ArgumentCompany insightArtificial intelligenceBankingTechnology
The vertical AI revolution belongs to firms with proprietary data, not to the tool makers
Rich Fairbank · Capital One's 2024 letter from Rich Fairbank · 2025
The idea
Capital One's 2024 letter splits the effect of artificial intelligence in two. The horizontal revolution is off-the-shelf tools used across every industry, and its gains are shared by everyone. The vertical revolution, inside each industry, Fairbank expects to be driven not by the technology companies but by firms with deep proprietary data about their own customers and their own risks. For a lender that means decades of credit outcomes. The letter also describes credit discipline as zigging while the market zags, flagging score inflation in 2022 and pulling back when others did not. It is a frame for thinking about which banks a technology shift favours.
In any industry, ask who holds the data the models need; the shift favours them more than the vendors of the models.
When it works
Bank and fintech coverage, the question of who wins from AI in finance, and any moat built on data.
When it may fail
Proprietary data can be made less valuable by better public data or by regulation, and the letter is a chief executive describing his own firm's advantage.
How to use it
For a company claiming an AI advantage, ask what data it holds that competitors cannot buy, how far back it goes, and whether it includes outcomes rather than just activity. A data moat with no outcomes is a log file.
Where it came from
Letter Capital One's 2024 letter from Rich Fairbank · Rich Fairbank
What the source says. Fairbank splits AI's effect into a horizontal revolution of off-the-shelf tools and a vertical one driven by companies with deep proprietary data in their own industry, and describes Capital One flagging credit score inflation in 2022 and pulling back in some segments.
Connected ideas
Says something similar
What banks actually deploy is narrower and more governed than the consumer conversationInstitutional AI is scoped, governed and data-limited; expect to work inside those limits and to be able to explain them.
Says something similar
A model trained on a biased outcome will reproduce it at scaleA model predicts the past decisions in its training data; if those were biased, the model is a bias machine with a straight face.
Concepts
Put it to work on L3VLUP
More like this
- Company insightArtificial intelligenceLetter located
Inference has overtaken training, and the flywheel runs on tokens
A compute vendor's case is a flywheel from agents to tokens to compute; the buyer's case is the return on that compute, and both have to be read together.
From NVIDIA's fiscal 2026 letter from Jensen Huang · Jensen Huang
- ArgumentArtificial intelligenceVideo located
Which parts of the analyst job compress, and which get more valuable
Prepare for the parts of the job the tools make scarce, judgement and client sense, rather than the parts they make cheap.
From Goldman Sachs CEO David Solomon on AI, Debt and America’s Future · Prof G Markets
- FrameworkBankingFilm
A bank run is a maturity mismatch, not a solvency problem
A bank fails when its liabilities come due faster than its assets, which can happen to a solvent bank.
From It's a Wonderful Life · 1946
- FrameworkArtificial intelligenceVideo located
A model never sees words, only tokens, which explains most of its odd behaviour
Reason about a model in tokens: that is what it reads, what fills its window and what you pay for.
From Most Devs Don’t Understand How LLM Tokens Work · Matt Pocock