Applied AI in Regulated Markets

Most AI demos die on contact with a real P&L. I spend my days on the other side of that line.

At Verizon, I work on the commercial and deployment strategy for applied AI inside a large, regulated business. The interesting question is rarely whether a system is technically impressive — it usually is. The question is whether it is worth moving into live operations, and what it takes to keep it there. That answer almost never turns on the model. It turns on the workflow the system has to live inside, the incentives of the people who actually use it, the failure modes a regulator will ask about, and whether the economics still hold once the novelty is gone.

The demo is the easy ten percent. The ninety percent is everything that happens after: the integration that no one wants to own, the edge cases that surface only at volume, the executive who has to defend the cost line, the compliance review that asks what happens when the system is wrong. Most of the value, and almost all of the failure, lives in that ninety percent. Learning to see it early — to tell the difference between a capability and a deployable system — is most of the job.

That is also how I invest. Through Fervor Capital, where I am Founder and General Partner, I back founders building vertical AI for regulated, complex industries: financial services, healthcare and life sciences, enterprise and industrial software, and climate and sustainability. These are the markets where the hard parts I deal with as an operator — compliance, workflow, trust, real unit economics — are the moat rather than the obstacle. Domain depth compounds. A thin wrapper over a general model does not.

When I evaluate a company, I am not asking whether the technology is good. I am asking two questions I would have to answer about my own work. The first: is it worth building — does the problem matter enough, and does the founder have the conviction to stay in a hard, regulated industry long after the novelty is gone? That conviction is what carries a team through the unglamorous ninety percent. The second: does it survive contact with a real P&L? Who has to change how they work for it to succeed, and why would they? What does it cost to run at scale, not in a pilot? What breaks when it is audited? Founders who can answer both tend to be the ones building something that lasts.

I have spent around thirteen years in and around large technology and regulated businesses — Verizon, Oracle, Cigna, AT&T, T-Mobile, and TIAA — and earlier built an early-stage fintech of my own. The throughline has not changed: the distance between something that works in a deck and something that works in production is where the real work is. I have learned to live in that distance, from both sides of the table.

If you are building applied AI for a regulated market, I would like to hear from you.