The most common question I get from investors looking at NaS_OS is some version of: “Why not go after the global banks? That’s where the budgets are.” It is a reasonable question, and the short answer is that the global banks already have AI tools and are paying enterprise prices for them. The longer answer is that the bottom tier of the M&A market is where category-defining companies tend to start, and there are structural reasons I think that pattern repeats here.
I want to walk through that reasoning, because the question of which tier of the market to serve first is the most consequential strategic decision a vertical AI company can make. And the conventional answer, go where the budgets are, is often wrong.
What Stripe and Square got right
Stripe did not start by selling to enterprise CTOs at Fortune 500 banks. They started by making it possible for a developer to take payments with seven lines of code, and they grew up through the long tail of small online merchants and indie SaaS companies. By the time they were ready to land enterprise contracts, they had a product so good that the enterprise sales motion was almost beside the point; the developers inside the enterprises had already heard of them.
Square did the same thing in physical retail. They didn’t pitch the chains. They put a $10 card reader in the hands of a coffee cart owner in San Francisco, then a hairdresser in Brooklyn, then a few thousand small merchants who had never accepted cards before. The chains came later, and on Square’s terms.
The pattern is not “start small and grow up.” The pattern is “find the segment whose problem is most acute, whose decision-making is fastest, and whose feedback loop is tightest, and build there until the product is dominant.” For Stripe, that was developers. For Square, that was small merchants. For an AI tool in M&A, I think it is boutiques.
The structural advantages of boutiques as a starting customer
A 15-person boutique advisory firm does roughly 8-30 mandates a year. The managing partner usually founded the firm and signs every check. The senior associates and analysts are doing the actual work and feel the pain points daily. The feedback loop from “we tried it” to “we know if it works” is days, not quarters.
Compare that to a bulge bracket. A new tool has to go through procurement, security review, model risk management, the AI governance committee, the technology architecture group, and a pilot program with a single team that may or may not generalize. The decision cycle is 9-18 months. The feedback loop from product to user is mediated by multiple layers of internal stakeholders.
If you are building a product that needs to evolve quickly with customer feedback, and any AI product does, selling to bulge brackets first is a structural mistake. You will iterate too slowly to get the product right.
Boutiques also have better economics for the early customer cohort. The contract values are smaller, but the gross margin is higher because the cost to serve is much lower. No custom deployment. No dedicated AI strategist. No 6-month onboarding. The boutique signs up, the product works within a week, the firm uses it on the next mandate, and the founder calls me directly when something is wrong.
The structural disadvantages of selling to global IBs
The cost of an enterprise sales motion is not just the long cycle. It is what the cycle does to the product. Enterprise customers ask for features that serve their internal procurement requirements: SSO with their specific identity provider, deployment in their specific cloud region, customization for their specific compliance framework. Each ask is reasonable in isolation. Cumulatively, they pull the product toward serving the procurement department rather than the user.
I have watched this happen to other vertical AI companies. The product launches with a clear point of view, gets pulled in five directions by the first three enterprise customers, and ends up as a worse product for everyone. The boutique tier doesn’t ask for those features because they don’t have the procurement infrastructure to require them. They just want the work to get done.
There is also a positioning problem. If you sell to bulge brackets first, your product is shaped around their workflows. Those workflows are different from boutique workflows in ways that matter: the IM templates are different, the equity story patterns are different, the data room conventions are different, the financial model structures are different. A product built for Goldman Sachs is not the same product as one built for a 12-person boutique in Madrid or Chicago.
What this means for NaS_OS
We are building NaS_OS for boutiques first because that is where the product can be best, fastest. The companies that already raised hundreds of millions of dollars to build AI for the top tier (Rogo, Hebbia, Harvey) are doing important work, and they are well-suited to that segment. But the layer below them, the 5,000+ boutique firms doing the actual mid-market deals around the world, is structurally underserved and is exactly where I want to build.
There is a version of the future where NaS_OS eventually moves upmarket. There is also a version where we stay focused on the boutique tier indefinitely because that segment alone is large enough to build a real company. I don’t have a strong view on which one is correct yet, and I think founders who have strong views about their five-year roadmap before they have product-market fit are usually wrong about the five-year roadmap.
What I do have a strong view on is the right place to start. It is the place where the customer feels the pain most acutely, makes decisions fastest, and gives you the cleanest feedback. For AI in M&A, that is the boutique tier. The banks can wait.