Can an AI agent produce board reports?

It can produce the draft, from approved data, with every figure traced to its source. An AI agent can assemble the finance pack, operational metrics and risk data, draft the tables and summaries, and check consistency across papers. What it cannot do is own the message. The executive presenting each paper approves it, and the company secretary releases the pack.

What makes board reporting different from other reporting?

Board papers are read by people who are legally responsible for the decisions they make on them. A director relies on the paper being accurate and complete, and on the executive who wrote it standing behind it. That makes ownership essential. A paper with nobody's name on the judgement in it is a weaker paper, however good the numbers.

Board reports also mix sources: management accounts, operational data, compliance reports, risk registers, project updates. Each has its own owner and approval. The agent's job is to pull approved versions together and make them consistent, not to create new figures.

What can the agent prepare?

• Finance section. From the approved management accounts, as covered in how to prepare the monthly board finance pack.
• Operational metrics. From the same definitions used in daily operations reports, summarised to board level.
• Compliance and risk. From the approved compliance and risk reports, which have their own rules set out in whether an AI agent can produce board compliance reports.
• Consistency checks. The same figure must say the same thing in every paper.
• Draft summaries. Factual summaries of what changed, for the executive to edit.
• Action tracker. Status of actions from previous meetings, drawn from owners' updates.

All at depth 2.

Which step keeps the separation of duties?

Two approvals. The executive who presents each paper approves its content, including every judgement in it. The company secretary, or whoever manages board business, releases the pack after checking that each paper has its approval. The agent prepares for both and does not send anything to the board.

There is a subtler line too. The agent should not decide what the board sees. If an operational metric turned red this month, the agent includes it; it does not drop it because the paper is getting long. Omitting a bad number is a decision for the executive, and one they answer for.

An illustrative example

Take a regulated lender with a monthly board. The agent drafts the pack from the approved management accounts, the arrears dashboard, the complaints report and the risk register.

Its consistency check finds two problems. The finance paper shows loan book growth of 12 million over the quarter, while the risk paper, using an earlier data cut, shows 10 million. And the complaints figure in the chief executive's summary differs from the compliance report because one counts complaints received and the other complaints closed.

The agent reports both to the finance director and chief risk officer, who agree to use the later data cut and define the complaints figure as "received" in every paper. It updates the drafts. The chief executive rewrites the summary in their own words. Each executive approves their paper. The company secretary checks the approvals and releases the pack.

What controls make it safe?

Where Pryme fits

In Pryme, drafting board papers and sending anything to directors are separate capabilities, and the second is outside the reporting agent's role scope. Board members also need to hear how the agents themselves are performing; the agent can assemble that from the verdict record, as discussed in how to report on AI agent risk to the board.

When the board is deciding on agents themselves, the board paper on AI agent deployment template gives a structure. The wider finance controls are in AI agents for finance: books you can audit.