The Model Is the Bank
What happened when financial institutions put language models into production — in underwriting, compliance, fraud, and the call centre.
Read the issue
In this issue
Underwriting After the Language Model
Lena VogelCredit decisions are being made on documents no human will ever read.
What an LLM Can and Cannot Do in a Bank
Devin RaoA practical boundary, drawn from two years of deployments.
The Compliance Copilot
Nadia FaroukThe most boring AI use case in finance is also the most profitable.
Synthetic Data and the Model Risk Committee
Sarah LindqvistHow do you validate a model trained on data that never happened?
Fraud Models That Explain Themselves
Hannah ColeA decline you cannot justify is a customer you are about to lose.
Interview: Building an AI Risk Function from Scratch
Priya MenonShe was employee one. Eighteen months later the function has thirty people.
Customer Support Was the Easy Part
Omar HaddadDeflection rates went up. Then the hard conversations were all that was left.
Vector Search Meets the Transaction Ledger
Ravi IyerSemantic retrieval over financial records is useful, and quietly dangerous.
The Cost Curve Nobody Budgeted For
Tomas AlvarezInference is cheap per call. Nobody modelled the number of calls.
Model Drift in a Credit Cycle
Maya OkonkwoEvery credit model is trained on a period that has already ended.