How to Review a Board Pack with AI
To review a board pack with AI, start by confirming what material the system can read and what decisions the meeting must make. Review the papers individually, preserve their source references, compare related claims across the pack and check the arithmetic. Finish with a short, ranked set of questions that could change a decision.
The test is whether the review improves your preparation for the meeting.
Imagine a pack in which the chief executive’s report describes a project as on schedule. The capital expenditure paper requests an additional allocation. The risk report records a delayed regulatory approval. Each paper may be understandable on its own. Read together, they raise a consequential question: what has changed, and does the recommendation still hold?
AI can help connect those pieces. To be useful, the review needs a process that follows the evidence all the way to the decision.
Start with the decisions the board must make
Before asking for analysis, establish the meeting, the committee or board involved, the reporting period and any areas requiring particular attention.
Then identify what each paper asks of the reader. Is the board being asked to approve expenditure, endorse a direction, note performance or resolve an issue? Does the paper make that request explicit?
This distinction helps focus the review. An information paper and an investment proposal deserve different questions. The latter needs enough evidence to examine the proposal, its alternatives, its assumptions and the consequences of committing.
The UK’s Financial Reporting Council makes a useful point in its governance guidance: board papers should make clear what directors are expected to do, and significant proposals should explain how they were developed and challenged. That is useful practice to draw on, while applying the governance requirements relevant to the organisation being reviewed. Source: FRC Corporate Governance Code Guidance.
An effective AI review should make unclear decisions visible early. “What precisely are we being asked to approve?” can be the most valuable question in the room.
Confirm what the AI actually received
A board pack can contain searchable text, scanned pages, charts, tables and appendices. Successful upload does not establish that every part became usable evidence.
Ask for a map of the papers and their page ranges. Check that the map covers the intended scope. Identify unreadable pages, omitted sections and material that exists only inside images.
For a workflow that works from extracted text, a chart’s visual message may be absent even when its title survives. A sentence such as “performance improved, as shown below” is inadequate evidence if the underlying chart was never read.
This matters particularly when interpreting silence. A report that found no issue in a paper it could not read has told you nothing about that paper.
Confirm which page numbering convention the review uses, too. A PDF’s page count and the printed page labels can differ. A source reference earns its place by helping you find the evidence quickly.
Review each paper before joining the findings
For each paper, establish its purpose, its main claims, the assumptions behind those claims and the information needed to assess them.
Then examine the gaps. Does the recommendation depend on a forecast without an explained basis? Is a material risk described without an owner or response? Does a proposal discuss its preferred option while leaving the alternatives unexplored?
Keep the language precise. “The supplied paper does not include a downside scenario” is a finding about the evidence. “Management has not considered downside risk” is a broader claim that the pack may not support.
That discipline makes the eventual questions fairer and more useful. It lets the board ask for what is missing without accusing management of something the review has not established.
A worked example: three papers, one investment decision
Consider this fictional pack. The figures and page references below are illustrative.
The executive summary on page 8 asks the board to approve a distribution project costing R12 million. On page 46, the project paper lists R8 million for equipment, R3 million for implementation and R2 million for contingency. On page 71, the risk report describes customer demand as provisional.
Three findings deserve attention.
The cost components add to R13 million. There is a R1 million difference from the request in the executive summary. That arithmetic is straightforward, but its meaning needs checking. Is contingency included in the approval amount? Are the papers from different versions? Do the amounts use the same tax basis?
The demand assumption needs a status. Provisional demand may be a reasonable basis for a scenario. It should be distinguishable from contracted demand, especially if it supports the investment recommendation.
The proposed resolution needs to connect to the evidence. The board should be able to tell which amount, scope and conditions it is being asked to approve.
A useful review turns these findings into questions:
Which project cost is the approval request based on, and how does it reconcile to the components on page 46?
What proportion of the demand assumption on page 71 is supported by customer commitments?
If demand is delayed, what happens to cash requirements and the proposed implementation schedule?
Those questions give management something specific to answer. They also help the chair focus the discussion on issues that could change the decision.
Separate arithmetic findings from interpretation
There are two different tasks in the cost example. Adding R8 million, R3 million and R2 million is a calculation. Deciding whether that total is directly comparable with the R12 million request requires interpretation of scope, period and definitions.
Software can check arithmetic over extracted figures. Someone still has to establish that the figures were read correctly and belong in the comparison.
The same distinction applies to repeated metrics. “Revenue” in one paper might refer to the group, while another uses the same label for a subsidiary. A difference is a reason to investigate before treating it as a contradiction.
The review should preserve that distinction in its wording. It should also show which checks ran, which could not run and where the evidence was incomplete.
How ExecutiveNavigants approaches the review
The ExecutiveNavigants Board Pack Evaluator maps the pack into sections and works through them individually. It saves section analyses and extracted figures, then applies a calculation engine to the accumulated register.
The checks include totals, variances, ratios and comparisons of related figures. The report combines the findings with section analysis, decision clarity and ranked questions carrying page references.
This architecture lets the review build on an explicit working record as it moves through the pack. The calculation engine operates on the figures transcribed into that record. Its checks do not independently prove what was printed on the original page, so extraction accuracy and the interpretation of a metric still need attention.
The tool also surfaces gaps in coverage. Its continuity review concerns what the supplied pack says about earlier meetings; it does not establish the contents of a previous pack that was never supplied to the workflow.
These boundaries help a director use the output appropriately: as preparation for informed challenge and judgement.
Take a short question brief into the meeting
The final output should help you prioritise. Keep detailed findings available, but put the consequential questions first.
For each priority question, retain the source reference, why the issue matters and what information would resolve it. Distinguish a decision blocker from a clarification that can follow after the meeting.
A discrepancy that changes the amount being approved deserves attention before a minor labelling issue. An unsupported assumption that could change the investment case deserves more discussion than an editorial inconsistency.
The value of AI-assisted board preparation is visible when the meeting starts with better questions, clearer evidence and a more precise understanding of what is being decided.
Explore the Board Pack Evaluator at ExecutiveNavigants.