Today, I am hosting Chat Joglekar, the CEO of Baton and a former VP at Zillow, to talk about how he built an AI-native marketplace for small business acquisitions.
Baton (which raised Series A earlier in 2025) helps small business owners sell their business at the best price by aggregating demand from buyers, ranging from search funds to individual accredited investors. It makes money through commissions from each transaction like any investment bank or business broker.
I personally loved interviewing Chat, because AI native market places are a breath of fresh air compared to the Enterprise AI space, which recently feels overhyped. This interview proves that many business models are viable with AI.
AI has been instrumental in scaling Baton. Baton has automated the majority of onboarding and due diligence process (with just the right amount of human-in-the-loop), helping it grow its marketplace to 2,000+ active listings. Baton is using AI in every department, not just GTM or coding, and achieved 10x growth with minimal opex increase.
In financial domain, trust and safety are paramount. Chat will also talk about how he balances heavy AI automation with maintaining the “human touch”.
My favorite topics from this episode were:
AI agents are eating SaaS budgets: we discuss how AI agent spend is going exponential, and there’s no limit in sight.
How multi-modal AI has changed document intake, and made it accessible to even startups without buying enterprise products for OCR / document processing.
How we are in the “$3 uber ride” stage of coding AI agents, and just how much pricing power Anthropic / OpenAI has.
How low code AI tools might be on the way out.
Which roles you absolutely cannot automate in the foreseeable future.
The shift from billable hours to flat fees in services.
If you care about AI-native operations, marketplaces, or the future of services, this episode will challenge a few assumptions you probably hold.
Consider supporting Enterprise AI Trends by upgrading. Limited time offer - new annual plan members will get a 30 minute 1:1 call with me.
Timestamps
- [00:00:53] Intro & Why Build a “Zillow for Small Business”
- [00:02:21] Baton’s Business Model and Typical Sellers
- [00:03:38] Lessons from Zillow & Bringing Offline Supply Online
- [00:06:43] Small Businesses as a New Asset Class
- [00:08:57] Market Landscape: BizBuySell, OffDeal & Verified Data
- [00:11:14] Pricing Tiers, Deal Timelines & Buyer Personas
- [00:19:37] Solving the Marketplace Cold Start & Matching Supply and Demand
- [00:27:38] Going AI-Native Across Valuations, Listings and Internal Tools
- [00:43:45] Humans in the Loop, AI Judges and Data Reliability
- [00:49:30] AI Search, AEO, Outbound and SMB AI Adoption
- [00:58:02] Irreplaceable Human Roles, AI Economics and Baton’s Long-Term Vision
- [01:03:12] Trust, Human Touch & Baton’s Long-Term Asset-Class Vision
Transcript
[00:00:53] Intro & Why Build a “Zillow for Small Business”
John: Great to have you in this podcast episode for Enterprise AI Trends. Super excited to talk about Baton. Am I pronouncing that right, Baton?
Chat: Yeah. Yep. We’re trying to pass the baton from seller to buyer.
John: I think that’s a perfect segue into talking about Baton and the problem that you’re solving. Could you tell us a little bit about yourself and what Baton does and what’s the thesis?
Chat: So I’ve been at a bunch of different startups and was catching up with a friend near the tail end of COVID, and he was looking to buy a small business, which I thought was this crazy, weird, and interesting thing when I heard it, and started to talk about it with him.
And near the end of the conversation, he said, “Actually, talking to you reminds me, the thing that’s missing in this space is the Zillow for small business.” And I’d been at Zillow for over five years, and so it triggered something in my head. I started to think about it and do a lot of pattern matching to what real estate used to be, pre-Zillow, Trulia, and PropTech.
Ultimately, we felt that it’s a kind of opaque, illiquid market where trillions of dollars of small businesses are changing hands. But generally, owners are not getting a great outcome ‘cause they’re selling in a thinly traded market, or someone’s coming and reaching out to them and they decide to sell, but they’re not really running it like a process that you would think of in traditional M&A, or honestly in real estate.
[00:02:21] Baton’s Business Model and Typical Sellers
John: Yeah. Could you just tell me about what Baton’s business model is and how it makes revenue, and maybe go into who the typical customer is? I understand that it’s a marketplace, so on the buy side and the sell side…
Chat: Yeah. We think about it like this: in this market, what’s missing is quality supply. So we’re really focused on the supply side, which in this case is small business owners. We’re trying to empower them, one, with a free valuation. Two, when they decide to list, we actually act as essentially the sell-side advisor, almost like an investment bank for Main Street.
And we take a success fee at the end of a successful sale. There’s a nominal retainer that we charge along the way just to make sure everyone’s engaged and motivated. And our typical—I would say like four out of five—sellers are baby boomers who have run their businesses for 20-plus years and are now looking to retire, spend more time with family and grandkids, and have built something amazing.
But again, to the prior question, they don’t really have a great option unless they know a business broker or unless they’re big enough to where a bank is actually interested.
[00:03:38] Lessons from Zillow & Bringing Offline Supply Online
John: So I understand that you were a VP at Zillow. What kind of insights or experiences have you had that allowed you to scale Baton, and how did that help you when you first got going?
Chat: Yeah, I think a lot of the archetypes in the markets are similar. My mom was a real estate agent, gosh, 40 years ago. And when she would show people around, the only place you could see what the value of your home was, or homes in the area you were looking, was to walk into a brokerage and print out stuff off the MLS.
And that’s how small business transactions are today—it’s really unclear for a small business owner. And so I think doing the same thing, like what Zillow was able to do initially via the MLS and just getting listings live.
I actually helped launch and scale the new construction marketplace, and I think there are a lot of parallels between that marketplace and Baton, because a lot of new construction community listings were not actually online. You would drive by a community and be able to walk in, but people moving from a different city couldn’t really see it. And so a lot of it was focused on getting that supply online and legible.
And if you could, then the buying community—the demand—could see it and start engaging. We saw that to great success in the new construction marketplace at Zillow, and we’re running the same playbook here at Baton.
John: So that’s interesting because if it’s a new construction, obviously there’s not a lot of public information. And even if it’s a small business, you don’t necessarily know who owns that nail salon or a logistics firm, a local freight broker—something like that. How does that actually work to get that digitized and online? How do you actually do that at scale?
Chat: Yeah. At Zillow, we actually used a preexisting XML feed that some new home communities were using, and we made it the standard and asked that they would submit stuff in that standard. That was helpful. But they had to produce it. There was another middleware provider that we ultimately acquired that was running that standard.
For Baton, everyone kind of has QuickBooks or a P&L. And so we’re really about capturing that information, which then on our side is in a structured form where our process can ingest it, understand the operating income, add-backs, and the adjusted cash flow for the business.
Now, when you go on Airbnb or you go on Zillow, you know the thing that you’re looking at is pretty legible. You can look at it and see pictures. You know how many bedrooms there are and bathrooms there are, granite countertops, et cetera. And that’s been through over a decade of work that Zillow has done, and we’re on the same path to be able to bring all this inventory to a legible place, which is really easy for the demand side to peruse.
[00:06:43] Small Businesses as a New Asset Class
John: Is it fair to say your ultimate vision is to become a Zillow for private businesses, typically SMBs that are trying to sell? And I think the main vision is that—the whole “silver tsunami” of retirees selling their business and trying to get liquidity?
Chat: Yeah. Ultimately the vision is to create the next great asset class for everyone. Because if you think about it, everyone thinks of their home as an asset. Stocks and bonds are assets—things that are transparent and liquid and legible. And small businesses are a trillion-dollar category. There are tons of them that trade every year—trillions of dollars of small businesses trade hands—yet it’s stuck in this kind of opaque, illiquid arena.
So I think a marketplace is our take on the best way to expose this, make it a lot easier for owners, buyers, even capital providers, to understand the asset value. And if they do, that makes it a lot easier to lend on, a lot easier to build financial products on, a lot easier to buy and sell.
John: The issue is that the information about these businesses is not even online, and therefore you can’t even do anything. You can’t even get started on underwriting asset-backed securities on top of this because, unlike real estate, there’s nothing known about these businesses.
Chat: Exactly. There’s no history. There’s no standard way of reporting. No one has collated this.
We have close to 10,000 financial records now on businesses that we’ve done valuations for. That’s now a cohort of information that we have that we can use to build better models. We can start to really get better valuations out for small business owners, which ultimately is how these things trade, right?
If every small business in America wanted to sell their business for $10 million, regardless of their adjusted cash flow, there wouldn’t really be a market. But if we can actually use comps and use multiples to make a market between the demand and supply, that’s the key.
[00:08:57] Market Landscape: BizBuySell, OffDeal & Verified Data
John: You mentioned that you’ve done valuations for 10,000 businesses, and I think that’s a pretty staggering number. So we’ll get to that in terms of what the funnel looks like. But before that, this whole category of trying to broker the sale of private SMBs has been around. There are sites like BizBuySell.
And I personally recently covered a YC-backed company called OffDeal that does something like this in this space. So how does Baton position itself, and what are some of the key players in this market?
Chat: Yeah, I think BizBuySell is the clear leader, but we think of them as a listing site—maybe in some ways similar to Zillow, right? Zillow is matching agents to people looking for homes. Very similar: BizBuySell is primarily brokers who are listing something that they’ve acquired as a mandate.
We really felt like, when we looked at it, the job to be done is to successfully sell the business. Because for most sellers, it’s their first and only time that they’re going to go through this process. And for many buyers on our platform, it’s the first time that they’re buying a business. So actually getting that transaction to close is the key job to be done.
So actually being a sell-side advisor and doing the whole kind of soup-to-nuts transaction—I think OffDeal does something similar. And so I think the ability to bring technology and scale to bear has been good.
And then BizBuySell, for example, isn’t verified data. An owner can go on there and list, just put in numbers for their cash flow, their revenue, et cetera, and it’s up to the buyer to reach out, get an NDA, get the financials, and then understand if the cash flow number that the owner has stated is true.
We’re again trying to handle a lot of that overhead and prep work on our platform so that we are doing the valuation. We see verified P&Ls and financials, we see tax returns and can reconcile them to the P&L before a listing goes live. So that’s all the kind of work that we’re doing ahead of getting something to market.
[00:11:14] Pricing Tiers, Deal Timelines & Buyer Personas
John: Wouldn’t that incentivize you to get larger transactions, or to skew toward doing more larger notional transactions in that business model?
Chat: Yeah. We have three pricing tiers: Light, Pro, and Premium.
We’ve built them so that smaller businesses can sign up for Light and get a service that’s appropriate for an under half-a-million-dollar valuation business, if you will. Kind of one-to-five-million-dollar businesses, one-to-eight-million-dollar businesses, are best on our Pro plan. And then much bigger businesses are on the Premium plan.
Ultimately, sure, it is certainly better for us—we make more money the larger the business is—but I think we’re really driven by our mission to empower every small business in America to get to their next stage and honestly keep these businesses around in communities. So I think what we’ve discovered is that we can drive, with scale, a really tight, cohesive sales process for a $200,000 business as much as a $2 million or $20 million business.
John: Just curious—we’re gonna get to all the AI and the tech enablement later, but in terms of just processing and deal lifecycle, does the size of the business in terms of notional have any correlation with how long it takes to close or find a buyer?
Chat: Yeah, that’s a good question. I would say not really. I think the financing is probably the biggest underpinning. If someone’s getting an SBA loan or external capital, it tends to take longer than if someone’s doing a cash offer or they have committed capital and can move quickly. That means that the diligence process is generally a lot quicker.
But we find that the rest of the process—going to market, being listed, and getting to under contract, getting LOIs and under contract—is pretty similar, whether you’re a small business or a large business.
Granted, I do think larger businesses get a lot of engagement and interest. But even though they get a lot of interest upfront, there’s a lot of negotiation to get to the right offer. So it might be a couple weeks spent doing that. Ultimately it works out to about six weeks from listing to LOI, and then generally it’s about two to three months in diligence.
And we’ve had ones that are a few weeks if it’s an all-cash offer, all the way to six months if it’s financing and something goes sideways and we have to find another lender, all that kind of stuff.
John: There are a couple failure modes of how a deal can not work out. And it seems like one of the big ones is financing. Is it fair to say that the typical sort of lifecycle of this matching is anywhere from a month to—like you said—a couple of months or even a year? And most of the time is spent in finding the money to buy because a lot of these are LBOs or…
Chat: Yeah, a lot of them are SBA loans, and someone’s getting approved by a lender. Our average, I think, is a little under six months right now from basically signing with us to closing. On average.
John: Could you comment a little bit about who these buyer personas are? Like, who are the people?
Chat: Yeah, mostly owner-operators—folks buying their first business, coming out of tech, coming out of finance. Instead of working for someone else, they’re entrepreneurial and they want to work for themselves, and they’re making a bet on themselves. But similar to any other startup, they’re de-risked by the fact that this business is maybe throwing off half a million dollars in cash flow, so they can come in and take it over operationally, grow with sales, grow with marketing, whatever it is.
Largely owner-operators. And we do have some professional buyers—private equity, search funds, roll-ups, holdcos. They tend to just, in numbers, be a much smaller percent of the buyer pool, but they tend to actually outperform in terms of the number of deals they can win, ‘cause they can move quickly and they’re looking for something.
[00:15:31] What Types of Businesses Sell & How Auctions Drive Multiples
John: What are some popular categories? And maybe you can comment about what makes an ex–tech worker, like someone in a big tech company, equipped to run a nail salon.
Chat: Yeah. I don’t know if a tech worker is well equipped to run a nail salon, but I would say our most popular category overall is services businesses.
So either home services—like B2C services—or B2B services: a coffee roaster that sells to B2B customers; a CNC machining company that has a defense contract that’s got durable cash flow. I think those other B2B businesses like warehouses, distribution companies, et cetera, are probably the majority.
Then we have some retail, a smaller amount of hospitality, some healthcare, but I would say largely service businesses are what we see. Or more scaled retail, like an art gallery that sells for two to three million dollars.
John: It would be very interesting—and maybe you guys can publish a research piece later—but it would be interesting to look at the multiple dispersion between these categories in the private markets as opposed to the public markets, where you see crazy multiple dispersion among services and within services: there’s software, there’s defense, and there’s crazy change.
It would be interesting to know if a stable defense contract would make the multiple a lot higher as opposed to, say, a retail…
Chat: Yeah, totally. I think we’re starting to get to that point where we can break stuff up by different categories—the category behind the category, right? Like “defense contracts” as a category. I think we’ve had a handful of those across the history of the company.
Generally, it’s a pretty tight multiple dispersion. But then as you get into the details of the business, there are some unique factors which will, one, drive a liquid auction for the business, which then enables the multiple to grow and increase. If there’s five or six or 11 offers on a business, obviously you’re going to see more of an increase in the multiple as people make bids on the business.
John: You mentioned an auction. I think that’s very interesting. Could you comment about, when a new listing comes on board and the seller is interested in selling, exactly how much time does Baton allocate to get as many offers in as possible? You mentioned an auction—how does that work?
Chat: Yeah. Generally, we’re trying to get every business, as much as possible, to multiple offers. Whether I’m calling it an auction, it’s essentially getting to multiple offers. Because what that effectively is, is great price discovery, right? If you’re getting three offers from buyers and they’re all—let’s call it like 3.6 million, 3.7, and 3.4 million—for a business, we have a pretty good sense that this is the market price.
There might be different terms, there might be different earn-out structures, et cetera, but you can start to negotiate and think about the structure from there. But ultimately what we find is when we get more than one offer, there’s a very high probability it’s going to go under contract, it’s going to close, because we’ve done our job as an advisor in getting to the correct price discovery.
We get a good chunk of folks that start getting interest and are priced appropriately to a place where they’re getting multiple offers. And a lot of that is, in some ways, just a transaction funnel of how many people see the listing, how many people go into the data room and check it out, how many people schedule a buyer–seller meeting, how many people make offers, et cetera. You just work your way down the funnel. And we know if we get a certain amount at the top, we’re going to feel very confident that we are going to drive them through the process and get to a sale at the bottom.
[00:19:37] Solving the Marketplace Cold Start & Matching Supply and Demand
John: So what are some of the biggest challenges that the sellers face, as well as the buyers? We can segue into just the challenges of building a marketplace like this, because you have to bootstrap both sides. How did you prioritize which side to seed first and all of that?
Chat: Yeah. Maybe I’ll answer the second question first.
I think in any marketplace there’s the cold start problem—when there’s nothing in the market, how do you get demand to show up, and how do you get supply to sign on? So that’s where we focused early on. And the biggest challenge in cold start is, what is going to be meaningful?
In this market, this is a supply-constrained market. So we focused a lot on the supply—aka small business owners—giving them the free valuation, giving them value, then helping them and supporting them in the sale. And then supply begets demand. More demand then turns into more supply, because we were like, “Hey, you’re going to come on this platform and you’re going to get 500 or 1,000 buyers that are interested.”
John: Just for the audience: supply in this case is the businesses that are trying to sell, and the buyers are the business buyers.
Chat: Yeah, exactly. They’re looking for a business or something to buy.
I think the biggest challenge for the supply side initially is even understanding what their business is worth. Most small business owners don’t have a clear sense of what their business is worth. So we’re trying to unlock that first and get them on the platform.
Once they’re listed on the platform, quite honestly, one of the challenges is you’re a busy small business owner running a business. That takes a lot of time and energy. And so also trying to sell it is a challenge. Small business owners that are doing this on BizBuySell—“for sale by owner,” if you will—it’s pretty challenging because you have to put all the financials together, you have to answer all the questions. That’s really where we try to step in and use humans as well as technology to make that more seamless.
I think from the buy side, if you talk to any buyer, they’ll say by far their biggest challenge is deal flow—just seeing quality supply that they can make offers on. Most buyers love our site. They say it’s one of the best sites they’ve ever interfaced with. It’s super easy to look through listings. They always say, “Hey, I wish you had more listings.” And it’s our job to get more supply.
But also, a lot of the really good supply comes off the market pretty quickly. Back to the real estate analogy: if you have a home in a great neighborhood with great school districts and all this kind of stuff, it’s going to stay on the market for a couple weeks before someone makes an offer and it disappears and it’s under contract. We see the same thing—that really great businesses are on Baton actively for a matter of weeks before they go under contract and then they’re in diligence.
John: I guess by this point you have a sense of the types of businesses that are easier or harder to sell—or, I guess, the type of businesses that buyers may be looking for. Would that be correct?
Chat: Yeah. We have over 20,000 buyers on the platform that have registered and given us their criteria, so we have a really clear sense of what the buy-side community is looking for. The nice thing is it actually pretty nicely matches up with the supply we have—services and durable cash flow businesses.
We are hyper-focused on how we do a better job of matching. Is there a buyer out there that should be exposed to a listing, a piece of supply that, for whatever reason—I always use this example—they’re looking for a business in Northern New Jersey with over half a million dollars in cash flow, and we have a business in Long Island that is 490K in cash flow. We have to be able to merchandise that to that buyer because they can make the decision that actually that’s a 45-minute drive, I’m willing to do that, because otherwise this is a great business.
I think it’s a lot of “How do you improve the matching so that you can get buyers good listings and you can increase engagement on the supply?”
John: That’s interesting. I did not realize that location is one of the key factors, but it makes sense. If you’re buying a factory, you might want to check it out before buying it and maybe spend multiple days on it, so it makes sense.
Chat: Yeah. I think if you ask a lot of buyers, they’ll say they want to be driving distance to the business so they can pop up or pop down in case anything goes wrong or they need to deal with something. They want to be close.
We also see people that are willing, once they acquire a company, to move to that location. We have this business in Albany, New York, and someone from LA acquired the company and then moved to Albany. People will do that almost like moving for a job. It’s similar, but they want to be in a location that works for them. In this case, that person didn’t have a family as far as I know, so they were able to pick up and move. Whereas if you have a family, you might want to stay in the same area.
John: Yeah. I assume a lot of these buyers—and I personally know a lot of search fund people—they actually skew quite young. Younger than you would think. The fact that you said this person moved kind of supports that impression.
But just going back to growth topics on the buyer side, what are some criteria that you’re incorporating, maybe programmatically or using heuristics, to qualify a good buyer persona? And then also for sellers: what kind of criteria do you have at the top level to ensure that you are attracting the best people on your marketplace?
Chat: Yeah, for us, we have an onboarding flow for buyers. We ask some simple but key questions. We make sure that someone has proof of funds before they’re making an offer on a business, so it’s ensuring that we’re getting quality offers and not a teenager who can just sign up and make offers on 37 businesses or something.
That’s actually worked really well. We haven’t had to implement a ton of KYC-type things because people are obviously signing a platform NDA, so they are held to an NDA and confidentiality when they’re looking at listings. It’s been really good—we have a high-quality buyer pool, and our sellers have consistently told us that they really appreciate the quality of demand that we are able to source and bring.
From a seller perspective, it’s folks that are motivated. Maybe it’s unsurprising, but people that have the motivation to sell and have reasonable legibility on their business are the canonical ones. We know those things are going to sell. If someone’s interested in selling but doesn’t have to sell, they’re going to wait for the right offer. They may get a good offer but not have to move.
So we think about it as: what’s a seller’s motivation? And similar to the buyers, we ask a bunch of onboarding questions in terms of your timeline for selling. We prepare a lot of financial data ahead of that. We’re trying to get motivated, serious sellers to interface with motivated, serious buyers so we can move really quickly in the process.
[00:27:38] Going AI-Native Across Valuations, Listings and Internal Tools
John: So switching a little bit to AI—and I think this is a good segue—how are you currently leveraging AI for Baton, and what are some home-run use cases for building?
Chat: I think we use it kind of soup-to-nuts everywhere—everything from outbound, sourcing, and go-to-market, to sales, to operations and getting listings live. AI writes the listing descriptions ‘cause it’s pulling off the owner’s video interview, their survey questions, so we can write them automatically. We have humans in the loop along the way, but our valuation report is done with AI. We’re pulling adjustments via AI and technology.
We’re able to do outreach to small business owners. I think that’s probably the biggest slam dunk for us—just being able to programmatically do outreach to owners at scale without having a team of 20 SDRs writing emails. We’re able to do that.
The other big slam dunk is on operational overhead to get reports and get all the listing prep done. Traditionally, for an investment bank or for a business broker, it’s hours or weeks of work. We’re doing it in a matter of hours because we’re able to pull it together really quickly and get a listing live.
John: Just to quantify the impact: pre- and post-AI agents. Let’s say 2024—you guys have been around since 2022, I believe—you’ve grown to having due diligence done on 10,000 listings.
Chat: Yeah. It’s hard in terms of people, ‘cause we don’t necessarily have the counterfactual of “We had a team of 100 and now we only have a team of 10.”
The one thing that sticks out in my head: it used to take me, pre-AI—even in the tool—probably 20 minutes to do a valuation. And now all that stuff has been productized and we’re using AI for adjustments and all these other things. I think it takes 15–20 seconds. The longest part of the process is a human who just reviews what the computer does to ensure that it’s buttoned up before we share it with the owner, which is a matter of a minute.
So we’ve taken something like a 20x improvement, right? Just with the ability to use AI and leverage these things. I think it’s probably similar in the listing creation. It used to take a couple days or several days to get all the things together for a listing and put the CIM together, if you will. Now that’s done in 20 minutes or something like that.
John: So aside from time saved—I think the hot debate right now is: how do you measure the ROI of these AI use cases? I personally think that time saved sounds great, but it’s more exciting when you can say, “Hey, there was a lift in some top-line number,” or a pipeline number.
So what would be a way to quantify the impact in terms of your pipeline relative to the amount of human labor that goes into tracking it? Because at the end of the day, it’s flipping the equation, right? Time saved is—if you just switch the denominator with the numerator…
Chat: Yeah. I think we’ve over 10x’d the listing value that we have on the platform with maybe 2x the amount of people. So it’s not 10x the team to get 10x the listing value. We’re seeing real gains because we can have the humans spend time on the stuff humans need to spend time on, which is connecting with the small business owner, building trust, explaining the valuation, as opposed to all the time and energy that was being spent on busy work, if you will.
It also enables us to do things that aren’t blocked by humans, right? We can respond to customers when humans are sleeping, because we have messaging and AI able to answer questions without humans having to show up in the morning and kick the process off or answer the question at eight in the morning.
John: So it seems like—assuming that number of listings is one of your KPIs and that has 10x’d with 2x increase in the input—a 5x lift, what would you say are the pre- and post-AI-agent differences? At a high level you can go into as much detail as you want, but before and after: how are things different?
Chat: Yeah. I think before, it was pulling a bunch of different… Imagine: here are the seven or eight different places we have information about this business. You, as a human, are going to go and look at it and read it, listen to it, do all this kind of stuff, and then go into our platform and start creating a listing.
A lot of that now—one, just with the ability to write software so that something the owner said in their onboarding survey is automatically in the listing—saves time. But also, owners end up with maybe a 30-minute video interview, and we have a transcript from that. They’re answering about 50 or 60 questions explaining a bunch of stuff about the business.
AI essentially takes those two things and, in a matter of one second, has created an amazing description—with a prompt that we’ve built over time. Essentially a prompt with an AI agent that judges the versions that come out. We have a really tight, well-done teaser description that used to take—we literally had humans that we were paying as content writers—over an hour or two to do that. Now that’s done cheaper, better, faster—the whole thing.
So I think that’s the kind of, back to your point, exciting piece. The exciting piece is less about taking this process and making it quicker. It’s actually, to me, about net-new things that we bring to the table in the next year. We just launched a beta of AI search, so you can search through all the listings and search for stuff like, “I want accounting businesses in the Northeast with over half a million dollars in cash flow,” and it will search for those and give you the results.
Rather than building 87 different filters—that would be the traditional way we would handle that—and you would type in a keyword plus, “Is there East Coast? Do we have New York City?” Some of those things. Ultimately, creating net-new experiences and interfaces that engage with the proprietary data that we have on the platform—which are all the files, all the information that the owner has done—is a way that we can actually be a marketplace and expose that in an easy way so a buyer can ask whatever question they have about the business and get an answer.
Because it’s in video 10 of 12 in the video interview. They don’t necessarily have to listen to that video interview. They can ask the question and then click in and see the full context.
John: So it seems vibe coding especially, and coding AI, has allowed you to do more of these operational-excellence and internal-tooling builds that allow you to ship more features onto your platform. So I guess some of the AI impact can be quantified in terms of a KPI, but some of it is just more value created on the platform, faster.
Chat: Yeah, and it’s just the ability, right? It’s like we give people laptops instead of slide rules, or whatever it is. We want to give people the tools to do their job.
Our salespeople do call prep for all of their calls just using AI. They’ve built a prompt so that they can do research, which every seller wants to do, but when they’re on 15 or 20 calls in a day, that was harder to do manually. Now they’re able to do it. So they’re empowered to use the tools in front of them.
John: Did you guys build your own sort of sales coaching/Gong-type app, or are you currently using a vendor?
Chat: Yeah, we use Gong. But then we’ve built something on top that takes in the Gong transcript and does a bunch of other stuff on top. So we’re using a call-recording vendor, but then we’re taking the transcripts and using our own custom agents from there.
John: You mentioned earlier that you have a seller interview where you’re getting the seller to record themselves answering a bunch of questions. Can you tell me when was the time when you were saying, “Okay, AI is good enough to do this”? At what point did you feel like, “Okay, it’s inflecting and we need to… this is good enough, we’re gonna go AI-native”?
Chat: Yeah. I think it was actually one of the key use cases that made us feel we could jump in, burn the boats, and go fully AI-native. We had a contractor writing the listing descriptions for us, and they were listening to the video interview. Then we actually just built—it was a ChatGPT way long ago—we basically said, “Copy and paste the answers from the owner survey, copy and paste the transcript in, and then make a description.”
When we started doing that, initially, that was an order of magnitude better and way quicker, fundamentally. Then we tried it on a bunch of listings. We had the human doing it in parallel and the machine, if you will, and we did it for a few weeks. We basically realized that in every single case, the machine was better. It was literally writing a more descriptive, more detailed, and more buyer-friendly description than the copywriter was.
Because it’s a perfect use case for AI, where AI can read the full context of all that information and, with an appropriate prompt, actually produce an amazing description. From that moment we were like, “Oh, what if we use AI that’s trained on the thousands of times that we’ve manually done adjustments in valuations? Can we use AI to actually have the system do it automatically?” And it did a great job.
So we’ve found these use cases. Nowadays we just start from the get-go with AI because it’s a lot quicker. Back to the AI search thing, we’re not building filters and then layering on AI to see how it does it better. We just start with AI.
John: In terms of AI vendors, are you guys all-in on one vendor for now, or are you using multiple companies? And how much credence do you give to the idea that the cheap models from, let’s say, China are actually going to really displace the foundational labs in the U.S.?
Chat: Yeah, that’s a good question. We use almost every model here for a variety of different things. We find Claude good for certain things. We find OpenAI good for other things. We’ve actually found Amazon Nova has been our sweet spot lately.
We haven’t done a ton with the Chinese-lab models, because we’ve found that between some of the Flash or Nova models, from a cost perspective and latency perspective, they’re actually performing really well. Over time, exactly to your point, we’ve tested some of these other models for interesting use cases like reading a PDF, and when we’re getting into a new use case, we’ll try everything and see.
But I feel like there will always be a use case for the frontier. TBD if, in five years, we need the frontier stuff or that’s reserved for cancer research and rocketry and deep-tech science research—maybe that’s where frontier models are. For most of the other stuff, it’s okay to be on second- or third-generation.
I think that’s a question. But like any technology, I think companies find a way to use every layer of the stack. There are things that the frontier models will do in five years, from an AGI/ASI perspective, that will be fundamentally things that we have to internalize at Baton.
John: Just to get a sense of your AI spend and AI adoption—you don’t have to get into a specific number—but I assume that now you’re spending more on AI tokens than SaaS, or is that still not true?
Chat: I think that’s probably true. Maybe it’s still even, but certainly from a growth perspective I would say we expect the token costs, or just AI costs overall, to dwarf SaaS.
Early on, as you bring new platforms in, we need Gong—that’s SaaS. But now we’re not really investing in Gong per se, because we have the licenses that we need. We’re using the transcripts to do a lot more on top of it. And then we may move to a different call-recording solution that’s free by next year, because most of the value is in our agent, not in the Gong SaaS platform.
John: So that’s interesting. Do you think that vibe coding and coding agents has made you second-guess spending additional capital on add-ons from your SaaS providers? If they want to upsell you on some AI feature, does that make you think, “Oh, okay, how long is it going to take us to code this ourselves?” Do you ever think that?
Chat: Yeah. We started with Retool as a good example—a way to almost create an intranet or an admin portal. Now we’ve realized with coding agents and Claude Code, we’re able to build those things to be much higher fidelity for our use case versus using the components Retool forces you into.
So Retool AI is not something we’re super interested in, because we can actually build something that’s much more tailored to our use case. A good example is spreading financials between P&Ls and tax returns. To build that in Retool is honestly very complicated because you’re pulling stuff from various different parts of the database, whereas we can vibe-code that in 30 minutes and it looks great. Our transaction advisors can say, “Hey, I wish this was different,” and we can change it. It just allows us to build our own custom software a lot quicker and better.
So yeah, there’s core infrastructure: having a CRM is maybe a foundational thing, plus outbound and call recording or some of these key insights. But once you have a core infrastructure for go-to-market or the transaction, we feel like we can build the rest with AI.
[00:43:45] Humans in the Loop, AI Judges and Data Reliability
John: Earlier there was something interesting you said, which is that after a process is turned or redesigned into being an AI-native process, you still have some things that humans need to do to stay in the loop. Can you tell me a little bit about that, especially in the context of, say, one flow? What are some things that humans are still doing that you think are hard to replace in the medium term, or permanently needed?
Chat: Yeah. I think there are a few use cases. For example, redacting financial records of addresses, Social Security, etc. We have found that there’s not an amazing, bulletproof, 100%-correct solution in AI yet. Do I expect that eventually we’ll have AI do it and a human review? Sure. But right now, we have a pretty good process that is not that expensive where a human is doing that. Longer term, we expect that to be automated.
I think the other main area we have humans in the loop is to check the work that the AI has done. Primarily, one, because it’s non-deterministic, it could get worse over time. The other piece is we’re changing so many parts of the funnel that maybe something changed further upstream that makes the AI’s answer worse. So we just want to have a quick check. Many times it’s like a five-second human-in-the-loop check, but it enables us to ensure that there’s quality all the way through.
Most of these are getting to a point where it’s one final human-in-the-loop check before it gets shipped out or whatever. That’s the goal. Hopefully over time it becomes a human reviewing the work of dozens or hundreds of agents and having an admin dashboard where they’re able to see almost like an AI judging other AI agents and saying, “We’re 99% confident this was done right. This one is 87%,” so that the human goes and looks at that and sees what the AI thinks may have been wrong.
John: So right now you’re checking the final output usually. And just for this audience, these would be things like the listing descriptions and the financials due diligence.
Chat: Financials, the adjustments that the system picked out of their financials. Is meals and entertainment, interest paid, depreciation—have those been found as add-backs? It’s a really quick way for a human to look at the two things—the PDF and what ends up in our system—and just validate that it didn’t miss anything, and then say “Correct,” or “This is what’s wrong,” which then makes the system better.
John: Would you say that AI—especially reasoning models—are good enough to process tabular data or even data inside spreadsheets for you to have it check whether certain numbers are being pulled out of spreadsheets? Would you say that’s reliable enough for you?
Chat: It’s reliable enough to pull; it still needs to be checked. Back to your earlier question about when we went AI-native—it used to be pretty bad. We didn’t have a lot of confidence in tabular data and CSVs and Excel files; it just did a bad job. You would upload a PDF, which is a P&L, and then ask it a question like, “What’s operating income?” and it would just make up a number. That doesn’t happen anymore. It’s actually pretty good at P&Ls and CSVs and Excel spreadsheets.
But, given the complexity of some of the Excel spreadsheets that we have—where there might be things on multiple tabs and it’s quarters or months over a year and multiple years—it makes sense to have a human check what ended up in the database. Over time, to my prior comment, it’ll be AI doing the work, a different AI—an AI judge—looking at the PDF and looking at the stuff in the database and making sure it matched, and then a final human to check. Because, obviously, we don’t want financials to be wrong.
John: I think a lot of people don’t realize that, especially with the Gemini OCR ability, it has really solved a lot of tabular data extraction. And I think a lot of companies are still paying a lot of money for enterprise OCR solutions, or enterprise data-extraction solutions—like ridiculous amounts. But I think the recent multimodal models really solved that. I think that’s not priced in.
But to your point, since we’re in the finance domain, even for compliance reasons, you may still need someone to stamp it.
Chat: Yeah.
John: But instead of doing a lot of intermediate-step checks, you just need the final thing. And then you might have LLM agents embedded within the process to keep checking. So that, I think…
Chat: Yeah. AI judges, we’ve found, are actually really good. You just make the chain. It’s almost like, imagine you had unlimited resources. You do one job, which is this agent; you do this other job, which is judge that agent. It actually enables you—even if we had contractors—to tell them to do the whole thing and give them five steps.
But now we can build AI agents that are the specific steps, because we know that it’s much easier to judge quality when someone has one thing to do. An AI agent has one thing to do; you trust them to do it a lot better.
[00:49:30] AI Search, AEO, Outbound and SMB AI Adoption
John: Switching to go-to-market and things like AI visibility: AEO is hot right now. We can do a rapid fire here. Are you bullish or bearish? Is this just SEO rebranded? And how is it affecting marketplaces? Obviously SEO is important for marketplaces; AEO might be too.
Chat: Yeah. It’s interesting—exactly to that point. I think listings, we’ve found really good results from AEO for more of a, “Hey, where’s a place that I can sell my business?”—more of these general questions.
I think it’s a little less clear for someone going into ChatGPT and asking, “I want an HVAC business with over half a million dollars in cash flow.” We’re still trying to see how anyone ranks, really. But I actually think we get a ton of folks that say that they’ve found us on ChatGPT, Gemini, Google. So we are very focused on performing well there.
John: How much do you think the mix has shifted from traditional SEO to AEO?
Chat: Maybe it’s still a minority, but I find that it’s a more considered minority, and the conversion on AEO traffic is much higher. If they’re actually doing that research versus clicking on a link, they’re generally almost thinking about the problem a bit differently. They see us as one of the two or three options that’s presented by Perplexity or ChatGPT, and then they come in and actually go all the way through the process and talk to us, et cetera.
So I think there’s a lot of legs on AEO. It’s unclear how the dynamic changes, but I think very quickly—especially with Google and AI Overviews and Gemini—a lot of people are starting to just naturally use AI to search. It feels like that’s where the puck is going. We’re focused a lot on that, because I think if you do AEO stuff, you get a lot of SEO benefits anyway. And if you’re doing SEO stuff, it’s going to help AEO. So we’re doing it all so that we are discoverable.
John: In terms of doing cold outbound, programmatic outbound—what are some things that you’ve learned? And are you building the whole stack internally yourself, or are you using popular tools like Clay or things like that?
Chat: Yeah. We use a bunch of popular tools: Clay, Apollo, Smartlead, Instantly. It’s like any and every tool. We’re trying to be good at the stuff we know we can be good at, which is who we want to target, how we want to target them, and the content that we’re sending out. That’s really where we focus virtually all of our time, as opposed to building a platform to actually send emails or do some of the outbound stuff at scale. So we just leverage a lot of tools, and Clay we’ve found works.
John: What are some of the things that you wouldn’t yet offload to full-blown AI SDRs? Or have you already automated the whole thing and don’t have any human in the loop?
Chat: Yeah, we have humans in the loop later in the process. For this, because it’s cold, we actually have to explain what Baton is and who we are. I think the AI SDR doesn’t do an amazing job yet of that.
If we were selling, I don’t know, ERP software, an AI SDR probably works fine because anyone getting pitched understands the product; they’re just hearing about some new vendor. I think we’re unique in that the people we’re outbounding to—saying, “Hey, are you interested in getting a valuation or selling your business?”—are like, “Who are you?” So I think it’s worth it, from our perspective, to ensure there’s a human in the loop.
We try to offload it and get as much of the conversation to be on AI as possible, but eventually we have humans stepping in when we see that the conversation gets stuck.
John: As the dissemination of these AI tools in the market increases, do you feel like all these growth channels and tactics will stop working or become too crowded? Or are you currently not seeing that so far?
Chat: Yeah, it’s a good question. I think naturally it will in crowded markets, or for folks that—as a startup CEO, I get pinged about a lot of stuff as a Series A startup CEO. It’s probably crowded for me; I have Superhuman and auto-archive turned on, so every pitch and every marketing email I’m not seeing. Sometimes I miss things, but I’ll go in there and look versus getting all of that in my inbox.
So I do think it gets crowded, and you have to have multiple channels: partnerships and referrals and outbound and LinkedIn, hitting people in multiple different ways. We think about it as a bit of retargeting: calling people that have expressed interest in addition to just emailing them, but also reaching out to them on LinkedIn—the traditional full-stack outreach.
John: Since you deal with SMBs a lot, where is SMB AI penetration at? Certainly not as much as the tech/VC land that we’re in. What would be your sense—is it 5%, 10%?
Chat: Yeah, I think it’s pretty low. I think people are using it to be more productive—to do research, to do simple tasks. I don’t think, from a process automation perspective—answering the phone or dealing with inbound or outbound requests as, say, an HVAC company or as a traditional business—it’s widely adopted yet.
We find that a lot of the folks we talk to, because they’re later in their careers, 20–30 years in, the amount that they want to invest to grow the business when they’re actually thinking about selling tends to be less. What we find is a lot of the buyers are acquiring a business with the stated intent that they’re going to leverage AI once they acquire it. Because they can say, “Actually, then I can bring in an AI outbound dialer. I can bring in an AI receptionist so that all these calls…”
One of my friends who bought a commercial HVAC company—he bought the company a year ago. He has started to implement a bunch of tech, something as simple as Slack for communication across all the team members, which they didn’t have before, all the way to getting AI into the platform. I feel like it’s this next generation that’s going to do a lot of that: they can take over a business that works and then layer in AI.
John: It seems like selling a call-answering AI agent to an HVAC company is the most popular AI startup idea. You obviously see… maybe not. So the question is…
Chat: Yeah—what’s the moat?
John: Yeah, what’s the moat? In terms of that specifically, because it is so popular, is it just that we think it’s a crowded market? Or is this still blue ocean outside of the SMB world?
Chat: I think there are so many SMBs that if you’re horizontal and vertical—like, if you want to do lawn care and you say that your AI agent is way better at lawn care because it’s trained on lawn care—you can probably go after all the lawn care folks and have a reasonable business. I don’t know if that’s a VC-exit business versus you can bootstrap that, because you can build an agent pretty efficiently. That’s a good question.
But I think the VC play would be something like ElevenLabs that can be integrated across 400 different industries and has legs in SMB and enterprise, and across all channels.
[00:58:02] Irreplaceable Human Roles, AI Economics and Baton’s Long-Term Vision
John: Since you’re running an AI-native firm at this point, you must have some opinions about what roles you definitely cannot do without—human roles in these AI-native types of companies. What are those roles? And do you really subscribe to the idea that there’s going to be a ton of 10-person unicorn companies out there?
In your context, it would be: can you grow 20x from here without adding headcount? Is that impossible? What views do you have on that?
Chat: Yeah. I think my take is, the thing that AI does not do right now is build and elicit trust. In our industry, you have buyers and sellers that are trusting the marketplace and the platform and the people that are engaging with one of the most important financial transactions of their life.
I think all of the associated support activities can be AI, but you want to connect with a human and have someone tell you, “This is the right offer for you and you should move forward.” I think those are these canonical folks that will exist.
I also think AI is good with the data it has. Humans that can see where the puck is going and what needs to change, and look ahead into the future—that’s a bit of strategy and operations—those are super valuable because I don’t think AI is great at that yet. Maybe it becomes great; you can imagine it can start to do a much better job because it can read all that stuff. So I think net-new ideas and solutions are still a place where we’re investing in humans.
To the 10-person unicorn question: I think for the right industry, the right market—Perplexity just raised at a billion-dollar valuation and they have, what, 50 people, or 100? Gamma has 50 people. I think in that archetype, where it’s PLG and some of that stuff, there will be 10-person unicorns.
For things where there’s more human involvement and you’re asking for support and help—whether it’s 10 people or 50 or 100—that’s different. I do think we will be able to scale and grow with much less headcount growth than we would have. Back to one of the first comments: we would have 10-plus, 20-plus SDRs here without AI. Our cost structure would just look different. We’d be raising a bunch of capital, we’d be growing into this market. Right now, we don’t have to do that. We can invest the time and energy in software to support our growth.
John: Let me ask you a hypothetical question. If Cursor—or whatever coding agent you’re using—5x’d the pricing and the industry just 5x’d, what would you be doing?
Chat: Nothing different. It’s totally worth it.
John: 5x?
Chat: You’d probably have to 100x it, maybe.
John: Wow. So you think that right now we’re in the three-dollar Uber-ride era of coding. Or I guess AI.
Chat: Yeah—the Max plan for Claude, which gives you Claude Code for a hundred dollars a month, is like a two-dollar Uber ride from here to JFK.
John: That’s the big debate right now. I think it’s ridiculous to even be having this debate, but a lot of people have it, so it’s good to have an actual CEO talk about that.
Chat: Yeah. You think of Cursor, or an engineer—the productivity on their time—and just making an engineer 10% more productive. Most engineers are like twice as productive with these tools. You don’t want that entire incremental gain to go to Cursor and the tokens and all this kind of stuff. But at $200 a month versus 20 bucks a month, that’s still a no-brainer. At $2,000 a month, it’s probably still a no-brainer. At $20,000 a month…
John: Speaking of which, as the amount of AI in your firm increases—like you said, from a customer-perception standpoint, trust is still important—how do you make sure you balance doing things really AI-natively, where eventually consumers just expect everything to be AI-generated, but they still need that trust? Especially for you guys, it’s important to be able to get you on the phone or something like that.
How are you making product decisions to maximize trust while maximizing AI?
Chat: Totally. I think fundamentally, people should be able to reach humans anytime. We want the humans available for our customers on either side of the marketplace as much as humanly possible. So it’s about taking away all the busy work.
I talk about it as giving them an Iron Man suit. Ultimately, the more we can do that, the more trust we build. Because I think today, if an owner or a buyer literally knew that we had an AI agent scheduling their buyer–seller meeting, I don’t think they actually care. If they couldn’t reach their advisor or their advisor was unavailable, they would get upset, right? Because they’re putting their time and energy into us to help that owner sell their business.
So I think we want to preserve human time for the other humans on the platform that we want to make a connection with and build that trust.
John: That’s great. So wrapping up: where do you see Baton in the middle of next year, and ultimately, after you create this new asset class, what are some further things you can see Baton growing into? What other adjacent markets do you see?
Chat: Yeah. In the next year or two, it’s really about becoming the de-facto, well-known place that small business owners and buyers can come to see all the listings and move really quickly through the process. It’s the single best place to buy or sell a small business. That’s the next year or two: we solve that brand promise that we are the best place to buy or sell a small business.
Ten years from now, or whenever, with the asset class, I think it gives us the ability to create financial products for both the small business owners that are better than the loans they have access to today, and for small businesses in general to have better access to capital.
I think buying and selling small businesses becomes a great path to entrepreneurship. There are financial products like a small-business ETF, where someone can invest in small businesses through Baton, and it’s actually helping solve the financing part. We are probably participating in some way, shape, or form in the financing, where we can underwrite deals better than SBA lenders can, at that level or above. Because this is a more deeply understood and legible asset class, you have a lot more capital from the financial markets coming in.
John: It almost seems like it’s useful to have a platform like Baton, or the marketplace, facilitate the loan underwriting and do this vertical integration, because you were saying that is the key risk factor in closing the deal and it adds all sorts of delays. So I think that kind of makes sense.
Chat: Yeah, maybe.
John: Yeah, maybe there’s some hidden reason why the industry hasn’t done that, but…
Chat: I think part of it is there just hasn’t been scale. Most business brokers that talk to an SBA lender are doing one deal a year. So they talk to their preferred lender, and that lender is doing a handful of deals a year. There’s no need to think about this at scale.
Whereas we will, in 10 years, have tens of thousands of businesses selling on Baton, which then becomes something that is material. Then the ability for us to underwrite 10 billion, 20 billion, 100 billion dollars worth of GMV—the financing part of that is a key part of the equation.
I think until we are at a billion or two billion in GMV, that’s probably not something we need to focus on. But at that point, we have enough volume where we can actually say, “Okay, if we can make this process 50% better, or 2–5x quicker, whatever the value prop is, then we end up being able to speed up more transactions and say, ‘Okay, we’re going to go from 20,000 to 50,000 transactions in a year by being able to speed this process up.’”
John: There’s a ton of startups, especially, that are benefiting so much from AI. And I think incumbents… It’s just crazy in the sense that I think incumbents know that they could do this, but they don’t have the financial incentive to do it.
Chat: Yeah.
John: For now. You’re not getting paid to 5x your productivity or whatever if you’re running, say, a customer support org, unless your incentives are aligned.
Chat: Yeah, totally.
John: It’s like lawyers, right? They get paid by the hour. Do they want to do stuff in five minutes versus 50 minutes? They’re actually negatively incentivized.
Chat: Yeah. My neighbor is a partner at a law firm and he made the comment, when he was over for dinner, that he’s shifting to flat fee. They’re a big M&A firm.
John: Oh, wow.
Chat: Because then they’re incentivized along with the customers. They’re like, “Okay, it’s 10K or 100K,” or whatever it is. We’ve started using Crosby Legal, and it’s like a hundred bucks for them to review and redline an NDA. It doesn’t matter if it goes back and forth 10 times; it’s still a hundred bucks. It’s way cheaper than me going to our corporate counsel for a partnership.
John: So they’re doing some AI stuff with human in the loop, right?
Chat: Yeah, totally.
John: But even then, a hundred bucks is aggressive because…
Chat: Yeah. Eventually it’ll go down.
John: Yeah.
Chat: A hundred dollars… that’s my problem with AI-enabled services, because sometimes the human in the loop gets very tricky, especially if you have to be accountable for the result, where the result has some financial…
John: Yeah, there’s some legal or financial…
Chat: Yeah. Then it’s sometimes hard to compress your cost structure down too much.
John: Yeah.
Chat: Even for us, with some of the contractors that we’ve had in the past, they were getting paid by the hour. Then we said, “Actually, this is insane. We’re going to pay you per unit of work,” which was, for example, getting the video interview ready to be on the listing.
John: Yeah.
Chat: That’s 20 bucks; it’s whatever. And then it’s just done. They’re actually incentivized to do it as quickly as possible, versus spending two hours getting the video together and it costs us $40 when it doesn’t really need to.
John: That concludes this podcast episode. But thanks so much. I learned a lot, and I’m sure my audience will also have a ton of comments.
Chat: This was great. This was fun. Yeah. Yeah, this was great. I think all the questions were on point.
