Talking about models in the regulator's language - an AI-forward bid that finally sounded regulator-ready

The buyer was a UK asset manager procuring a technology platform that involved model-based decision support. The firm bidding for the work had a strong technical proposition and a weak model risk narrative. Two previous bids to related buyers had cleared the technical read and missed on the second line of defence read - in both cases the feedback described the model risk position as generic AI marketing language.

The response described the underlying models as AI systems and the risk position as robust. Neither term appears in the regulator's model risk supervisory vocabulary. The second line of defence read the response and correctly concluded that the firm did not routinely think about its models in the language the sector expects.

The engagement rewrote the model section against a fixed template: each model named, tier assigned under the firm's own model risk framework, validation status stated, monitoring cadence stated, human-in-the-loop position stated. Where the firm did not have a mature answer for a field, the response said so and stated the compensating control. Nothing was left as robust.

The governed advantage on this engagement was the refusal to accept vague answers on fields the sector had specific vocabulary for. Every model in the response carried a named accountable owner, a signed source for its validation status, and an audit-trail entry recording the last review. The response was legible to a chief risk officer without a translation layer.

The bid was won. The buyer's chief risk officer's post-award note flagged the model risk narrative as the deciding factor on the second-line read. The firm has since standardised the template across its financial services bid pipeline.