Skip to content

Podcast

Meet the Slowest Startup Incubator in the World—Pumping Out Billion-Dollar Companies

AI & I

Source ↗ ← All highlights
  • Run Companies To Real Revenue Before Scaling
    • Bolton and Watt runs each startup to $5–10M in revenue before hiring a CEO to scale further.
    • This “slow incubator” model focuses on early execution mastery and then hands operational control to hire-stage leaders. Transcript: Dan Shipper Yeah. Dan Friedman So we try to start a new company every couple of years, often in like a really niche vertical that somehow combines software services and some real world component. And the idea is we come up with idea, we run ourselves through five or 10 million revenue. And then we go find a CEO who’s better than us to take it from this sort of the next 10x and we remain involved as a partner to the company for its life really involve board members who have Spent thousands of hours thinking about the whole competitive landscape the company competitors all all that stuff um so it’s a really different relationship than a traditional Incubator which may say okay here’s a million bucks here’s an idea i’ll help coach from the sidelines but we’re actually like in the seats um and that means we’re really concentrated Um we’ve started two companies to date and we’re about to start our third and just to give you an example of the types of things we like to do um our first company is called moxie and it helps Nurses open their own med spas so these are nurses who are doing aesthetic, medical aesthetics like Botox, filler, lasers, etc. And we’re sort of her back office that helps her stay compliant, grow her business, and really everything you need to do with med spa. So we now have hundreds and hundreds of these med spa clinics across the US, all in with nurses. So that was our first business we started about three and a half years ago. And then our second one is a contemporary funeral home. And what that means is we have no physical real estate whatsoever. We do arrange everything online over the phone. When we have sort of in-person funerals, they’re generally at wedding venues that we booked out year in advance saturday night but totally open tuesday morning at nine um and we’re Now the largest um provider of funeral services in california um and just about to launch a bunch of new states so we have a taste for these sort of like weird businesses that are like not Yc zeitgeist that have really rural notifications and the sort of reimagining these types of bundles. Dan Shipper I love it. Wait, tell us, where is Moxie stage wise? Sam Gerstenzang Sure. So Moxie is a Series C company, you know, into the tens of millions in revenue and 600 plus customers. Team globally is like 200. (Time 0:03:02)
  • Moxie Launched Pre-ChatGPT And Adapted Later
    • Moxie launched just before ChatGPT and scaled to Series C with 600+ customers and ~200 people.
    • The company began without AI in mind and later adapted AI as an operational tool rather than a core design choice. Transcript: Sam Gerstenzang Sure. So Moxie is a Series C company, you know, into the tens of millions in revenue and 600 plus customers. Team globally is like 200. So it’s like a comfortably mid-stage company. And vis-a AI, we launched it some number of months before the release of ChatGPT. And so it is hilariously a sort of like just before AI company that had none of that in its conception. (Time 0:05:10)
  • Get Reps By Specializing In Early Stage
    • Specializing by founding stage yields repetition-driven advantage: they repeat the zero-to-$1M playbook across companies.
    • That specialization produces rare early-stage reps most founders never get. Transcript: Dan Friedman I think to some degree, it’s an intersection of where we thought would be the most fun and also we thought the most value would be created. I think we looked at a bunch of other incubators and I’ll put aside for a second, but I think there was someone who wanted to have a bunch of ideas and like let those go and sort of see. And we realized that I think to make to maximize the success of every shot, we realized that we needed to actually go, you know, eat the glass, figure it out, you know, push the boulder Up the hill, figure out if the boulder wants to go uphill, as Dan often will say about that sort of early product market fit journey. And so we thought that was the place where, okay, if you could ask someone else to do that, you could like pay McKenzie to come up with startup ideas. But the hard piece is like figuring out like, okay, what’s actually market signal? How do you make changes when no data quite says what to do? And if we could get really good at that, we get permission to do all kinds of other things as well. And so that was sort of like the origin of it. And I think Dan and I both feel really lucky that it worked so well the first time, because I think if it didn’t work, people would be like, oh, your model doesn’t make sense. And now we can be like, oh, actually it has. Sam Gerstenzang We’ve done this two times and we’re going to doing it in that way. The only thing I’ll add is, and maybe the years is a little bit of a gross heuristic, but my experience and I think a lot of my founder friends’ experience is, before this, I built one company Over 10 years. And years four through 10, there was a little bit of a constant existential question of, am I doing the thing that is most interesting and most useful? And am I spending my time the right way? And I’ve sort of like learned the physics of the business. And now I feel like I’m in purely execution mode. And that feeling of like a little bit of existential dread, one way to avoid it is simply to not be doing, not be responsible for years four through 10, or to vice versa, be responsible For years four through 10 across five companies by the time we get there. And I think that I’m sure it will have some other challenges that, some other psychological challenges, but we are, I think for me, at least, I feel that we’re here optimizing for like What’s most enjoyable and exciting for us. Dan Friedman We also sort of like pretend on the found journey that’s like the same skill set all the way. And I think like what you do and how you do it really depends on the stage of the business. And so there’s actually a lot of value specialization to like, OK, we know what this looks like going from like zero to 1 million revenue over and over again. And here’s what takes from one to 10 or the systems or the people, how do you build intuition? And so we’re also just like getting a lot of reps in, in that early stage that very few people actually get to do that and see what success looks like on the other side. (Time 0:06:28)
  • Prefer AI Durable Over AI Native When Core Work Persists
    • They distinguish AI-native startups from AI-durable ones and prefer AI-durable: use AI to speed ops without changing the core service.
    • Example: med spas won’t be replaced by robots soon, but AI can optimize marketing and support. Transcript: Sam Gerstenzang Like, there’s two good companies to start now. There’s the AI native company that pushes the ball forward inside of some category, or there’s the AI durable company that effectively uses AI where the core of the machine is not going To change. And if you look in our first two, there’s no such thing as an AI native criptory. It’s just like this going dramatically. Yeah, exactly. We’ll put it on the blockchain and use AI. And we’re not expecting a robotic injector anytime in the next seven to ten years. And so the core work of a med spa will be, you know, med spas themselves are actually like conveyors of the latest technology, the latest medical technology. You know, when GLP-1s come out, like med spas are one of the early adopters of actually spreading it to their communities. And, but at the end of the day, like what happens inside the walls of med spa is not deeply impacted by AI. However, there are like around the edges spots where it can really matter in terms of reaching the right customers, serving them and communicating with them effectively at all hours Of the day at an affordable price for the business. And so we want to be great deployers of AI inside of our operation. We want to help our partners deploy it to the maximum effective degree. We want them to be on the early edge, not the bleeding edge necessarily. There’s no need for them to be there. They can always just be a few months later and not take risk with their customer relationships. (Time 0:15:10)
  • Use An Agent To Automate Customer Discovery
    • AI transformed their customer discovery by automating prep, synthesis, and hypothesis-tracking with an agent called Matthew Bolton.
    • The agent reads POV, hypotheses, and recent call transcripts to update evidence for/against each hypothesis. Transcript: Sam Gerstenzang I could speak to the like company discovery process where I think we’ve actually seen probably the greatest transformation, which roughly like maps to the more green fields, the better AI can be basically. And like, whenever I talk to my founder friends that are seed stage, they’re like, oh my God, our engineering is 10x faster. And then I talked to the like, you know, series D friends and they’re like, we’re like 10% faster. What is everyone talking about? Which is the sort of classic dichotomy right now. In the new company discovery process, we are, roughly speaking, every stage has been rethought. So the first step is like, let’s find some verticals to go start to poke around in. And this used to be like a week of Googling and maybe like calling some friends to just get the basic facts. And this is now like a mega prompt to like generate a list of categories and a mega prompt to assess like what we think is a good business and like our particular point of view of what we’re Looking for and start to narrow in on a couple of different, a couple of different ones to go talk to real people in. And we did a, we did a little bit of an exercise this time around where we both did the like AI curation. And then I did like a human point of view. And I actually felt in that moment, like, do you remember the children’s story of the, it’s like a myth of the guy who was competing against the like automated tunnel builder machine To build a tunnel? That’s a real thing. Dan Shipper Like John Henry or something. John Henry. Yeah. Sam Gerstenzang Yeah. Well, it’s a real thing. It’s not like a guy fucking put a hammer through a mountain. Dan Shipper Well, he, he, I think did actually try to compete against the, the, automated thing and then died on it. Yeah. I’m pretty sure. Yeah. Sam Gerstenzang RIP Chuck Henry. Anyways, I was the John Henry in that story and fortunately came out the other end. And we did, as a group, select three categories. One of the three was like my human point of view. Two of the three were like the AI was like, no, no, no, these are screaming mashes. And then more interestingly, we built a insight. This AI ultimately forced us to move from Google Docs to Notion, which I was like fighting for years. And we built an agent identity that we call Matthew Bolton, which is like a horrible name because Matthew Bolton was the Bolton in a lot. And Matthew Bolton is our assistant in being really good in the customer discovery process. So he helps us prepare for every call and like looks at the persona who we’re talking to, looks at our current hypotheses and what our like validation focus is and basically says like, Here’s the areas to dive into. And of course we review and make sure we like talk about the right things, but he makes the prep more efficient. And then afterwards, the like ad transcript goes directly into a notion table. We run Matthew Bolt on and it regenerates a point of view on for each of our core hypotheses on the idea. What’s totally validated? What’s totally invalidated? Where do we need to dive in more? It pulls out the relevant quotes from the people. And so it’s been a huge hit in that way, where it’s totally missed for us, which is also kind of interesting. (Time 0:18:03)
  • AI Can Summarize Calls But Fails As Synthetic Customers
    • Matthew Bolton excels at surfacing quotes and evidence but is unreliable as a synthetic customer.
    • Synthetic AI customers tended to give uniformly enthusiastic answers, failing to capture real customer nuance. Transcript: Sam Gerstenzang This AI ultimately forced us to move from Google Docs to Notion, which I was like fighting for years. And we built an agent identity that we call Matthew Bolton, which is like a horrible name because Matthew Bolton was the Bolton in a lot. And Matthew Bolton is our assistant in being really good in the customer discovery process. So he helps us prepare for every call and like looks at the persona who we’re talking to, looks at our current hypotheses and what our like validation focus is and basically says like, Here’s the areas to dive into. And of course we review and make sure we like talk about the right things, but he makes the prep more efficient. And then afterwards, the like ad transcript goes directly into a notion table. We run Matthew Bolt on and it regenerates a point of view on for each of our core hypotheses on the idea. What’s totally validated? What’s totally invalidated? Where do we need to dive in more? (Time 0:20:18)
  • Don’t Reward Mere Use Of AI; Demand Better Output
    • Demand teams deliver their best work while refusing to award credit just for “using AI.”
    • Seed the right tools, show strong examples, and hold people to higher output standards rather than counting AI usage. Transcript: Sam Gerstenzang I love this. Oh, thanks. I feel like that’s actually really meaningful. You know, I got to say, Sam and I this morning looked at each other and we were like, are we sure we should be on this podcast? Dan Friedman I literally said to Dan, I was like, let’s make a list of all the places we’ve tried to use AI and it hasn’t worked. Cause I think, so it’s nice to hear that from you. Sam Gerstenzang Exactly. Exactly. We’re like from the guy himself. The so you, you know, ultimately you just say like run PNC analysis and it runs through all these different steps. And one of what feels like personally meaningful about this is we try to be intellectually honest with ourselves. This is like something we hold ourselves to. And I think you probably know this because you’re building new products all the time. But like the nature of starting something new is it requires like a manic energy and a little bit of a suspension of disbelief because there’s just like no reason any new company should Succeed or any new product should succeed. And this keeps us like really rigorous and fact-based, which is what we aspire to. And we can balance that with like our own, you know, we can be sycophants to ourselves and ask it to remain fact-based and balance these two opinions. But it, it like totally helps. Dan Shipper Do you find that it’s, it works with the like, cause I find if you’re, if you ask it for reasons against or for, it’s going to come with, it can come with anything. So like, do you find that it is actually good at weighing evidence for you? Or are you, you’re just surfacing it and then like making your own conclusion based on the evidence that it surfaces for and against what you, what you believe? Sam Gerstenzang If we try to ask it an opinion on a high level question, we, I have not, like, I have not given, I’ve not trusted it with that. Or I, you know, I tend to take those results with skepticism, but I think it’s really good at like finding the quotes, you know, at the end of the day, to support a hypothesis in a perfect World, what I want to bring to Sam is here’s the three key things that must be true in this idea. And I’ve got three quotes from different people that like directly speak to each. And this will just like much, much more efficiently help you get there. (Time 0:24:50)
  • Drive Adoption By Showcasing Workflows And Comparing Outputs
    • To drive AI adoption, surface clear examples and reward better results not AI badges.
    • Find early adopters, share their workflows publicly, and compare outputs to raise team standards. Transcript: Dan Friedman Yeah, I think it’s really interesting. Like inception company, series A company, series C. Um, and I think there are like two parts of this of like one, how do you actually get people to start using AI? Uh, which we should talk about. And then, then I think the second piece is like what’s actually worked and what hasn’t. Um, Dan and I were talking about this the other day of like, should you have an initiative, like an AI initiative? Is that a good idea or a bad idea? And I think my perspective was like, it’s a bad idea because you don’t want to sort of lead with the hammer. And, you know, we all remember the time, like, NoSQL, everyone was putting everything in NoSQL, whether it belonged there or not. Everyone was building a Slack bot, you know, 10 years ago, whether one needed it or not. But, like, you do need something to kind of, like, shock the system. You need something to be like, okay, great, there’s, like, a new tool set. And I think, like, the point of view I’ve come to is you shouldn’t give anyone credit for using AI, but you should make sure that the expectation is that they use AI or sorry, the expectation Is they’ll deliver the best product and output knowing that AI exists. And so to do that effectively, you both need to like sort of seed what are the tools you can use and give a lot of good examples. And you need to start like demanding that when you see the results from someone on your team, that they’ve actually used the best possibilities. But you don’t get any points for generating a bunch of copy that’s clearly written by AI and it’s bad to read, right? Like you have to best the copy. (Time 0:32:47)
  • Use Throwaway Apps To Unlock Immediate AI Gains
    • AI enables fast greenfield experiments and throwaway apps that bypass full engineering integration.
    • Example: a simple app resolves hospital or town names for Meadow Memorials instead of heavy integration work. Transcript: Dan Friedman So a few examples of this, like really good at generating landing pages and like pushing our thinking there and coming up with stuff, a ton of work to like integrate that back into Webflow And have it fit with our system and have a consistent header and footer. And so there’s almost like two phases. There’s like research development and then there’s like production. We’ve had a number of people on the team build sort of like throwaway apps that have been really useful. So for example, one of the challenges we have is for the funeral business, people call in and they might mention some town name and we know whether we service that. So designers spun up an app where anyone can type in a hospital name, a town name, and it resolves whether we can find it or not. Instead of integrating that and spending engineering time on figuring out how to deploy that safely, integrate it, all that stuff, it’s just a separate app that’s a link from our main One. And so we found that enabling those types of things has worked really well. And our engineers have gotten more productive, but a lot of the core things an engineer does is still the same. (Time 0:34:50)
  • Sam GBT Wrote Outreach In Their Voice
    • The talent team trained ‘Sam GBT’ on Sam Gerstenzang’s blog voice to outreach to candidates on LinkedIn.
    • That personalized AI outreach worked well and the creator had even forgotten it existed. Transcript: Dan Friedman Our talent team was reminding me the other day that they’ve made something called Sam GBT, which they trained on all my blog posts and they use to reach out on my behalf to potential candidates On LinkedIn. And so it’s like train my voice. I forgot it existed. And that’s worked really well for them. And so there’s sort of these like places where it’s kind of unlock enable into an existing system. (Time 0:36:26)
  • AI Helps Greenfield Projects Far More Than Mature Systems
    • Newer projects see dramatic AI gains; mature systems see incremental improvements (e.g., 10%).
    • They observe newer engineers on greenfield projects accelerate more than teams touching complex legacy systems. Transcript: Sam Gerstenzang Early word out of the Max Engineering is like, yeah, this is better, but not this is a step function, different experience. And we are doing a lot of work there to like retool in order to experience more benefit because we’re seeing exactly the like newer engineer working on more greenfield project moving Much faster than, know, working on something that touches multiple parts of the system. And that’s like, you know, 40 person product engineering team, more mature code base and so on. So we have not seen the night and day transformation that the X sphere is reporting. Dan Friedman I just, I, I, we saw a story like similar from on on, like, it feels like over the last year where everyone’s talking about like geo instead of SEO and like everything’s going to change And agente commerce. And I, and I think like, that’s one place also where we’ve seen much more incremental change. We’re getting more traffic from chat GBT. Um, but it’s almost like it’s another channel for us. And we have to think about the same way we do like a paid search. There’s like a cat and mouse game to figure out how to get free results. There’s gonna be a paid version of it. But fundamentally, like one, we don’t think people are going to buy a funeral via chat. And two, that may not be true for a lot of products. When I do a flight search, I still prefer to do that myself versus ask a travel agent to do that. And so for us, it wasn’t really a shift in the way we thought about marketing. (Time 0:38:39)