AI hiring in financial services is being shaped by regulators, not technologists. Here are the four roles every bank should be hiring for now, the three that look important but aren't, and the question that separates real AI talent from impressive resumes.
Most banks still hire for AI the way they hired for software in 2015: find the smartest engineers and let them build. That worked for software. It fails for AI in regulated finance, for one structural reason: this work is shaped by regulators, not technologists, and the people who can navigate regulators are rarely the people who train transformers.
SR 11-7, the OCC’s third-party rules, and EU AI Act provisions that reach any US bank with European operations all tightened around one theme: AI in customer-affecting decisions must be governable, auditable, and explainable. Banks that don’t hire for that gap learn it the hard way: a consent order, or a model pulled mid-quarter.
None of these are buzzword titles. All are hireable today, and most are deeply undersupplied.
What they doValidate, challenge, and govern every model that affects a customer or the balance sheet.
Why it matters nowSR 11-7 was written for traditional models; regulators now apply it to LLMs and gradient-boosted systems where the validation method has to be invented case by case. A senior MRM who has actually built and validated AI models is the most valuable human in your AI stack.
Red flag“Model risk experts” who have only validated linear and logistic regression. Ask them to walk through their last validation of a non-linear model.
What they doOwn the policy framework, the model inventory, the approval workflows, and the relationships with regulators and internal audit.
Why it’s separate from Model RiskGovernance is process work; validation is technical work. The same person rarely does both well, and pretending otherwise gives you incomplete validation defended by incomplete process.
What good looks likeSomeone who has shepherded a consumer-facing AI deployment through a federal regulator’s review and survived. That experience is the credential, not the policy degree.
What they doBuild and maintain the testing that surfaces fairness drift in production, run the audits, and set the thresholds that trigger retraining.
Why it’s undersuppliedMost banks bolted fairness auditing onto data-science teams as a 20% responsibility. It doesn’t work. Auditing needs sustained focus on the same models for months. It is its own discipline.
ToolingAequitas, Fairlearn, AIF360, and the proprietary equivalents.
What they doStand up the deployment, monitoring, and rollback infrastructure that makes AI in production non-terrifying.
Why “regulated” is non-negotiableThe standard playbook assumes you roll forward fast and watch. In a bank you can’t: every deployment passes a change-control board, every rollback is an incident report, and every alert must be meaningful yet quiet enough that people don’t ignore it.
It’s the single best interview question I know for AI talent in BFSI. Every candidate worth hiring has a story, and the story tells you what you need to know.
Prefer something to print or forward? The four roles, the three to skip, and the interview question, condensed into a one-page brief built to hand to a hiring manager.
The kind of writing on workforce, AI, and enterprise hiring you'd actually want to read on a Sunday morning. No vendor pitches, no ad copy, no fluff.