Why nine questions
AI is arriving in pension governance whether boards choose it or not. The Pensions Regulator's recently published AI plan makes the position clear: trustees remain accountable for decisions and outcomes even where activities are delegated. The board is expected to understand where AI is being used, establish governance around it, and test the risks on a continuing basis.
The practical challenge is knowing what to ask. Most trustee boards are not technology specialists, and nor should they need to be. They do need a way to assess whether the AI touching their scheme's data and decisions meets the standard the law and the regulator expect.
Amédée Levillain, CEO of Knowa, is publishing a series of nine questions designed for exactly this purpose. Each question is short enough to ask out loud in a board meeting. Each is designed to produce an answer that is either useful or revealing. Together, they form an initial governance screen: not a substitute for procurement or legal review, but a starting point that ensures the board learns something it needs to know.
The series is being published on LinkedIn over the coming weeks. The first two questions are live. This article summarises them.
Question 1: Is this built for fiduciary work, or is it general-purpose AI in a new wrapper?
Read the full post on LinkedIn
Modern general-purpose AI models understand pension terminology well. Vocabulary is not the test. The governance question is what surrounds the model: the scheme-specific records it can reach, the data boundaries it respects, the workflow it sits in, the controls on what it is permitted to do, the citations behind its answers, and the audit trail behind its decisions.
A platform built for fiduciary work treats the AI as a constrained component inside a controlled system. Explicit boundaries on what it can do. What gets logged. What is traceable back to source. Who is accountable for the output.
What a good answer looks like: a product team led by professionals who have been confronted with the problem, not only engineers. Domain depth shows up in small things (how late papers are handled, how actions carry forward across meetings rather than being treated as one-off tasks, how conflicts of interest are a structural feature rather than a checkbox).
Warning signs: the vendor cannot name a pensions professional who shaped the product. Output reads as articulate but generic, and could equally apply to a charity board or an NHS trust. Demonstrations rely on document summarisation, not on tasks specific to fiduciary work.
Question 2: Does this match how a trustee board actually operates, including how we work with our advisers?
Read the full post on LinkedIn
Pension governance is collaborative by design. Trustees deliberate together. They take advice from investment consultants, actuaries, covenant advisers, legal counsel, and the in-house pensions team. They sign off in committee, often by sub-committee first. The work is asynchronous, distributed, and shared between people who do not all sit in the same firm, the same country, or the same week.
Many AI deployments, including capable enterprise tools, still fail to reflect this. A platform that treats the board as a single user, in one organisation, working on one document, will quietly create more friction than it removes. Or it will silently exclude the advisers whose input shaped the decision.
This matters under TPR's accountability framework. Trustees remain responsible for AI use by their administrators, providers, and advisers, not only for AI they procured themselves.
What a good answer looks like: the platform reflects the quarterly cycle, the agenda, the sub-committees, the advisers. Controlled access for employee benefit consultants, professional trustee firms, legal counsel, and sponsor-side teams. An audit trail that records who saw what, and when.
Warning signs: single-user productivity tool with no native concept of trustees, advisers, or committees. No audit trail of access. Collaboration costs that discourage proper adviser participation. Data leaving the scheme's environment in ways the board cannot see.
Questions 3 to 9: coming soon
The remaining seven questions will be published over the coming weeks on Amédée's LinkedIn. We will update this article as each new question is released, building toward the complete governance screen.
The full Knowa white paper, The Architecture of Real Decisions, which provides the intellectual foundation for this series, is available to download on the Knowa white papers page.
Nine questions. One policy. The board learns something it needs to know.