The middle layer is paid to decline
Executives sponsor the programme and individual contributors adopt the tool. The layer between them is measured on this year's numbers, lends its scarcest people, and carries the failure if it goes wrong. Declining is the correct read of that arithmetic, not a mindset problem.
- A director asked to run a pilot is being asked to spend scarce people now against a benefit that lands later, elsewhere, on someone else's line.
- The cost is specific: the people who can define a workflow are the people who understand it, and those are the ones the function cannot spare.
- Change-management workshops and executive mandates do not touch the arithmetic. A mandate produces compliance, which looks like participation and reports like it too.
- The interventions that work are all changes to measurement: let the function keep the savings, fund the backfill, and put the pilot in the director's own objectives at real weight.
- Test it by asking what the layer is paid to do. If the answer does not contain the pilot, the pilot is a favour.
Adoption failures are usually diagnosed at one of two levels. Either the sponsor did not commit, or the users did not take to the tool. Entry seven deals with the second and shows that refusal at the user level is usually verification arithmetic rather than attitude. This entry is about the layer nobody diagnoses, which sits between them and decides far more than either.
A director is measured on this year's numbers. Running a pilot costs them their strongest people for two quarters, because the people who can define a workflow precisely enough to automate it are the people who understand it best, and those are exactly the ones the function cannot spare. If the pilot works, the benefit is a headcount reduction that lands on a budget line that may not be theirs, probably in the next financial year, and the transformation office reports it. If it fails, it failed in their function, in front of their own leadership, this year.
Set out plainly, that is a rational person declining. It is not resistance to change, not fear of the technology, and not a mindset problem. It is a correctly-read incentive structure, and the person reading it correctly is usually one of the better managers you have.
Cost, certain and immediate:two quarters of the function's scarcest people, taken from work that is already on the plan.
Benefit, uncertain and displaced: a saving that arrives later, is measured centrally, and is attributed to the programme rather than the function.
Downside, certain and local: a visible failure inside their own remit, in the year they are being assessed on.
The size of this gate is measurable. Gallup's Q2 2025 workforce study, 19,043 employed US adults, found that among people whose organisation is actively implementing AI, only 28% strongly agree their manager supports their team using it. Where the manager does support it, employees are 2.1 times more likely to use the tools weekly. The platform is bought and the mandate is announced, and roughly seven in ten people are looking at a manager who has not backed it.
The standard prescriptions do not touch any of this. A change-management workshop does not move a number on a scorecard. An executive mandate produces compliance, which is worth naming precisely because it is so easy to mistake for progress: the pilot is staffed, the meetings happen, the status is green, and nothing about the underlying trade has changed. Compliance reports identically to commitment right up to the point where the programme needs the function to do something that costs it something.
What works is duller and it is always a change to measurement. Let the function keep the savings it generates rather than harvesting them centrally. Fund the backfill, so lending your best person costs the function nothing. Put the pilot into the director's own objectives at the same weight as everything else they are assessed on. Each of these removes one term from the arithmetic above, and any one of them can be enough.
One organisation adopted only the first, letting functions keep the first year of savings from what they automated. Volunteer pilots went from two to eleven in a quarter. Nothing about the technology changed, and no additional training was delivered.
The diagnostic is a single question, and it is worth asking out loud in the room rather than inferring: what is this layer paid to do? If the answer does not contain the programme, then every hour the programme gets is a favour, and favours are the first thing withdrawn when the quarter tightens.
This is generally called a culture problem, which is a convenient diagnosis because culture belongs to nobody in particular and is therefore always somebody else's to fix. It is an incentive problem. Incentives have owners, and the owner is usually two levels above the person being blamed.