What AI actually costs a bank
Drawn from the public record of AI at US banks: what a use case demands to build, to run, and to be allowed to run, and why so much of the bill never reaches the business case.
Banks Bring More Rigor to a Loan File Than to Their AI
Every large commitment in a bank already survives an evidence standard. AI is the one that arrived faster than the standard could catch it. Four questions close the gap, and they belong to the CFO.
Read →Your First AI Use Case Is Buying Your Program
The most expensive model a bank will ever run is likely its first. The reason is an accounting choice, not the technology.
Read →Regulation Follows the Decision, Not the Technology
A frontier AI model drafting emails is not a regulatory event. A decades-old model denying mortgages is. The rules were filed under the decision, and they concentrate along what it can harm.
Read →The Only AI Cost That Gets Cheaper
Token sticker shock is real. It is also the only part of a bank's AI cost base riding a deflation curve. The ledger below is unmetered, priced in labor, and compounding.
Read →