Stop planning for a smaller finance team. Plan for a differently shaped one.

Every conversation about AI and finance headcount reaches the same tedious impasse. One side says the function halves. The other says nothing ever changes. Both are wrong in the same way: they are arguing about size when the interesting change is about shape.

The work AI absorbs first is bounded, repetitive and reviewable. That is a precise description of what a first-year trainee does. It is also, inconveniently, the mechanism by which trainees learn what a set of accounts feels like when something is wrong. You cannot develop the intuition that catches an odd creditor balance without having reconciled several thousand of them. Remove that apprenticeship and you have not saved money; you have deferred a capability gap by about eight years.

The firms thinking clearly about this are doing something counterintuitive. They are keeping intake broadly flat and changing what trainees do — more review, more exception handling, more exposure to the commercial conversation earlier. It costs more per head in the short term because a reviewer of machine output must be trained faster than a preparer of manual output. It is, as far as I can see, the only route that produces senior finance people in the 2030s.

The middle of the pyramid is where the genuine pressure sits. Roles whose value is coordination — chasing, consolidating, reformatting between systems — are exposed in a way that neither the trainee nor the business partner is. That is the honest conversation to have with your management layer, and it is not being had in most organisations.

One provocation to close on. If your AI business case is built on cost reduction, you will spend two years explaining to the board why the savings did not appear as savings. Capacity released rarely converts into headcount removed; it converts into work that was not being done. That is a better outcome, and a harder story. Tell the harder story first.