as into, uh, AI Twitter as you and I.

Sure. So basically the, the key, the thing that took trillions of dollars off the market or whatever was that what Claude and OpenAI, they're like, we need all these servers, we need H100s, the very best GPUs, and we need billions. Like, you know, the estimate is that, you know, hundreds of millions, if not a billion dollars plus has been spent on training these models and Deepseek. Had a hold my beer moment where they're like, $6 million. We trained a model using graphics cards that, you know, you might have laying around your house. And, uh, we didn't use any people to do this. There's usually people that are judging outputs. They use the models to do this. They also, they did some other kind of creative things, which is they said, here is the question, and they withheld the answer, and they gave the model rewards essentially for the right answers. And these are questions like 2 2 4, not what's the best strategy. Strategy to do blah, blah, blah. Right. And it's actually bad at that. And we can, I can show you some examples of that, but no reinforcement learning. Yeah. Reinforcement learning. Um, so, uh, and also they did sort of, I forget the exact name of this, but they, uh, uh, wait, I have it here. Mixture of experts. So when a question came in, they said, which expert, which expert, how should we route this question to which part of the model? Now, if they weren't training this on other models, they'd built their own base layer model, but this allows them to answer questions at a much cheaper rate. And I think that they could do this at time of inference too, which is just a cheaper way to do this than to do training. So I—

Time of inference being the time that the prompt is put to the model as opposed to when it is trained on the massive amounts of data.

Yeah. So I think that the thing about this is I pasted in a prompt. So I asked Claude Sonnet 3.5, which is my favorite model, O1 Pro, which is the $200 a month model. And the, uh, and Deepseek the same question. And I asked Claude to help me come up with a question. I ask a revenue question, which is, and actually Kyle, I'm curious to get your answer before we see if it's like Kyle AI, Deepseek, or, um, if you had complete control of Salesforce's entire marketing budget tomorrow and were mandated to drive 3x growth in 18 months, what would be the most radical yet executable shift you'd make?

Wow. Um, I would refactor their outbound BDR motion because this might be just like availability bias, but Lemkin tweeted about, uh, the prospecting email he got from a Slack rep. I was just like really poor and uninformed, and, and I felt really bad for that Slack rep, but probably shows that there's a structural issue in their prospecting engine because it's not very data-driven. It seems to be like old-school predictable revenue spray and pray. Totally. And I'm a big outbound believer. If I had to do it in 18 months, you'd probably go spend 3 to 6 months rebuilding that entire infrastructure and then seeing a monster improvement in BDR performance that could drive the growth.

But yeah, so I think that— so I'm going to give you the worst to the best answer according to Jordan AI, and I will tell you I think actually Deepseek came up with your answer or a version of your answer.

Oh, that's the worst one.

Are you telling me that is the worst one? But you actually— so to be fair, like, I'm the guy in the basement. I'm working on the latest AI models. Like, I haven't had to run teams like you had run teams. So, you know, so your answer may be the most practical, but I liked Claude's the best for sheer creativity. So Deepseek, basically the suggestion was switch to a PLG motion to reduce CAC, have a Salesforce Lite and free tier offering. Strings, do AI-driven nurturing and usage-based pricing experimentation, which I think they are doing this usage-based pricing.

Salesforce would be all usage-based and that's where everything's going.

Yeah. Yeah. So AI-driven ABM, AI, uh, CRM repositioning, uh, developer ecosystem growth. That's kind of the, uh, as the ChatGPT of CRM. So that was like kind of one way to, to, once it was just, oh, one pro had a very interesting idea. I didn't dislike it. It was like kind of crazy. Shift Salesforce from an enterprise-focused suite into a frictionless growth engine. And how it works is basically make Slack the default place, capture all the information in Slack, and then on the backend, translate that into Salesforce. Um, and so that solopreneurs, solopreneurs, solopreneurs can just use Slack and then Salesforce will just be being built in the background with all the structured data. I was like, ooh, that's like crazy clever. But I thought Sonnet was the most creative. What Sonnet said is, uh, what you should actually do is promote yourself to mid-sized companies, 500 to 2,000 employees, offer them complete digital transformation package, free implementation, dedicated consultants, AI specialists, and you take 2 to 3% of the business to do it. And I was like, I was like, that's like a VC model. And Salesforce like has the money to do that. Um, you know, only a certain amount of them are going to pop. You can undercut HubSpot. and, uh, and also you're dogfooding your own solution, which is always the good way. And so I love that because that forces Salesforce to move from a, uh, software service to path product at, or, uh, uh, what's the, what's the new thing? I forget the name of this thing where they invert it.

It's, um, oh, like outcomes as a service. There's something.

Yeah, yeah, yeah, yeah, yeah. That like, yeah, it's like, yeah, exactly. So it's like Salesforce is gonna have to dogfood their own stuff, use their own stuff, go from zero to one, and in doing so, they can implement and build all the AI to make that easier. And say, we're just going to give you operation outcomes. And so I think in terms of like sheer creativity, I was like, you know, and you might not— that would take a massive thing to be able to get a business to do something like that, I think. But this is why I am not— there's two reasons that I'm not worried about this deep-sea thing. The first is that Claude is ultra, ultra, ultra creative. It does better writing, it does better creative thinking. ChatGPT, DeepSeek, they're really good with these very structured tasks like write code. And there will be a lot of productivity that's on top of that. But the creativeness, I think, will be so much more important, especially if you're going to have these models talk to people. And the other thing is Jevons paradox, which is that as costs divide in half, usage more than doubles. And that means you think about electricity or the first Tesla electric car, right? How expensive that was. Uh, miles per gallon is a good example of this. So as the miles per gallon goes up, actually more gas is consumed. And so when AI gets to be really, really, really cheap, it's like you can use it for all sorts of things. It unlocks all sorts of use cases. Um, and so I think that we shouldn't be worried that, uh, NVIDIA stock will forever stay low. Like, this is a thing that's here to stay. There will always be room for these ultra creative models and applications. And I think that, that cost, driving costs down is actually going to be net good for everyone. And the large investments in AI are not overplaced.

Yeah.

And if we think about with these reasoning models, it's like 6 to X, 6 to 10X the token consumption than like a traditional ChatGPT-4 model. And so if everything's moving to reasoning and reasoning is the thing that is going to be the core driver, and reasoning could be built into basically every business workflow imaginable. The consumption, the inference consumption, like the consumption will flip from— it was always, it was all about compute, compute, compute was the thing that we were scaling on training. And now all of the scaling and compute needs to be at inference. Then we could see a 1,000x climb in compute consumption, which I agree. Like, I don't think this is an existential crisis for the frontier models or NVIDIA. Do you buy that they built it for $6 million?

They open source everything, so I don't have to buy it. I don't have to not buy it. It's being replicated right now. So we'll just know the— we will legitimately know the answer to this question in weeks of time, maybe not weeks, but, but that's the beauty of open sourcing everything is that we won't ever have to guess like legitimate US companies are going to be able to replicate this or not and say, well, you know, because they have a parent company that has a bunch of extra stuff. I know that you had mentioned a quote from, I think the, the Scale AI founder that's like, hey, yeah, they have, they got a bunch of stuff for export controls. They didn't want to talk about it.

I think that's the 50,000 secret H100s that flowed through Singapore was what everybody was saying.

Yeah, I mean, I mean, yeah, like there's probably some truth to that, but, but we, but the thing is we don't, we don't have to adjudicate that truth to know if what they did was possible. We will be able to answer that question. And so if people can just be patient while we try to replicate what they did, because that's the hardest. I mean, with, I, I have a social science background and that's there's a huge, uh, replication crisis in the social science world, which is like, we can't replicate anything. It's like, well, it turns out that all of our favorite studies— yeah, yeah, yeah, yeah. It's like, it's like, uh, every 5 years eggs are either the worst thing for you or the best thing for you. Um, but this is not like that. This is something that we will be able to say, actually, yes, you can do this, and you can run these models locally, and you can, um—

I mean, the output models clearly are very small. People are running them on a stack of Mac Minis or, or desktop Macs, and it's working. The question is getting to that model where way more GPUs used to produce that output. I, I am in the suspicious camp.

Yeah.

Like, you know, well, the other Stargate is announced and, you know, your hold my beer comment is, is great. You know, they, they released the first teaser of this on Christmas Day, and then this next announcement is right around this Stargate announcement. And it's, you know, if they were to have, you know, massaged, uh, what is in that academic paper through— because they're—

or shorted, right? It's a— I mean, the other thought is that they shorted Nvidia stock, they released this thing because they're a trading firm. Yeah, if you can move a market, that, that's like, oh, okay, now we have incentive, right? Like, why would they do that if they were just a research firm that weren't— that wasn't associated with some— so anytime, you know, you look into people's motives, right? There's a great quote: don't try to get a man to understand something that his paycheck depends on him not understanding.

That's so funny. I, I quote this all the time.

Yeah, that's, uh, Upton Sinclair, I believe, is the author of that.

Upton? Yeah, Upton Sinclair.

Yeah.

Um, okay, cool. So I got my, uh, fix of deep seek. DC conspiracy.

Yeah. Yeah.

So I want to start with where revenue leaders should think about AI in their business. And I want to go from sort of broad strokes, conceptual framework, and then narrow down and hopefully give people some concrete next steps to take and actually give them something they could take to their RevOps or DataOps team or a consultant and get pretty tactical. So let's just start with like state of the market and especially on, on like revenue generation. Like where are we today and what does AI change about where we're going over the next couple of years in terms of pipeline generation?

Well, what a loaded question. There's a bunch of pieces in there and I want to make sure that we're starting in the right place. Let me talk about kind of a little bit of the broken paradigm of the past where our frameworks fail. And then you can ask me additional questions to elaborate. We are still running on this model of AE to SDR ratio. We're using tools like Apollo and ZoomInfo, and we're using tools like Instantly and SmartLead now for email sequences. And we're trying to personalize our way out of the problem. We're trying to say, Kyle, you wear a shirt, I wear a shirt, we're basically the same person. And then you just launch into some deranged rant about you because you don't know who you're sending to. You don't have any understanding of the prospect. And when we do this at low scale, our solution is to increase personalization. Like you could say, Jordan, I'm so sorry that last Friday at 8 PM your grandmother died and then you went home and cried and then had a beer and then brushed your teeth. None of that has anything to do with my desire to buy or my propensity to buy anything. And so we need to first destroy that email.

Would get your attention, though, that's for sure.

Yeah, that's— well, that's people. People think that attention is intention, and that's totally wrong. And so you need to— I know you like that one. Yeah, that's—

yeah, I'm writing that one down.

Yeah, right. And writing for the soundbite, right? Or, or that the other— the thing is that the impressions on LinkedIn are different or are the same as the impression you make on someone. And these are different things. And so really what we need to do is, and the hard thing, Kyle, is you look at ChatGPT and you ask this question to me. It's like, am I the problem? Is anything can go in that box. It's not like Apollo where you're like, filter, filter, filter, export. Oh, thank God. It's like, I got my personas. I target restaurants between 50 and 100 employees in Texas. I just now just need to send them a message and my job is done. And so ChatGPT is a different beast, but we can make this problem much easier on ourself if we ask this question, which is, let's say that you as, uh, the famed CRO of owner.com had unlimited amounts of time to write one email. And I use email only because intent is vague. Uh, tools like Google, obviously people are searching for your solutions, so you have that for free. Facebook is using gigantic machine learning models to optimize for clicks. You can't control any of that stuff. You're outsourcing your targeting to the large companies that just want your money. But an email is actually something that you have to send to an individual person. You have to know why. It doesn't actually matter if that channel is not your best channel. It's just a good lens to look at the problem. So the question is, Kyle Norton, if you had unlimited amounts of time to write one message to the platonically perfect prospect, that you knew was good from public data, what would you look for and what would you write?

Uh, well, for us it's mostly about hypothesis of need. Like, specifically, what can we infer from their public data that tells us they have a problem that we can solve? And then we want to, uh, as quickly as possible highlight that problem, show that we can uniquely solve it and then build credibility in that email through, you know, a series of, of other pieces of information that we might know. They, they know a customer that we work with. They're related to somebody that we've done great things for. That would be the rough sketch.

Yeah. And that's like distance, right? You've mentioned something. It's like, I know about Taco Time because it is close to me. It's a well-known restaurant. It might be their pricing on apps. So, you know, if there's a delta between that price and versus the menu. So you know that there is some discrepancy there. Uh, you might have some understanding of their traffic. You might look at reviews on the rating sites to know how well, uh, they're doing on the various delivery sites. But you see how you began from a different place, and this is a little bit easier for you because you only target restaurants. Uh, this is harder for horizontal SaaS because it's like, who do you help? It's like, 2 teeth and a smile is all we need. And, and that's the, that's the challenge is that horizontal SaaS is selling a little bit of everything to a little bit of everyone. So they cannot develop these systems that you have developed because a CRO has to do what is best for the organization, but worse for his job. And that is pick, and that is pick a segment and say, we're not going to focus on forget 9 out of 10 of these segments. We're going to focus just on this one for the next 9 months. And that is probably super dangerous to their job because it's like, you picked wrong. Get that guy gone. But, and the likelihood you picked wrong is high. But otherwise, what's happening is that you will go an inch deep and a mile, an inch deep and a mile thick, right? Mile long. Inch deep, mile long. I got it right. Sorry, sorry, 5 centimeters or something for the Canadians out there. So, so that's the challenge is that really you don't know where to start with AI because you really don't know where to start with your customers because you're trying to take the sum of everyone that has ever bought from you and go outbound to them, go market to them. And the best thing you can do in that case is talk about you. Because you don't know about them because you haven't picked a segment to know enough about. And the only reason that you can do this at owner.com is because the whole organization is aligned around a very narrow ICP and persona. And you can say to the organization, we are not going to— I'm going to start day one and we're not going to generate revenue for 6 months because we need to build our data layer. And so this is the thing, the 2 questions that segment. For those that are so unfortunate as to work in a horizontal B2B SaaS, you know, sorry, uh, but if that's the case, pick a segment. Like, go look at your CRM, look at, you know, talk to the CSMs, uh, the, the, the customer success reps, uh, go look at the industries, the use cases, the personas, pull this information and say, this is the place where has the biggest possibility. We have high win rates and large dollar values. And then go say, great, now how do we— now that we have picked that, you can run with the second question, which is where do we get the data where all you have to do is re-describe the data? Here's all the reasons why I think this message is most relevant to you. And if you can do that, you can organize all your marketing, all your sales, everything around that list. Because as I say, the list is the message.

I was waiting for it.

Yeah, yeah, yeah, yeah, yeah. So So that means that everything is easier. You know what to say, but you, but you should just— this is a great thought exercise to just do manually. And if you can do it manually, then all you need to do— and I say this like with some air quotes or something— is then figure out how does AI build into each of those components. That's a data layer component. That's like a routing component. That's a scoring component. But you cannot know where do I start with AI without inventing the perfect message. The perfect message infers the perfect list. The perfect list infers the perfect data. The perfect data infers the systems that you need. And then you can say, how do I make AI make that all happen?

Okay, so I want to juxtapose the predictable revenue sales acceleration tool driven 2015 to 2023 period of time versus now. And it, and it feels like the macro changes to me are, uh, we are saturating every single market and every single buyer with way more messaging. We have less money to spend on BDRs and a bunch of things and just there's less intent in the market because we're post, post ZERP and budgets are tighter than, than ever. And, um, we're now in a world where, uh, the expectations to raise money and compete are higher than ever because smart companies are building with AI, they're marketing with AI, and, and it feels like the competitive landscape has taken a step forward. And the thing that worked previously was spam an entire market with a fairly generic message that us-centric because deliverability was decent and reply rates were decent. Like I remember in 2015 when I first got my hands on Salesloft, you could just like send what I'm sure was crap if I looked at it now. You just send like 1,000 emails out from, from your— a couple Salesloft accounts and you just had people being like, yeah, I'll take a meeting, I'll take a meeting. Like it was bonkers. Yeah, but that, that is so— that is so not the buyer these days that now the, the level of sophistication, uh, needed to get a buyer's intention and translate that into an intention to evaluate is, is different. So that seems to be me, like, the before and after. And now maybe we can deconstruct— I'd like to go through, like, more, more of that waterfall for people. So You started at the output and worked back. So let's start at the foundation now. So in your mind, step 1 is understand your ICP, like understand a smaller portion of your ICP and get narrow. Would that be job 1?

Yeah, really.

Job 1 is build the data to figure that out.

Well, it depends if you're vertical SaaS or you're not. So let me answer the question for vertical SaaS. Uh, let me ask a question for horizontal SaaS because that leads in the vertical SaaS. So horizontal SaaS, multiple use cases, multiple personas, multiple problems. So what you need to do is not try to go to market with your— we're in a— like, let's say you're like an AI workflow. We're like an AI workflow solution for sales, for marketing, for— and so you can't just target by persona and say, well, for sales, this is the message, and for marketing, this is the message, because your understanding is not deep enough. So you need to say, okay, great, look, we'll get to all those other segments, but let's pick the segment that we think has the most juice, where there's the most conversion, the competition, and that's a persona, that is a segment. Sometimes it's industry. It's usually a version of the product or some product feature, right? So in your case, it, it might be the website or whatever. And so once you say, great, this is the first thing we're going to experiment like this is—

but to get there, you need the data to understand your customer base and what like the healthiest mix of customers with the most ROI And yeah, yeah.

And so, and this is actually a really good thing you can load into ChatGPT, like literally export your— if you're, if you're the wild person from the Wild West and you don't care about like ChatGPT training on your CRM data, just upload that whole thing to ChatGPT and, and say, and this is the series of prompts you should run. You should say, I'm a CRO. I, I want to think through these problems. Here is my data structure. I'm trying to identify the right segment. In this sheet is the use case, is the segment, is the persona that bought, is the like total ACV, the time to close. We've got a lot of really great data. What I want you to do is help me think through how I would go about structuring this data to be able to pick a segment. So to have ChatGPT help you write that query and then say, now go do that, ChatGPT, and ChatGPT will like create a bunch of charts and stuff and you'll say, okay, great.

And use 0.1 for that because you want to use the reasoning capabilities.

No, it doesn't matter. Like ChatGPT-4.0 is fine. Actually, I don't know that 0.1 will even do a lot of— you probably actually can't use 0.1 because you have to have it write code. Yeah, there's like a bunch of limitations. GPT-4.0 is fine for this. And actually, this is a case where I would use GPT versus Claude. Claude can't execute code, but ChatGPT can both write and execute code. And it has this new canvas feature where it can just keep iterating on code, which is really helpful. So it doesn't have to redo it every time. So you basically can say, okay, well, I don't even know what questions to ask, but I say, here's my goal and here's the data input. You tell me what questions to ask, how I should structure these questions, what I should look at. It'll invent things. And then you have to be really judicious because they are bullshit machines and they are amazingly good at bullshitting. And so If you have to like squint, you should be like squinting when you read that thing and they're like, yeah, the first thing I would wanna know is blah, blah, blah. And you're like, mm, no, not that. You can use a tool like superwhisper.com, which is just, you can just talk. ChatGPT, if you're on the desktop app or the phone, there's like the little record button. So that's really amazing. Unfortunately, this is why I use Super Whisper. While it's writing, you can't use it. So sometimes you could also have it on your phone and the desktop and record while it's going. There's some little tricks like that, but really then what you're going to say is that, okay, great, I have some conviction here. You're then going to go sell your leadership and say, I think we should focus on this segment, but by the way, we are not generating pipeline. We are learning to run experiments quickly. Uh, and that is a total different thing, which is to say that it's going to take us some time. You know, you and I had this conversation is that, you know, when you started, I'm like, how fast was the system set up? Was it a week? Was it 3 weeks? And you're like, it was 6 months. And that's just with one segment, with one persona, with one, you know, with one use case here.

So with a badass data ops team.

Yeah. Working full time.

Working full time on it. Yeah, it was.

Yeah, you had people around you. And so, so like you came in in a really great environment and most companies don't have that. They don't have a go-to-market engineer. They have a RevOps person, which just is rearranging deckers, the Titanic sometimes just put this over here and here. It's like, well, why? And it's like, I don't know why, I was just told to do this. So, so really you kind of need someone that can connect strategy and connect the data. But once you understand that segment, you get everyone's alignment on it. You say, is this the first thing that we want to test? And then you go deep, you go pull in transcripts, you can dump a bunch of transcripts into chat. I would use Claude for this. Go dump your whole, like, Copy all your website, copy a bunch of content, um, have it detail your ICP and persona. So you want to warm up the chat always.

And you got videos on all this on your YouTube that we can link?

Yeah, yeah, I have an aggressive amount of videos on this stuff. Um, and it's like a lot. So basically what you want to do, you should think about these models in a bell chart sort of way, which is that you don't want to say, who's my ICP? No, what you want to do is you want to paste in all of your, the content of your website, a bunch of chats and say, great, now that you have all this context, define my ICP, define my persona. Now, based on this, where, and based on this data, where are the best places where I should be focusing? So instead of just asking it for an answer, you really need to warm up the chat with context. Who are you? What's the market conditions? What are other buyers, uh, saying? What are, like, we copied and pasted, I did a go-to-market cannonball where we try to launch a whole go-to-market motion in an hour for, for owner.com the other day. And we spent most of that time just dumping context in a cloud. We copied all your Capterra and G2 reviews in there. So people are like, oh, well, before it was this, and customers say really, really specific things. So that's kind of where you need to begin.

Are you being paid paid what you're worth? Pavilion's Compensation Report has become an industry standard for go-to-market leaders to benchmark their compensation, equity, and other employment terms. We're refreshing the report for 2025, and we need your input today. To get early access to the results, take the anonymous survey at joinpavilion.com/com— that's C-O-M-P— or click the link in the show notes.

So get the data, synthesize it. AI can assist there. And you would bring in both customer data and opportunity data because you want to understand propensity to buy, who's the most likely to spend money and who's most likely to grow and renew and be a successful customer. Is that correct?

Yeah. And you could ask too. The kind of nice thing is every time you think you're ready to go, ask ChatGPT or Claude what a better question is. So you're like, what data do you need for me to accomplish this goal? Here's my goal. I'm trying to do blah, blah, blah. What do you need from me? And it will tell you. And so just take a step back and you want to improve the model reasoning. You can do this, uh, the, uh, that I call the MPP, the Metaprompt Process, where you can say, here's da da da da. Help me.

This is my most, like, helpful thing that I've probably ever learned from you is this Metaprompt Process. So I would say if there's, like, for the listeners, if there's one thing you take away and get good at to make you better at everything else. This metaprompt process is the thing. So, all right, fire away.

Yeah, amazing. Well, I'm glad to hear that. So basically, the idea is you just want to give it all your context, tell it who you are, what you're trying to accomplish, and then say, do not answer this question under no circumstance. What I want you to do is to help me create a metaprompt that will, that then will answer this question. So you write the prompt, you're good at writing prompts, it understands. And You'll see that the prompt does so much better because it thinks, it slows down. And this is absolutely key to get the best out of the models because the models will invent things. And by thinking about the thinking, and this is exactly right, you, you, all you have to do is take all of the money OpenAI invested into O1 Pro. That's all it does. It just, it breaks things down into steps. And so what you're doing is you're essentially getting $200 for $20 a month with Claude and you're saying, you help me think through the best way I should ask you this question? And it will change the quality of your outputs dramatically.

So then we get the list. Now we have the right segment, persona, industry, the version of the product we want to focus on. Now it's acquiring data for people in that, in that ICP, new ICP.

Yeah, yeah. So you want to— this is where the go-to-market engineer comes in. And I think people are going to have a lot of a struggle integrating the go-to-market engineer into their workflow. Because it has to come from a segment and a strategy against the segment. And then you need to figure out what data is really going to move the needle for us. So let me give you a really, really tactical example for a horizontal SaaS company so you can understand how this works. We did a go-to-market cannonball for a company called Shovels. They are a permit API. And so who can use a permit API? Everybody and their mother, right? Like, who can't use a permit API, right? Like, yo, Kyle, like, you want to know what's done on your house? Use a permit. API. It's like, you want to know which restaurants are filing? You, you like, owner.com could probably use permits, right? Like, it's so crazy the amount of people can use permits. So, so what are you going to say? The, the default message that most people in the predictable revenue model are going to be like, hey, Kyle Norton is a company that does things and I've got permits and here's— it's going to be terrible. But let's think about a much, much, much, much, much, much narrow case. So what we did is we ask an inverted question. Where is this data really, really helpful? And you can actually dump— in this case, this is a data product. You could dump a bunch of this, this data into ChatGPT. But we honed in on roofing distributors of GAF. So GAF is the largest manufacturer of roofs in the country, and they have a distributor network, and that distributor network sells to roofing companies. And so these are the people that they'll like— they're going and knocking on doors at Joe's Roofing and Bob's Roofing and Larry's Roofing. Which they're all named that, and they are saying, hey, you should be using GAF in your roof installs, don't use my competitor. And here's why GAF is the best and it's the cheapest and whatever. So, so what we did is we built a system that queried permits in an area. We did two things. First, we scraped all the GAF distributors. And the nice thing about this is I can say you are a GAF distributor. You are on the GAF website. I don't have to take some roofing distributor list from some generic place. I already have one thing to say, but it gets better than that because I know your territory right now. I have the person, right? I know that this is a person that works at this GAF distributor and they're in Arizona, so I know they're in Phoenix, Arizona, so I can query my permits database. This is all programmatically and say, give me the biggest roof installers in Phoenix, Arizona. And it gets better than that because inside of the permit data, it actually says GAF, uh, and it gives the competitors too, and it gives an estimate. That means, Kyle, that I can send a message like this. Kyle, I don't know if you know this, but you are one of the top roofers in Phoenix, Arizona. From the permits that I looked at, it looks like you've installed 40,000 roofs in the last year, and those roofs, uh, are an average of $50K each. But I noticed only 5% of them are using GAF. And I think that you can do— like, here's the other things that we can benefit you. So now I can tell you exactly who's going to benefit from— like, if you focus on anyone in the Phoenix area, focus on Joel's Roofing because he's done the most amount of roofing. He is using some— yeah, the least amount of gaff, right? And so I can say to that person, and I can say to the distributor, yo, focus on Joel's Roofing. Joel's Roofing has done the most in Phoenix and he's only using GAF for 5%. Kyle, you already have his contact information because I know you sell in the area. If you want more insights like this, let me know. I've got 10. I've got the other top 9. I'll send them to you. Just let me know. That's an email that you would pay to receive as the GAF rep. But if I was selling permit data, I could never write a message like that.

Yeah, that makes a lot of sense because you're now down to such a small cross-section of the market that you can be so specific about some unique insight because the list is the message.

Yeah. And you could think about this too. I mean, in B2B, we do insane things that B2C would never do. I had a conversation when I ran my direct mail business with an AI camera company that sells these consumer cameras for like $20 a shop. And they were looking at direct mail and they said to me, Jordan, how much does this cost? What's conversion rates? And they said, we can't do this. I said, well, why not? Because if it costs more than $3 to acquire a customer, we'll just provide that as a discount on our website and increase sales that way. I'm like, oh my gosh. Like in B2B, you can throw money away. It can be ridiculously ineffective, like an offensive amount of money. To, uh, to be able to close a customer. And if you have that offensive amount of money, you can do way more interesting things. And so I, my sort of prediction, my thought about how to use AI is if you get really good at going narrow, narrow, narrow, narrow, narrow, way, way, way faster, that means building the data pipelines, knowing what data to build, extracting the customer insights, figuring out where the data is, and get really good at that motion, which is a brand new— every piece of that is a brand new motion for companies. But if you got good at that, shovels could dramatically increase their dollar value by doing roofers, HVAC, et cetera, and go that narrow. And it's like, even if you blew money on this, and even if at the end of the day, because they had to cold call and send a piece of direct mail, the, the CAC to conversion rate is going to be high enough that you will convert like bonkers.

And I know a company you've worked with that you ran a similar style of campaigns and they had to turn it off because they had too much pipeline for their reps. So it's certainly an approach. And this is largely the approach that we've taken as well. I think like, right, we just happened to get there before you and I spent started spending so much time talking about this stuff. And it seems like a— it seems to me like the place that people should be moving, you know, like it's all about the data. Like so much of our success is about making our motion so efficient because our data is, is so much different than what else is.

Yeah. And I think that the other thing to think about is these signal platforms are not the place to go. That—

okay, say more.

Yeah. So you're kind of being sold a bill of goods. With a lot of these signal platforms because what they're trying to do is they're trying to sell you in the predictable revenue model, which is like, oh, you know, the problem with Apollo and ZoomInfo is that they don't have the data that you need. You need signal data. You need to know when someone's hiring. But that is a continuation of the 'I saw that you' mindset. So those signals are 'I saw that you're hiring.' 'I saw that you raised money.' I saw that you woke up this morning, right? It's all I saw that you, but for your case, there's not a damn signal provider in the world that could sell you the data that you want, Kyle Norton. Because you, because everyone is a snowflake with their data needs. And that's a good thing. That means that you can say things that your competitors can't. And this is all, this is why you haven't seen a lot of these signal companies blow up is that the more people use them, the worse they get. Like, so their failure is designed into them because some of the people that use them are using them all in the same broken way. And for them to grow means for you to shrink. Hmm.

I'm trying to think through— this is a really interesting opinion. I'm trying to think through of like how that would not be true. Is there— are there intent signals or trigger signals that would tell you something that is leading to a hypothesis of need?

Sure, sure. I mean, look, there's nuance to all of this. So think about this. Think about the just raised thing, right? The way that most people—

let's choose a better one than that though. Like that is like clearly garbage.

But, uh, well, well, well, well, well, let me actually—

I want to steelman the case for single platforms.

Well, we could use a, uh, a just— well, well, actually, I do think that this is a good case because everyone will believe it sucks. But let me tell you a way in which the just-raise thing could not suck.

Okay.

So, so, so this is, this is a good example of when to use this thing and, and why the way in which you implement it matters a lot. So let's imagine, let's imagine a legal company, right? And one insight that I don't know if this is true actually, but, but I think something like this is true. Once you raise a certain amount of money, you have to do all these extra legal things. Yeah. Like you have all these extra problems that you never had before. And lawyers know. And so, so like, let me invent a Kyle Norton has just crossed $100 million in funding, and we have worked with plenty of startups that use Coulee Go. And when they use Coulee Go, it turns out that the, the, the, in the next 6 months, they spend $2 million just on legal fees doing X, Y, and Z. We have built a process just for companies that have just crossed over $100 million. To remove all of the money that you're going to pay lawyers to photocopy your equity docs or whatever. So that's like an example where you could buy an off-the-shelf signal, but you don't have an off-the-shelf insight. And that's the difference. And so—

What do you mean by that? You can buy an off-the-shelf signal, but it's not— Yeah, explain that.

Yeah, so an off-the-shelf signal is when someone just raised. So you can buy that from Harmonic, you can buy that from Apollo, you can buy that from Crunchbase, right? Like anyone can access that, but that insight is not an off-the-shelf insight. Knowing that there's some unique thing that happens to companies at some particular dollar amount, that means that, that you're not saying what, what most of these you just raised money messages look like is Kyle, you just raised money. I like money. Can I have your money? That's like, you know, that's the, the insight is like, Kyle, you must be growing. And as a growing company, You might need marketing services, and I'm a marketing services company that has— doing marketing things for growing companies like yours. We work with Salesforce, Facebook, and Meta, and we do— you know, it's like, that's what that always sounds like. But it's a different thing if you say, um, Kyle, after $100 million, Kyle, have you, uh, did— have you sent the $2 million to Cooley and Go yet for them to reincorporate your board docs after the $100 million? Because every founder that we've ever encountered has. And I don't mean to pick on them. I have no idea what they charge, but, but that's the kind of message that you could send if you had some unique counterintuitive insight. Almost everyone does this after their Series A and they spend so much money. And by the way, you just, maybe that money has just gone out the door and lawyers won't tell you, they won't tell you how much it's going to cost. And so that is an off-the-shelf signal with a non-off-the-shelf insight, but that rarely happens. It's much better actually to build. you know, unique data sets, because that, that insight that you have might actually be when they hire their first in-house counsel or something. So, uh, and the data becomes your moat if you match it to a counterintuitive insight, usually. Hmm.

Um, I want to ask one more question on this topic and then move on to some of the personal AI application stuff. I'm sure what many people are feeling is a sense of significant overwhelm. You're throwing so much at folks and I've spent so much time trying to untangle this stuff and blowing up your iMessage and trying to grapple with it. But where do people— so where do people start? So most people I talk to are not even like starting to scratch the surface with applying AI in their business. Like, where would you say is the first place if I'm like, okay, Jordan, I want to go down this path, but it sounds scary and intimidating. Is my first job hire a go-to-market engineer? Is my first job to try to like run this data this data process myself, like, where, what advice would you give them to, to begin down this path of getting to this place where they have hyper-targeted, hyper-targeted segments of their market with hyper-specific messaging and these unique insights? Like, what's the starting point?

I think that the best starting point here is really to start dumping context into Claude. And what you do is you copy and paste your website. You get as much context into that window as possible. 200,000 tokens. It's like 100— I don't know exactly how many pages. There's a lot of pages of data. Just pop it into Claude and you can ask, use the metaprompt process and say, what are basically, given all these customer reviews and transcripts, et cetera, tell me what are the public. And actually I should, I have this prompt. I will share this prompt as part of the process because they can use this. In the show notes. It's like a 2-page prompt basically. And I have spent an ungodly amount of time working with Owen Pro and Claude to improve this prompt. Uh, but the way that this works is it will tell you what data sources you should focus on. And actually I'm selling this like $10 thing, which will create a go-to-market job, uh, engineer description. I sent you one version of it, um, that will like say, here are the data sources, here's where the market's going, here's how to think about those data sources, But this is really the best place because once you know where the data exists, then you can, you can literally just have someone in the Philippines or in India or in South Africa go manually do this, prepare all of the data for the STRs, and then try this at low scale. No AI, zero AI. And so you first need to make the move from a non-data-driven world to a data-driven world. And then you then need to deconstruct. Okay, great. Now that I live in a data-driven world, I kind of know this is working. SDRs are sort of calling people that have the highest version of this problem because people are doing their research ahead of time. Then you can start to say, great, let's build a workflow. You can use a tool like Clay to do this, N8N. There's a lot of these workflow tools and say, all I want to do is get this piece of data from this website all the way to my CRM. Here, then I want to get this piece of data. And so you can sort of layer this on. You don't have to do it all at once and just chip away at this. But don't try to solve a pipeline problem, try to solve a data problem. And before you solve the data problem, know what data you should ask. And so revenue leaders need to get into the weeds, go sit with the SDRs and say, hey, if when you're researching folks, what are you looking for? What is the— and they're going to be like, I, I take this corner, I go here, I go around this corner, I go here, I log into this thing and it's this thing. If I see this thing, it changes everything. It's like, okay, great. Remove that process, have an outsourcer go do it, and then say, great, now we're going to save some time and move that process from an outsourcer to AI. How can we keep lowering and lowering the cost? But you have to start there. You have to know what to look for, uh, uh, first.

And should you get your RevOps leader to do that work? Your data biz ops? Like, who's going to go do that?

Well, I think the— I think probably what's going to happen is that the CRO needs to find the segment first. Like, you have to pick the segment first because I don't think most CROs can do the, the, the technical work you're talking about, though. Well, I mean, yeah, you can have your RevOps person export the CRM or do this analysis and say, hey, yo, yo, person, I want to know what's the best performing segment with the best like, tell me the product they bought, what is the title of the person that bought, or you can include the buying committee if you'd like, what is the average deal size, and, uh, what is the time to close? And then go get that information and say, great, this is the segment. And once you— and you can go chat with, with a couple of SCRs, say, what, what are your easiest to close sales? And they're going to tell you it's X, Y, or Z. So you have to do that work first, and then you need to say, okay, great, this is the place, this is the narrowest place that I'm going to begin. In. And once you know the narrowest place, then you can start to say, okay, great, how do I create a step-by-step workflow? And the SDRs can just record a Loom.

Yeah.

And then you can just hire an outsourcer to go say, every morning what I want you to do is like search Google Maps, put this here, put this here, get that ready for the, the, the, the SDR. You guys have a version of this that's just all programmatic, but that's just a longer investment. To do this, you have to make a 6 to 9 month commitment to build the systems. You have to do process changes for people. So I'd just create a little pod and see if you can get it to work. And that's going to be the seed for which you can then start to distribute a brand new way to do this.

Yeah, I also think that the thing that helped us go quicker on it is You got to have somebody good in data ops.

Yeah.

Or biz ops. Like, that'll be the thing. Or, I mean, like, hire you. Like, you'll go do it.

Yeah, sure.

Like, either, either somebody needs the external expertise to come in to do it on a consulting basis, like I know you do for a bunch of mutual, uh, friends and, uh, portfolio companies, or somebody has to be full-time dedicated. In my opinion, somebody has to be full-time dedicated to this project. I don't think revops, revenue leaders, CROs can do this and go through all the prompt engineering and whatever, whatever. I am—

But they do have to pick. But you didn't have to pick. You didn't have to pick because your company picked for you.

We did actually do more picking than you may think.

Oh, amazing.

Yeah.

Who didn't you pick? Can you share?

Like a bunch.

Yeah. So this is the thing.

There's always more than there's like 500,000 independent restaurants and we narrowed that down to like our top 30,000 based on all this stuff. We built machine learning models, like we did hardcore picking, but certainly easier in vertical SaaS than—

Yeah, yeah, that's what I meant is like the picking has another layer to it if you're horizontal SaaS.

That is somebody has to go do the data work and the infrastructure, the analysis, the research, the— especially because you're going to use AI to do a lot of this workflow, there needs to be somebody technical.

And then how did you, how did you know, how did you develop the heuristics to go from, uh, half a, half a million restaurants to 30,000 restaurants? Like, sure, the machine learning, whatever, but, and like, yeah, I guess there's a, a blind way to do this, but presumably if you looked at one of those 30,000, which of the 50,000 you as a CRO could describe the reasons why.

Yeah. Yeah, for— I mean, I did a very basic, discrete version of this like almost 10 years ago in my first tech role as a head of BDR, where I pulled a big, huge spreadsheet of every lead that had been touched over a 30-day period. And I directly knew— I talked to a bunch of AEs on like, who are your good ops? Yeah. And like built sort of a framework and then basically said it was just like a monster pivot table and like with a bunch of VLOOKUPs to tell me, okay, like of all the activity that's happening in my BDR team, what percentage of that are going into the titles, functions, and companies that my AEs tell me are good? That was like the easiest heuristic.

Yeah.

And so that, that could be a an exercise people run, I will warn you that the results are going to horrify you. Like, what you're going to learn is that you're producing a bunch of pipeline that your AEs don't actually need and you're just wasting like massive amounts of time into tier C accounts and crappy seniority leaders. Like, it's not great.

You're exactly right. And this is one of the frameworks that I invented to do exactly what you're talking about, because you do— this is a a thing of narrowing. And so you as a CRO, uh, this is the 5 C's of Truth. You can start with a customer and market context, right? You're going to have some understanding of that, and sure, you're going to chat with your AEs, etc., to be able to do some piece of that, and that's going to be, uh, super important. Then you're going to go to your CRM, and this is where you can say, great, this is sort of the customer and market context. Maybe you actually also chat with your AEs as part of this, and the CRM work can say, hey, go validate this hypothesis. I chat with the customers, these are most hungry customers. These are where my AEC are good fit. Go to the CRM, then you get to pick a segment and you can go to this. The, this is usually the CSM, sometimes CEO if you're a smaller company. This might also be AEs where you go validate the on-the-ground stuff and they say, yeah, that matches my intuition. You can summarize that all with Claude and ChatGPT and say, I think that these are the heuristics that make a much better restaurant for us than other restaurants. And then go back to some of those closed one customers and describe that reality to them and see if they agree with you, because then you're anchoring yourself back on the customer context. And so you're going to go to one of those, you're going to go to Terry's Taco Time and say, yo, Terry's Taco Time, like my understanding is you were doing these ordering things that your business had started to expand locations, that you were you were looking to add digital ordering, you had hired someone recently that, you know, was going to work on your loyalty stuff. And was this all true? And they're like, yeah, but this is not true and that's not true. And then you actually are anchoring to customer buying behavior of your best customers that is sort of rounded out by the organization. And but, but you have to narrow and then you have to map the narrowing to the data. And only then can you start thinking about how do I operationalize that data, etc. So you can certainly leverage your team to do this, but you are the one that's going to have to synthesize that information. You can get your team to collect it, but you have to make some choices and then you need to sell that externally, right? My guess is that when you did that narrowing exercise that you got your C-suite involved and say, hey, we're spending a bunch of time on accounts, we're going to because that feels like a loss. $500,000, right? That's what you sell to VCs. $500,000. Oh my gosh, they could sell anything so big.

Yeah.

Oh my God. It's like, it's like, have you ever eaten a meal? That's our customer. If you've ever eaten a meal at home, it's like, how many you eat 3 times a day? That's a 3x bigger market than I originally thought because you're having breakfast, lunch, and dinner. And so, so then you narrow that down. You say, actually, no, it's really only people that— and the VCs are going to be like, well, this business is totally— it's like, no, no, No, no, no, this is not a message to the VCs. This is about the customers and who actually can drastically benefit. And if you can't describe to them with public data why they can benefit, then you have no business talking to them at all. And that's the key difference. The narrowing is the description of the problem based on public data, and the description of the problem is not only why they close, but all of the people you shouldn't focus on that the whole organization is Great. There are 470,000 restaurants that we're going to spend no time on at all. Are we good with that? Yeah, we can come back to them, but these are the 30,000. And you know why? Because we went through a damn exercise to determine why. And that exercise, that internal exercise, is also what you say to them, right? These are all of the reasons why I think I did all of my homework on your business, and these are the reasons why I think you can benefit. Profit. And by the way, my guess is within that 30,000 that you have that segmented further, that you know the top 100, you know the next 1,000. And my guess is that you're— Kyle, you're going to do something different for the top 100 than the top 1,000 than a top 10,000.

Yeah, um, really interesting. Is this part of the whole framework in your Cleygent course, or is that more technical?

Yeah, that's a good question. So I have, uh, I have a Cleygent course which is one Godfather's worth of content. So 2 hours and 55 minutes of, of content. Uh, so it's me and Martin Scorsese. We're basically the same. I mean, we both created, um, you know, there's one Oscar between us.

A 3-hour masterpiece.

Yeah, yeah, yeah, yeah, exactly. So, uh, yeah, between, between the two of us, you know, there's, there's an Oscar. Uh, who's to say which one has it? Um, but that's going to teach you a lot of the logistics of building these AI agents, how to prompt AI, how to talk to the strategy piece in there.

Or like, can this—

so the The strategy piece is not in there. That's my next course. What I just showed you, this 5 Cs of Truth, that's who to target and what to say. I've been working on this since October, since like the last course launched, October.

And so I saw the slide, I was like, man, I went through, I went through your collision stuff. I was like, I don't remember this. This is good. Like, I like, I like, that's a great framework. Okay, cool.

Yeah. So there's a, there's a lot of that I have and I'm, I've, I'm creating a bunch of and I think that like, yeah, there's just like a bunch of frameworks. I can share more of the stuff, but that course is going to probably be $1,000 with all the prompts. Um, I'm— so I think I'm gonna have a $149 version, which is just like none of the prompts, and then a $1,000 version with all of the prompts. And I think there's about 60 prompts right now. So, um, there's just a lot of— like, I have prompts for this whole 5C framework. So, um, basically I have a whole interview guide. So how can you talk to your customers And how can you talk to customers that really don't want to reveal their secrets or you don't have a good relationship? So how do you warm them up? So, like, asking a controversial question, like, we've been noticing specific industry trend. Some leaders think common view, while other leaders think alternative view.

This is my favorite discovery framework because it's called— I call it the strut your stuff question because you frame up something that's commonly discussed in the industry and you're trying to get them to like strut their intellectual stuff and like, oh, some people say this, but like, what do you think? And everybody wants to like have a hot take.

Yeah, yeah, yeah.

The reason I'm asking is I think what especially my audience is going to struggle with is like the framework. I think CROs are going to have to get involved, make, make some important decisions, but probably not into the prompt flow and doing the work themselves. You know, they can outsource that to their nerds. Um, and but the, the question, and I think the thing that people are really interested in learning about is strategically, what, what is the, what are the things that I need to orchestrate? Do I need to go hire an AI, a GTM engineer today? Do I need my RevOps team to do this? Does this need to come from BizOps? And, and so it's, we're, we're giving some people some of that, but, uh, it's cool to see that you've got more coming there because I think that's the missing piece because then you can take your Claygent course and like people can go execute that, but unless they know the strategy behind it, I think it's not as useful. So, I want to skip forward to how— so, I think this level of depth is probably best placed to have an expert do it, either hire an outside expert like yourself or hire an expert full-time on the team and make a big investment in terms of making this like a massive business priority and have, have a data ops person run this project. But I do believe that revenue leaders have to build a fairly meaningful literacy.

Oh yeah.

With AI and tools and like understanding the capabilities, or else you won't be able to be imaginative enough. Like, I think I end up with a bunch of these ideas that I want us to pursue because I'm actually in the tools myself. I use it a ton day to day. So, but you know, like ChatGPT exploded, what was it like 2022 November, something like that. So we've had like 4, more than 4 years and like that month I like got a, I got a subscription. I've been playing around. I've got, I've built like Clay workflows with your help. But there's, I would say the majority of folks that are like just starting to scratch the surface and maybe use ChatGPT. To like use instead of Google, but like with the same search that they would dump into Google, they're dumping into ChatGPT, which is obviously not really the way you want to use it. So where do those people start? You got to like super simplify.

Should we do this live? Like I can share my screen and I can just like talk to Claude and you can give me an example of a problem that you, you might be thinking of is like you could put yourself in the shoes of a non-AI novice. Um, was that helpful or do you want me to like go over the frameworks here instead?

Like, where should they go to learn? Like, do they need to take the MIT AI starter course?

Is it YouTube?

Like, where, like, what, what, how should people pursue?

I, I am including, because the, the, the application layer is the thing that's moving the most, right? So tools like Bolt.new or Cursor, uh, can create, uh, full engineering shipped products for for you. And so that, and that kind of stuff is really always changing. And there's tools like Gamma that can create presentations for you. And so the application layer is the thing that's churning the most and is the hardest to keep up with it. The underlying models, uh, you know, we had an operator release, which is the, the model can go— I actually had ChatGPT tell ChatGPT to log into ChatGPT and enter prompts into ChatGPT, which is like a whole other layer. Um, but, uh, But that stuff changes a little bit less frequently, and it's easier to stay on top of. So I actually will link some YouTube channels that I watch to stay on top of stuff. But really, this is more an update to the way that you think, the way that you talk to technology, and not necessarily any tools themselves. Tooling is for tools is my kind of mantra here. I think that, yeah, I have a couple mantras like A, the process over prompts. So the process matters a lot over any individual prompt. Tooling is for tools and you are the guide, not the river. And so that's the key thing is that if you're like, where do I begin? You just you can do things like you can hit record and just talk about your day into Claude. Here's what I'm doing. Here's how I'm spending my time. Um, you know, yeah, generally if I think about it, I, I probably do like a lot of task management. I'm not really sure about blah, blah, blah, just literally stream of conscious into this thing. And it turns out that any given token doesn't really matter because these models are good enough. So sometimes I'll have fun and just say things to amuse myself as I'm talking to the prompt. But you really just need to give it that context, and language is the very best way to do that. And so instead of thinking like in the previous world, where it's like, I go to Google and I'm looking at, uh, Toyota Corolla price, like that's what I want to know, um, you can ask a more complex question, which is, um, I have a Huffy 1996 bike, 3 kids, a dog, um, and a wife. Uh, we're trying to get to the other side of of Toronto. We want to know if all this will fit into our car, and we're a little bit concerned about the dog having enough space to move around. What is your recommendation for solving this problem? Like, that's the type of thing that you can ask. And so you need to open your mind up to a much different way to talk to technology and just stream of conscience. So you can ask AI, what, what are ways in which AI can help me with my problems? But in many of the things that we're talking about, Kyle, it goes back to the customer. So it goes back to the problem. So the fact that there is some brand new tool that does some amazing thing at half the inference costs, like, what does that mean for my daily life? Well, you might not even know until you ask the models what it can do. And that's just a great place to start and just get used to talking to them. I mean, I, you know, I'm here in my studio and I embarrassingly have more conversations with AI than people in any given day. And just And just that exercise, like anything else, gets me better at it. And I've had hundreds, maybe I think I have talked with thousands of hours with AI. And I just thought of this metaprompt process recently, and I'm sure other people have figured it out, but it takes some time.

I think one of the interesting things as well that folks can do is have one— like, I have one thread of all my AI questions, one chat with like everything AI related. And whenever I have a question, I go back to that thing. And you— and, and that thread is sort of a tutor now. I'm like, yeah, you know, when, when this reasoning model stuff came out, I was like, what the heck's the difference between inference time compute and training? Like, explain this. Like, what, what's the total difference? Like, what, like What is even a token? And you can just have, especially with, this is the advantage of ChatGPT, is voice mode, in my opinion. Like, that's, it's crazy that Claude doesn't have it, but to be able to take your desire to learn a new thing to ChatGPT and just like make it your tutor. Yeah.

And be like, voice mode is, you're talking about the speak where it can speak back to you.

Exactly.

Okay. Because Claude can record, but only ChatGPT has the voice mode. I think it's because of Siri and there's they can teach you a language, which is so helpful to be able to talk live to it.

Yeah. And so one of the things that I'll do is have ChatGPT break down the thing I'm learning. My brain really works in like taxonomy. I need to understand like, what's the shape of this subject? What are the chapters? What are the sections in each chapter? And then let's go and like, you know, discover the content of each of those sections. Like my— it's just like the way my brain operates. But I feel like it's a helpful way to say, hey, I'm trying to learn about—

I'm—

yeah, the other thing I like about ChatGPT is, uh, it's memory. You can, you can put things into memory. And so I have a, like, I have a thread with just— it's just health stuff. And I'll go in there and I'll just like yammer away and tell it about my, like, health situation and, and, like, just like my general health environment. This is how much I weigh, this is what I eat, and whatever. And so when I now have a health question, it's all in this one thread and all this stuff is baked into its memory. So you can say, I am the chief revenue officer of a Series B startup based in Austin, Texas. We sell to X, we sell to Y, commit this all to memory, and it will— it will— you can see in your ChatGPT settings like all of the things that are in, in its memory. And then you can say like, I'm trying to learn about, uh, how these models function. Like, tell me about the part, tell me What are the parts of this problem space that I need to understand in my business? So do you need to know why a vector database is different than a normal database? Like, probably not, but like, potentially very valuable if you're trying to— if you are more involved with some of the architecture decisions. But then you can, you know, understand the difference between discrete problem solving and decision making versus probabilistic reasoning, which is obviously AI. And why are those things different? What's hallucinations? So why do hallucinations, why are they important as a concept in AI? Oh, because it's a bullshit. It makes AI a bullshit artist. I think these are the, if I was going to give people, this is the advice I give all the time. It's like, stop using Google and just take all of your questions to ChatGPT. Try to use it as like your smartest friend that you can ask questions to, and like force yourself to learn how to use AI through AI. And then you start to like— you develop a— in, in my experience, develop a bit more of like a pattern recognition.

Yes.

Now I, now I know. And then follow some smart people on LinkedIn like yourself and others especially on X, probably more cutting-edge frontier stuff on X than LinkedIn. And so if you're getting started, LinkedIn's a great place. A few people on YouTube, learn the basics of prompt engineering, bingo bango, you're like pretty ready to— I think you're pretty ready to start to use it more effectively.

Do you ever ask it what information? Because this is something I use, and I'm curious if you have a variant of this, which is is what information, what other information would you need to know to be able to form an opinion? So I'm always worried about it giving me an answer too soon. Like, this is a pattern. AI is like, hey, how do I— and it's like, how do you make a car? Here's exactly how you make a car, all the steps. You're like, whoa, whoa, whoa, I haven't even told you exactly what I— so I like, do you ever ask AI to slow down?

I do, because I saw it in one of your videos. You know, like, tell, ask me, ask me what What else? What other information might be valuable for you to understand to help me work through this problem? I'll put that in pretty frequently. Your meta prompt process was super valuable. I think once you get the foundations, then following as many interesting people in a bunch of different places, like my X, my Instagram, my LinkedIn feed is because I like engage with so much of this content, it really pushes it to you. And so, I'm learning. I don't really learn that much on Instagram. There's interesting news, but if you just start to follow the right things so that information gets pushed to you, it becomes more ever-present and you start to use the tools more frequently, you can build up that sense.

Yeah. I mean, I would say you have to move from a consumer to a producer. I think actually you just can't learn by consuming. There's just not enough. You could never consume as much stuff because this whole thing is a dance. So if you wanted to get started in a really easy way, just literally grab the transcripts of all of your calls today. So you ever trained, Kyle Norton, you ever trained in SDR, trained in AE, you ever trained one of your managers? Take that thing, paste it into Claude, and be like, what did I do wrong? What else can I learn? What are the other concepts that I'm getting wrong? So you don't have to think about even talking with AI, just like, how do you talk to people? And you're going to realize that a lot of that is an overlap, that it's an exchange. And you can say, like, one of the things that I will do for both ChatGPT and Claude— so Claude, you can create styles. ChatGPT, you can create a system prompt. And so in that system prompt, I'm like, what I want you to do is always challenge me. Here are the mental frameworks to use. Use the 6 thinking hats and assume everything I type you, that there is a meta question, which is I don't even know if I'm ever asking the right question.

And is that in the set project instructions?

The project instructions are one, one place to do that for Claude. Yeah. And it's the system prompt for ChatGPT.

Oh yeah. Okay. Yeah. And that example is a great one. I'm pushing my team on this right now because we're doing a bunch of new stuff. And I say, I said to a manager literally today, like, just record that call because we're onboarding a new manager. And one manager has to offload a bunch to the other. And I said, just record the calls that you're on and then dump it into this prompt that I've got. It's a MetaPrompt that I built, which is basically like, you are an elite sales enablement specialist operating at the intersection of modern sales methodology and practical execution. And I went through the whole process. I'm like, give me the top thought leaders in this space And so I have it acting as those people with some of their, like, specific strengths, and then it transforms that transcript into a structured training guide. And I used ChatGPT or I used Claude to help me make, like, okay, what should the output— I said this exact thing you're talking about. What should the output of an enablement document be? Well, there's an executive summary. There's a methodology map. Happening. There's a tactical breakdown of, um, of that, of that topic. Then there are, um, the KPIs and, and the key things we need to track. And then there's a, uh, an implementation framework. And so now you can take that whole transcript, just dump it into this, into this prompt, and, uh, get something great out of it. And, and going through that process is like an easy way to start producing content, I think, um, and get exposure to the tools.

And you did something really interesting that you might not even have noticed. What you did is that you did something that makes these models perform really well, is you're like, who are the experts, what are their frameworks, and who has written the most about this?

Yeah, I stole that from you, obviously.

Yeah, yeah, yeah, yeah. Well, you know, I wouldn't be a LinkedIn influencer if I didn't find a way to compliment myself. Um, so, but, but, but this is really important because people have, you know, I'm always like, use April Dunford's frame framework to— this is the way that humans structure the world. And they write a lot about it. And they work— they try to teach you as a novice in any given area. So you can say, who are the experts? Write about their frameworks, explain their frameworks, create a panel of these people and have them debate about it. And then so you can basically get all these famous people for ridiculously cheap and say, okay, great. And that will help you with better outputs because you don't have to rely on your own thinking. You can rely on the whole corpus of the internet. And all you're trying to do is narrow ChatGPT, like focus on this corpus of the internet.

Yeah, that was one of the interesting breakthroughs that I've had recently because most of what's written about sales strategy and go-to-market is basically crap. Yeah. And so by prompting the model to focus on just specific people, as long as you have good taste in those people, then you get a way better output. And we're SMB focused. There's a couple of people that I think are super smart in this that actually want to build in a similar way that I'm building. Like Katie, Katie is one of— Kevin Dorsey is like one of the people that's in my prompt. Um, and because I, I, I would be happy if Claude acted like him. I also put my own name in that, in that prompt because I'm trying to like get it to do my stuff.

Yeah.

And so you can get it, you can tune an output. And the question for me and and we're going to run out of time— but is how do people get to this place faster? I think I want to pitch you on making a CRO AI course. Like, here are the 5 things, here's how you get started, here are the things to do, and here's how not to get distracted. Because for me, I had to wade through so many goddamn YouTube videos and different things and trying to figure it out and testing and trying it. There's got to be somebody that can accelerate it.

I mean, I think you and I I have to work on this because I, I'm having a hard time teaching revenue leaders to get there because they're just starting. So like one of the things that we're probably going to do a course for P Club, which is how to know which agents to build. So like you don't even know. So here's like the whole process to figure out how you would know.

But there's, there's something, there's a, there's one level of abstraction higher than that.

Totally.

Which is because you have to know, like, you have to understand what is an LLM, what, like, what, what is in this mystery box and what are the strengths and weaknesses. Oh, it can create things, it can synthesize data, but it is probabilistic and not discrete, so be careful. And here's where it's good. And I think that content doesn't have, uh, I mean, Katie and Jake Dunlap did a pretty good course on this. They're answering some of these questions. It didn't zoom down into the depth, I think, that where you're going, uh, but that's actually a pretty good starting place.

Well, yeah, you should, you should look at the, my course of Who to Target and What to Say, um, and just give me feedback about where the new one— this— yeah, yeah, the new one.

Yeah, yeah, yeah.

Okay. It's not, it's not done yet. I've been working on 3 months, and I would like it to close this gap. So I do sort of start out with like, where, where are we at right now? It's like, okay, Well, you know, here's like where we came from. Here's how we go forward. Here's how to pick a segment. But it doesn't— I don't have that translation layer that is like how do you operationalize this in your team? How do you— because it is a massive migration to move to this world. It's like a whole data thing that didn't exist before. It's like a bunch of tools that didn't exist before. It's like team members that didn't exist before. And so this, like, how do I adopt AI? You know, you're tempted to just put a bunch of AI stamps throughout your organization, but you need a whole transformation. I don't know that organizations are going to be able to do it effectively because even people like me that have implemented these things for customers have a— I don't have the translation layer to take a pre-AI Kyle and say, yo, Oh, let me walk you through. These are the exact same things. So you should probably look at this thing and say, where is it missing? Because I, I think it's just really hard to paint this new world because not only is it changing so fast, but you need a whole other foundation that is not the predictable revenue foundation to be able to even begin here. And that's the hardest thing, not how do I use AI. It's like, how do I get to a world where AI can do what I need to do? And it's not AI SDR, because what that means is—

notice that we didn't say that once today. That's not the, that's not the place to start. That's not the solution.

Yeah, yeah. So, but this is the problem though, is that people are trying to bolt AI into the previous system, and we need to, uh, think about— this is the moment for the go-to-market motion to have the same thing that we did with the engineering team and the product motion, moving from sort of the Lean Startup to the way in which we build products now. That has happened in the product org, in the engineering org. Multiple people can contribute to one thing with a tool called GitHub. And, and there's a product manager that determines what gets built in what order based on customer feedback. And so we just need to replicate all that stuff over into the go-to-market world. And it just means a ridiculously more nuanced understanding of who we're selling to and why they should care. And the hardest thing is operationally operationalizing who they're targeting and why they should care.

Over the last 5 to 6 years, my number one source of inspiration has been product management because I was at Shopify, super engineering-oriented, great product management craft. And I looked at it, I was like, yeah, this is really good. You know, there's like amazing strategic frameworks in here. And so I went down a big product management rabbit hole and it, it is definitely— there's so much for us to learn and go-to-market from product.

Yeah.

All right. I want to do quickfire because we're— we got to wrap. All right. What— quickfire number 1. What do you think separates a good CRO from a great one? You've now worked with a ton. So what are you— what are your senses for who is awesome at it versus not?

The most like Kyle as possible. System thinkers. So like, didn't you pay me to answer it that way? So it's system thinkers, right? People that realize that Tomorrow's revenue is violence toward a future system. And so those are the people that are like, I'm not building for the playbook tomorrow, I'm building for the process of the next 10 years. And those are the difference between great CROs and fine CROs.

Okay.

Process over pipeline.

What is the most common advice you find yourself giving to mentees? Just, just one. On like other leaders that you're trying to help out, people that are, that you're mentoring? What's the most common advice you give to founders too?

Um, well, probably I'm going to give it to founders. Um, uh, narrow your focus. Your eyes are bigger than your stomach, which is like, don't solve the problem for all restaurants. Like, if I was starting Owner.com today, it's like restaurants that are listed on 3 platforms and don't have a website and are selling just tacos. Like, that's where I would start. Um, and the best— I only know one founder, and they have grown tremendously, that took this advice. Um, and they moved from a $10 million valuation to a $200 million valuation in a couple of years, uh, because they just started axing things. So they had $170,000 in ARR from a segment, and they threw that ARR away because they're like, this is not the segment, we're focusing on this. And they really niched down their TAM was 2,000 companies.

And this is our mutual friend in New York.

Oh, yes, yes, yes. So, so, and they niched down and now they're going after the, like, most niche. And they went to a show and the most other technical person at that show was a print shop. And that's just because they found a segment of the market that no one focused on. And they're like, what if we gave you all of the latest stuff, all the latest AI, the best Silicon Valley had to offer, because the last people that helped you were like Oracle or something.

Yeah, this was super duper impressive. They're— they, they are ripping. And, and he—

and that's— and the way that we started, like, the way that we started, we ran an outbound campaign where we used AI. This is like 2 years ago, whenever they started. Like, we used AI to invent ideas about how AI could be used in their business. So it's like, Jordan, have you ever thought about— and this is like right at the perfect time. And so he was taking like 9 calls a day or something. And they then said, great, these are the 2 segments. They built a bunch of stuff. And then they said, not this, not this, not this. And what most founders, what like 99% of founders would have done is like, we'll just be a multi-layered product company and be like AI for everyone. And so you should go check out that website. I wrote all the copy and I had a lot of fun doing it. And it was relatively easy because because it was just to a very, very, very small segment.

Very cool. What's the hardest lesson you've had to learn?

Follow your internal compass. It leads to the best results, which is there are some things that I just, I have to look inward and follow that, even though it directs me to do things like, it's really easy to get swept up in what someone else wants of you, or what someone else thinks you should do.

Where the money's at.

Yeah, yeah, where the money's at. And so this is why, I don't know, a couple years ago, I stopped running outbound agency. And if you look at my a dollar. If you look at a narrow, it's like my revenue tanked, uh, and so it's like, oof. But then you look at the 6 months after that, and it was, you know, and I 2x my revenue this, this year, uh, my take-home pay, uh, from last year, just because I started saying no. So people come to me today and, and say, hey Jordan, can you do this? And it's like, I can, but I won't. Um, and that level of focus means that I can say to people, look, you're going to get a diluted version of me, and that's not what I want to wake up for. I don't want to be the email deliverability guy because there are people that are way better at that. And if I'm like stuck in the world like, hey, your SPF records are wrong, your DKM is like all messed up— no thank you. I will never be the best in the world about that. I'm going to be the best in the world at where does your data exist, how do you get it in your systems, and how will it transform your business. That's what I want to be the best in the world at.

Cool. Last question: what's the best thing you've read in the last 12 months?

Well, I'm gonna give you two things. Um, the, uh, Farnam Street's mental model post is fantastic, and it was really Canadian.

That's Shane Parrish.

Yeah, well, he speaks like, hi, this is Shane Parrish. Well, we're gonna learn tonight. He's got this like super monotonous— he's like, some people say that my voice is hard to listen to. I was like, I like it, but I could see that. Um, and then a video, uh, David Foster Wallace gave a great talk, uh, called This Is Water, and he opens it with the line, two fish are talking to one another and an elder fish swims by and says, "Morning boys, how's the water?" And the younger fish says, "What is water?" And so it's just a really good sort of talk about the banalities of day-to-day life and to realize that this is water, that this is where we live and this is how live. And it's best to lead with a sense of wonder. And so, I had so much fun, Kyle, sending you that video of that horse in Golden Gate Park in the middle of the night because this is water. Yeah.

Oh, it's so good. Well, this was great. I knew this was going to be a good one. And hopefully, this gives people some idea of how to navigate this very challenging space. Of transforming their organization with AI. So appreciate it, my man.

Thanks for coming on. What's the best thing we can do to end this that will make this the best episode of all time?

I don't know. Where should— end it as the best episode of all time. Give everybody your course for free.

No. Yeah, come to me and I will tell me the one thing you want to learn and I will send you a prompt. Or what I will do is if you send me a message and I don't get flooded with them, which, you know, I probably won't. Not many people actually take me up on this, but I will record a video of me solving your problem with AI. And just like I do for you, Kyle, and I'll just, I'll just make it public. So come to me with like, this is this thing that I'm struggling with, connect me on LinkedIn and say, say, I want a video on how to do blah blah blah, and I'll tag you in it and I'll make it public. You cannot— it cannot be private, but as long as it can be public— so don't share with me anything that you, uh, you know, would only tell, uh, someone on their deathbed. Um, say, here's the thing that I would like to learn, and I will create a LinkedIn video for you.

I can vouch, these are very valuable prompts and, and videos that I get back from Jordan. That's great. All right, let's end it there. All right, thanks Kyle Norton.

Thank you.

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 5-star review, share it with your network, and please join me next Wednesday for another great conversation.