Model Validator

The work is adversarial by design: reproduce the model, test it on data it has not seen, probe where its assumptions break, and write the limitations down in a way that survives a regulator reading it. Credit, capital, pricing, provisioning and increasingly machine-learning models all pass through here.

Statistics, mathematics, actuarial and data science backgrounds, and analysts moving across from credit risk or finance.

These are the shapes the round takes, not a question bank — banks phrase them differently and the wording is never the point.

What the panel is listening for. Independence with evidence — a validator who cannot say precisely what is wrong is as unhelpful as one who signs everything.

A credit model is a claim about how borrowers behave; you cannot judge whether the claim is reasonable without knowing what a real file looks like, which is why validators who understand lending are the ones who get listened to.

That is the general case as well as this one. A bank's technology is not ordinary technology and its finance is not ordinary finance, and the second and third lines of defence — risk, compliance, financial crime, model validation and internal audit — recruit heavily out of the functions, taking the people who can already read a business. The case drills are free and are the cheapest way to build that, whichever column you sit in.