How to prepare for a supervisory visit with an AI agent

Most of what a supervisor will ask on a visit is predictable from what you have already told them. The agent reads your past correspondence, returns and open actions, builds an evidence index for each likely topic, finds inconsistencies and drafts briefing notes for interviewees. Those people own every answer, and a named senior manager signs off the preparation.

What is a supervisory visit, and who owns preparation?

A supervisory visit, sometimes called an on-site review or inspection, is when a regulator's team examines a firm in person or remotely: interviewing senior managers, reviewing files, testing processes and following up earlier findings. Visits are announced with a scope and a document request, and followed by findings that may lead to formal action. The firm's preparation decides much of how the visit goes.

Regulatory affairs or compliance coordinates preparation, with a senior manager accountable for the relationship. Subject experts and interviewees prepare their own areas. The agent sits behind the coordinator. Its role is close to helping prepare for a regulatory inspection, applied to the full cycle from notice to findings.

What does the agent prepare, step by step?

1. Person, coordinator: logs the visit notice, the scope and the document request.
2. Agent, depth 1: reads the supervisory history. Past correspondence, findings, attestations, commitments and regulatory returns relevant to the scope.
3. Agent, depth 1: lists open commitments. Anything the firm promised the supervisor, with status and evidence.
4. Agent, depth 1: builds the evidence index. For each topic in scope, the policies, procedures, monitoring results, management information and sample files that show how the firm meets its obligations.
5. Agent, depth 1: finds inconsistencies. Figures, dates or statements that differ between documents, or between documents and what the firm has told the supervisor before.
6. Agent, depth 2: drafts briefing notes for each interviewee: the topics likely to come up, what the firm has said before, the evidence, known weaknesses and open actions. Not scripts.
7. Agent, depth 2: drafts responses to the document request, with the evidence index. The coordinator reviews, legal checks, and a person submits.
8. Person, interviewees: prepare with their notes and the subject experts.
9. Person, accountable senior manager: signs off the preparation and the document submission.
10. Agent, depth 2: drafts the action plan after findings are received. Owners commit to it.

Which controls apply, and what evidence does each leave?

Briefings, not scripts. Interviewees must answer in their own words from their own knowledge. Coached, scripted answers are noticed, and they undermine the individual accountability that regimes such as the UK's Senior Managers and Certification Regime exist to create. Evidence: briefing notes that set out facts and history, not answers.

Commitments checked first. A promise to the supervisor that has not been kept is the most damaging thing to discover during a visit. Evidence: the commitments list, with evidence of completion or an honest status.

Inconsistencies resolved before submission. Evidence: the consistency report and how each item was resolved.

Submission signed by a person. Evidence: sign-off by the accountable senior manager.

Known weaknesses disclosed openly. If the agent finds a gap the firm has not told the supervisor about, it goes to the senior manager, who decides how to raise it. Evidence: the decision.

What should you measure?

• Inconsistencies found before the visit, and any found by the supervisor.
• Document request items delivered on time.
• Open commitments at the date of the visit.
• Findings that relate to issues already known internally. These show whether the firm is candid with its supervisor.
• Action plan items closed on time after the visit.

Where does it usually go wrong?

Preparing a performance. The point of preparation is to know your own position and evidence it, not to present a better picture than exists. Agents that polish drafts can make this worse; keep briefings factual.

Sample files not checked. Supervisors pick files, often from areas such as AML or customer exits. If the sample files are weak, the policies do not matter. The agent can pre-check a sample against your standards; see preparing AML case investigations and preparing customer exit decisions for what good files contain.

Agents not in scope of preparation. If AI agents run parts of processes in scope, the supervisor may ask how they are governed. Be ready to show their bounds, certifications and records, and to name the person accountable for each agent.

Document requests answered too broadly. Under time pressure, teams send everything that might be relevant. The supervisor then reads material the firm has not reviewed, including drafts and internal debate. The agent should map each document to the specific request item it answers, and the coordinator should challenge anything that answers none.

Findings filed and forgotten. The visit ends, the action plan is drafted, and actions drift. The agent should track them to closure, ask for evidence when an owner marks one complete, and report progress to the committee until every finding is closed.

The stance on compliance agents overall is in AI compliance agents.