AI in the boardroom needs an institutional memory
A board paper records one point in a continuing decision. To understand that decision, directors may need the assumptions, challenges and commitments recorded across several earlier meetings. This is where I see a practical role for AI.
Start with the questions directors need to answer
The first task is to define useful questions. When did an exposure first appear? How has management’s assessment changed? Which mitigation was promised, and what evidence shows it was delivered? What was the rationale for accepting an exception? These questions give a purpose to the archive and a way to judge whether a new system is helping.
Keep the original evidence close
An AI-generated answer should point to the paper, page and meeting date supporting it. It should distinguish a direct statement from an interpretation and make missing information visible. Directors need a practical way to open the original evidence, assess the context and challenge the summary. A confident answer without a traceable basis should not become a board fact.
Preserve meaning as the framework changes
Institutions change their risk categories, ratings and reporting templates. A system that compares historical papers needs to recognise those changes. A lower rating may reflect a revised scale rather than an improvement in exposure. Useful institutional memory therefore includes the definitions in force at the time and the reason for any later reclassification.
Respect access and decision rights
Board material needs permissions suited to its sensitivity. Retrieval should respect those permissions, and sensitive records should remain within approved systems. Human owners must remain responsible for the risk assessment, the recommendation and the decision. The technology should make the supporting work easier to inspect, with a record of important changes.
Evaluate against real tasks
I would begin with a bounded set of approved historical papers and a set of questions whose answers can be checked. Reviewers can assess whether the system retrieves the right evidence, represents it accurately and admits when the record is incomplete. Useful measures include source accuracy, missed evidence and the effort required to verify an answer. Speed matters once the result can be trusted.
Build continuity into the meeting cycle
The value grows when each meeting adds a clear account of decisions, owners and unresolved questions. That gives the next committee a better starting point and creates continuity as people change. This is the direction behind OneBetaRisk: helping institutions connect their risk history with the decisions in front of them.