Question

A financial services company deploys a Copilot Studio agent that helps loan officers evaluate credit applications. The agent surfaces a recommendation ('approve' or 'refer for review') based on a custom model plus grounding data. During a compliance review, regulators demand that every recommendation be accompanied by the specific factors that influenced it, so that adverse decisions can be justified to applicants. Which responsible AI design measure most directly satisfies this requirement?

Accepted Answer

Configure the agent to return feature-level reason codes/explanations alongside each recommendation and log them for audit

Answer

Enable content safety filters on the agent to block harmful or biased language in the model output

Answer

Increase the model temperature so the agent generates more varied, natural-language rationales

Answer

Route all adverse recommendations through a human reviewer before they are shown to the applicant