Banks are embedding AI into model risk management, using automation to improve inventory, monitoring, and governance while maintaining strict oversight. Suresh Sankaran emphasizes that transparency, traceability, and human accountability will define succe
Banks
are moving beyond experimentation with artificial intelligence and into
practical deployment, as model risk teams look to harness automation without
compromising governance, according to a senior risk manager
Speaking
ahead of Advanced Model Risk Europe, Suresh Sankaran, Prudential Regulation
Lead at NatWest Group, said the most immediate value of AI lies not in
replacing human oversight but in addressing long-standing operational
weaknesses in model inventory management and portfolio-wide risk visibility.
“The most immediate value is in
inventory hygiene and evidence linking,” he said, highlighting how AI can
classify model artifacts, extract metadata, and connect models to risk themes
in a dynamic and structured way.
He noted that traditional model
inventory processes often fail because they either become outdated or overly
procedural.
AI, by contrast, can “do the dull but
essential reading at scale,” enabling teams to focus on judgment rather than
manual documentation tasks.
This shift allows institutions to
maintain a living view of their model landscape, including purpose, ownership,
and limitations.
At the portfolio level, AI is proving
valuable in identifying concentrations and shared risks across models.
Sankaran explained that clustering
models by common drivers, data dependencies, or regulatory touchpoints allows
firms to better understand systemic exposures.
However, he stressed that governance
remains central. “The objective is not to pretend AI can replace governance. It
is to let governance see the full landscape,” he said.
The challenge for banks is to balance
efficiency gains with regulatory expectations around transparency and
accountability.
Sankaran emphasized that automation
must be deployed as a controlled assistant rather than an autonomous
decision-maker.
“We are not outsourcing
accountability to a robot. We are outsourcing the first draft,” he said,
underscoring the importance of human oversight in all critical processes.
To maintain regulatory confidence,
Sankaran outlined a framework built on traceability, governance, and control.
AI outputs must be fully auditable, with clear links to underlying data and
documentation.
Workflow gating ensures that no
automated output becomes official without approval, while versioning and change
management treat updates to models and prompts as controlled changes.
Independent challenge also remains
essential. Even where AI generates initial analysis, second-line oversight must
continue to scrutinize outputs and validate conclusions.
Sankaran noted that regulators are
less concerned with efficiency itself than with the ability to verify how
decisions are made.
