The AI Board Member: A Field Report

Let me start with a necessary correction to our own shorthand. We talk internally about our "AI board member," and the phrase has stuck because it captures how it feels. But it is a metaphor, and as a former lawyer I should be precise: it holds no office, owes no fiduciary duty, and votes on nothing. Under the Companies Act, accountability sits with the humans, undiluted, and no amount of technology changes that. What we have built is not a director. It is something more interesting: a permanent, tireless presence in our governance that has changed how the actual directors spend their time.

Here is what that looks like from the inside.

What we built 

The idea began, as most useful things do, with an irritation. Our board packs were like every board pack I have seen in thirty years: assembled in the week before the meeting, describing a month that had already ended, read properly by some and skimmed by others, with the real questions surfacing too late to answer well.

So we started giving the machine the job the pack was pretending to do. Continuously, not quarterly, it reads what the business produces anyway: the management accounts, the pipeline, the delivery data, the commitments we made to ourselves in previous minutes. And it does four things with it.

It watches. Performance against plan, cash, utilisation, client concentration, the shape of the pipeline. Not as a dashboard waiting to be looked at, but as an active reader that notices when something moves in a way it shouldn't.

It remembers. Every commitment made in a board meeting is tracked until it is done or consciously dropped. This sounds trivial. It is transformative. Boards are institutionally forgetful, and executive teams know it. Ours no longer has that luxury, and I include myself in that.

It challenges the narrative. Before each meeting, it reads the board papers and compares the story we are telling with the data underneath. Where the two diverge, it says so. The first time it flagged that a paper's optimism wasn't supported by the trend beneath it, the paper's author was me.

It briefs the non-executives. Instead of arriving with whatever they gleaned from the pack on the train, our non-executives arrive with a short, machine-generated brief: what changed, what was promised, what doesn't reconcile, what deserves attention. The drop-in problem I described in the last essay, addressed directly.

We are now building the next layer: scenario work. Not predictions, but structured "what would have to be true" analysis around the decisions in front of us, prepared before the meeting rather than commissioned after it.

It is worth saying why we bothered, because the data on boards and AI describes a strange gap. Directors now rank deploying AI among their top organisational priorities and their biggest area of capital investment; two-thirds of them personally use AI tools to prepare for board work. Yet in the same surveys, only around a fifth of boards have any process governing that use, and fewer than one in ten report strong AI expertise around their own table — the lowest score of any competency measured. Among the largest listed companies, disclosure of board-level AI oversight has roughly tripled in two years, while the proportion of directors who have received any AI education remains in single figures. In other words: boards everywhere are racing to govern AI, and almost none are yet using AI to govern. The Diligent Institute's Dottie Schindlinger has predicted that high-performing boards will come to treat governance as "a continuous discipline" built on real-time data rather than periodic reports. We agree — so we tested it on ourselves first, which is the only honest place for a consultancy to test anything.

What it does well 

Three things stand out, and none of them is the one people expect.

It is incapable of tact. This is a feature. Humans soften. A finance director presents the numbers with a narrative; a machine presents the numbers with the narrative's inconsistencies attached. There is no career risk in a machine pointing out that the second half of the year requires a growth rate we have never achieved. Someone still has to decide what to do about that, but the observation now arrives unmanaged.

It never gets bored. The hundredth review of debtor days receives the same attention as the first. Human vigilance decays precisely where risk accumulates: in the routine, the familiar, the thing that was fine last quarter. Machine vigilance doesn't.

It removed the excuse of ignorance. Nobody in our boardroom can now say they didn't know, didn't see it, or weren't told. That has had a subtle but real effect on the quality of conversation: when everyone arrives with the same complete picture, the meeting starts where it used to end.

What it can't do — the warts 

Now the other side of the ledger, because a field report that reads like a brochure isn't one.

It cannot weigh what isn't in the data. It knows our pipeline; it does not know that a client relationship is fraying because of something said in a corridor. It can flag that a senior departure creates delivery risk; it cannot judge what that departure does to morale, loyalty, or the story people tell themselves about the firm. The most consequential decisions we have made this year turned on exactly those things.

It is confidently wrong in ways a person rarely is. Early on it produced analysis that was fluent, plausible and built on a misread of how one of our revenue lines actually works. A junior analyst who misunderstood the business would hedge; the machine did not. We caught it because an experienced human read it sceptically — chapter two of this series, applied. Every output it produces is now treated as a submission to be examined, never a verdict to be accepted.

It cannot own anything. When a call is finely balanced, someone has to carry it; legally, morally, and in front of the team who live with the consequences. A machine cannot stand in front of the company and say: this was my judgement, and here is why. Accountability is not a data processing task. It is a human act, and it is most of what a board is actually for.

It changed behaviour in ways we didn't intend. For a while, some papers started being written for the machine. This means pre-sanitised, defensively worded, engineered not to trigger flags. We had accidentally created a compliance reflex where we wanted candour. We now treat the machine's review as an input the author sees first, not a trap sprung in the meeting. The lesson: instrument your governance carelessly and people will manage the instrument.

And it is only as good as what we feed it. The month our delivery data lagged, its confident picture of the business was confidently out of date. Continuous assurance built on discontinuous data is theatre.

What actually changed 

The honest summary is this: the machine did not make our board smarter. It made our board's time more expensive to waste.

Meetings are shorter and harder. The supervisory layer such as the reading-back, the reconciling, the "whatever happened to" has largely gone, and what remains is the part that was always the point: judgement, argument, and decisions about the future. Our non-executives contribute more because they arrive knowing more. And the executive team, myself included, has learned that commitments made in that room are now permanent residents of it.

Would I recommend it? Yes! With the caveat that runs through this whole series. The machine amplified the experience and inquisitiveness already in our room. Point the same machinery at a board that lacks either, and you will get beautifully monitored drift: every exception flagged, no one equipped to know which flags matter. The technology raised the value of our people. It did not, and cannot, substitute for them.

The uncomfortable admission 

I will end with the wart I find most instructive. The hardest adjustment was not technical. It was mine.

A chief executive develops, over years, a quiet control over what the board sees and when. Not dishonestly, but through framing, sequencing, emphasis. The machine ended that. My board now sees the business as it is, continuously, without my narration. There were weeks when I found that genuinely uncomfortable, and I have concluded that the discomfort is the point. A governance system that never discomforts the chief executive isn't one.

Next in this series: if AI now does the work that juniors used to learn on, where will the next generation of experienced people come from? The apprenticeship is dead. What replaces it matters more than any technology decision we will make.