David Haber at a16z: In Vertical AI, Market Structure Is the New TAM
Why VC needs to rethink market size — and get smarter on market shape
This week on Verticals:
Why market structure is a better predictor than TAM in Vertical AI
What a16z looks for at Series A
The “messy inbox” thesis — why comms make great wedges
How AI services businesses are hitting software-level margins
Why traditional systems of record may go to zero
Trailer below — full episode on YouTube.
If there is a foundational thought in VC, it’s TAM (Total Addressable Market size). What was true in 1970 is mostly still true today: the bigger the number, the easier the pitch. But in Vertical AI, filtering on TAM alone is a mistake — here’s why.
The better question than simply “how big is this market” — or at least the question more investors should be asking in the current AI era — probes market structure. How concentrated or fragmented is the incumbent system of record? How does the software or service market you’re addressing currently monetize? What is the margin profile of your customers and what matters most to them: top of funnel, conversion, profitability?
On this week’s episode of Verticals, David Haber joins us to map out why TAM is losing its luster and why we’re seeing this meaningful departure from orthodoxy now. Haber co-leads a16z’s AI apps fund and sits on the boards of Vertical AI startups including Camber, Tennr, and Eve — three companies attacking vertical workflows in healthcare and legal.
David is no stranger to vertical business strategy. Earlier in his career, he founded Bond Street — a fintech company that aimed to reinvent small business lending — and sold it to Goldman Sachs in 2017. He then spent several years inside Goldman, watching how a 45K-person institution actually works. Even at one of the most technologically progressive firms in the world, he saw “expensive humans living in Excel… [not just] to do modeling, but using Excel to track work.” The software budget was a rounding error, but the cost of the work — be that labor spend or services to support them — was astronomical. That learning would come to inform a lot of his thinking as moved to the investor side — joining a16z in 2021 — and ever since.
Here’s a preview of the logic. When the system of record (SoR) in a given vertical is fragmented — think the long tail of EHRs in specialty healthcare — there’s space for AI-native companies to start by doing the work surrounding those systems and, over time, back into becoming the system of record itself. Camber is doing exactly this, with an initial focus on behavioral health: they wedged into revenue cycle management, automated the billing workflow, and turned a teens-gross-margin services business into one with more software-like fundamentals
Contrast that with a market where one player owns 50%+ share — say, Autodesk’s Revit in architecture, engineering, and construction. In a recent a16z thesis piece, Haber and his colleagues lay out three attack vectors against Revit’s dominance:
Rebuild it head-on
Build around it
Capture the services spend on top of it
But even they acknowledge the direct assault is the graveyard. When the SoR is that entrenched — taught in every school, embedded in every firm’s workflow — startups are forced into a complement position, building around the incumbent rather than replacing it. And that introduces a nasty dynamic: the more valuable you become, the more the incumbent wants to either tax you, build your feature in-house, or just flat-out strangle your business. As we explored deeply in our prior essay on disrupting incumbents, there are ways to fight Goliath — but not all battles are winnable.
This episode is brought to you by Parafin — the embedded capital platform powering financial products for platforms like DoorDash, Amazon, and Worldpay. If you have SMB customers, they need financing — why shouldn’t you be the one to provide it? Learn more →
The point is that different market shapes require different startup strategies. In concentrated incumbent markets, stay complementary as long as possible — don’t pick a fight you’ll lose while thinly capitalized. In fragmented markets, go deeper faster and own the end-to-end workflow, because there’s no single incumbent to retaliate. Eve, Haber’s plaintiff law investment, exemplifies the latter playbook: they own intake through outcome. The non-public settlement data they abstract along the way — what a specific insurer pays for a specific injury in a specific state — creates a compounding vertical data moat that no foundation model can replicate.
This matters for founders choosing where to build. TAM slides get funded; market structure analysis gets built. A $600B nurse payroll TAM means nothing if Epic owns the data you need to build a valuable product, accelerate distribution, and accumulate a data moat. A $2B niche with fifteen fragmented incumbents and heavy BPO spend — assuming there are some credible lanes of extensibility — has a much stronger profile for building category-defining Vertical AI.
So investors: over-pivot on TAM at your peril. And founders, consider whether the shape of your target market plays into your hands… or makes success an uphill battle.
Subscribe to Verticals to get new episodes every week, available wherever you watch or listen.