PrymeIntelligence

Agent governance·Template

Board paper on AI agent deployment template

A board paper on AI agent deployment template: the sections a board needs to approve an agent, how to fill each for authority and risk, and an example.

A board paper on AI agent deployment asks the board, or its risk committee, to approve an agent and records what it is accepting: the business purpose, the authority the agent will hold, the actions kept for people, residual risk against appetite, monitoring and accountability. The template and a filled example follow.

What does a board need to decide?

Not how the agent works. A board decides whether the firm should give a piece of software authority to act on its behalf, how much, under what controls, and who answers for it. The paper should be built around those questions, with technology confined to an appendix. The questions directors tend to ask are covered in what a board should ask about AI agents.

Most deployments do not need board approval individually. Use this paper when an agent crosses a threshold the board has reserved: customer-facing, moves money, makes or prepares decisions about individuals, or falls into a high-risk category under a law that applies to you.

When should you use it?

Before the go-live of any agent above the board's threshold, and before widening an existing one past it. The operational detail sits in the AI agent deployment plan; the board paper summarises it and asks for a decision.

After approval, the same structure works for regular reporting. Reporting on AI agent risk to the board covers the ongoing version.

How do you fill in each section?

Decision required goes first, in one sentence. Directors should know what they are being asked before they read why.

Authority requested is a short table of capabilities and depths. Avoid technical names; describe actions.

What stays with people is as important as what the agent does. List it explicitly.

Risk and appetite gives residual risk for each capability and cites the appetite line. Do not bury risks the second line flagged.

Evidence summarises testing: how many of your own cases, what floor, what failed and how it was fixed.

Rollout and rollback says how the deployment starts small and how it can be stopped. If the agent is released alongside the current process, with traffic moved over in stages, say so; blue-green deployment is one pattern.

Accountability names the executive who answers for the agent.

The template

No. Section Content Length
1 Decision required One sentence 1 line
2 Summary Purpose, benefit, key risks, recommendation 1 paragraph
3 Business purpose Problem solved, outcome expected Half page
4 Authority requested Capabilities, depth, limits Table
5 What stays with people Actions reserved for named roles List
6 Risk and appetite Residual risk per capability against appetite Table
7 Evidence Testing on own cases, results, open findings Half page
8 Customer and conduct Disclosure, vulnerable customers, complaints Half page
9 Regulatory position Rules that apply and how they are met Half page
10 Rollout and rollback Stages, triggers to stop, fallback Half page
11 Monitoring and reporting Measures and frequency to the board List
12 Accountability Executive owner, second-line view 2 lines
13 Appendices Technical design, full test report, risk assessment As needed

A filled example

A hypothetical building society, extract.

1. Decision required. Approve the deployment of a mortgage servicing agent to handle payment date changes and overpayment requests for existing borrowers.

4. Authority requested.

Capability Depth Limits
Answer questions about the borrower's own mortgage 1 Verified borrowers only
Change payment date 3 Once per 12 months
Process an overpayment within the free allowance 3 Within annual allowance
Process an overpayment above the allowance 2 Drafted; a person confirms charges

5. What stays with people. Arrears and forbearance; product switches; any request from a borrower showing signs of financial difficulty; complaints.

7. Evidence. 1,000 past servicing requests replayed. All capabilities met the floor. Twelve failures in the first run, all on overpayments spanning two allowance years; the brief was corrected and the re-run was clean.

12. Accountability. Chief operating officer. The second line supports deployment subject to monthly reporting for the first six months.

Banks supervised under model risk guidance may also need the paper to show how the agent fits senior oversight; see SR 11-7 governance: board and senior management oversight of agents.

Questions people ask

How long should the board paper be?

Four to six pages plus appendices. If the main paper is longer, the board is being asked to review detail it should delegate.

Should the vendor present to the board?

Rarely. The board should hear from the executive who will be accountable. Vendor detail belongs in an appendix.

What should the board approve: the agent or its authority?

Its authority. The paper should ask the board to approve specific capabilities at specific depths, so later widening comes back for a decision. That is the approach in agent governance is a bound, not a model.

Decide what your agents may do. Prove it. Enforce it.

Pryme Intelligence certifies each agent on your own cases, checks every action against five bounds at the moment it happens, and files every verdict.