Vertical AI Opportunity Finder
Build where the winning data is locked behind a company or industry's walls — foundation models won't go there.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 3
- Confidence
- 85%
Foundation-model companies can't and won't build most AI products, because there are far more smart people outside their walls than inside, and most of the world's valuable data is private — behind the walls of specific companies, industries, and governments. The startup opportunity is to take a broad smart model and tailor it with proprietary, use-case-specific data (often via fine-tuning) to beat the state of the art in a specific vertical.
Origin
Kevin Weil (CPO, OpenAI); invokes Ev Williams' maxim from Twitter that there are always more smart people outside your walls than inside.
Core principles
- 01No matter how big a company gets, there are more smart people outside its walls than inside
- 02Foundation-model companies don't want to and couldn't build most vertical AI products
- 03The models are broadly smart, but most valuable data is private and not in their training set
- 04Advantage comes from proprietary, industry- or company-specific data plus fine-tuning
How to run it
- 1
Target a vertical with private, valuable data
Find an industry or use case whose defining data or process lives behind company, industry, or government walls and is absent from public model training sets.
Pro tip The more the winning knowledge is locked away, the safer the space from foundation-model incursion.
- 2
Confirm foundation models won't chase it
Verify the space is too specific or too small for a general model company to prioritize — they lack the people, knowhow, and incentive.
Watch out Purely generic use cases a broad model already handles well are the ones most exposed to being squashed.
- 3
Tailor a broad model with your proprietary data
Take a strong general model and fine-tune / customize it on your use-case-specific data so it beats the state of the art on that vertical's tasks.
Pro tip Measure the gains with custom evals built for the vertical.
In the wild
OpenAI leans on its API and 3M+ developers precisely because it knows it can't build the immense number of AI products every industry needs. Weil frames the private, industry-specific nature of most valuable data as the reason vertical opportunities exist and are defensible.
→ A stated conviction that immense opportunities exist in every industry and vertical for outside builders.
Common mistakes
Building a generic wrapper a foundation model already covers
If your use case relies only on public knowledge the broad model already handles, you're most exposed to being squashed.
Ignoring the proprietary-data moat
Not leveraging private, use-case-specific data (and fine-tuning) forfeits the main defensible advantage against general models.
Is it for you?
Best for
Founders deciding whether and where to build an AI startup without being steamrolled by foundation-model companies
Not ideal for
Ideas whose entire value is generic capability already served well by a general-purpose model API
From the transcript
“there are way more smart people outside your walls than there are inside your walls”
“most of the world's data knowledge process is is not public.”
“there are immense opportunities in every industry and every vertical in the world to go build AI based products”
From the episode
OpenAI’s CPO on how AI changes must-have skills, moats, coding, startup playbooks, more
Kevin Weil (CPO at OpenAI, ex-Instagram, Twitter)