A concept for governed AI email drafting inside regulated banks — human-in-the-loop review, policy gating, and audit-ready records before any AI-drafted message reaches a customer. This paper describes the interaction design already validated end to end, and the governance engine planned to sit behind it.
Banks want the productivity of AI-drafted email, but an unreviewed AI send creates conduct, compliance, and reputational risk the moment it reaches a customer. Most AI email tooling is built for speed, not for the audit trail and human sign-off a regulated institution actually needs before a draft goes out.
Email Copilot is designed around a single rule: an AI draft is a draft until a human approves it. The full review workflow — a human-in-the-loop queue, policy and confidence-based blocking, an audit and accountability view, and an executive command view — has been designed and validated as a complete interaction flow, so the governance model can be evaluated and agreed before the drafting engine itself is wired in.
A fully interactive prototype of the review workflow exists, with every screen navigable end to end:
The workflow above is designed as the front end for a governance engine that is planned as, not yet built:
React + Vite client with Firebase Auth. Stage: Tech Ready — the governance workflow and UI are fully designed and demonstrable; the AI drafting engine, policy enforcement, and audit persistence behind it are the next build phase.