Straight answer
Is our AI spend paying off?
You cannot tell unless each use was given a measure before it started. Much AI spending sits in pilots, licences and staff time, reported as activity. Ask for the few uses that carry most of the cost, what each was expected to deliver, and what it has delivered so far. The answer is usually mixed, which is fine.
How can the board tell whether it is under control?
Signs it is under control
- Each significant AI use has a baseline and a target set before it went live.
- Management reports total AI cost, including licences, partners and internal time, in one place.
- Uses that are not delivering are stopped, and the money is moved.
Signs it is not
- Benefits are described as “hours saved” with no change in cost, output or service.
- Licences were bought for everyone and nobody knows how many are used.
- Every pilot is reported as promising and none has been closed.
What is the one question to ask?
Which of our AI uses has delivered what it promised, and which should we stop?
A strong answer
A short list with the original target, the result so far, the full cost, and a clear recommendation for each, including at least one to stop or change.
A weak answer
“It is too early to measure, but the feedback from users is very positive.”
Why is AI spend so hard to judge?
AI costs are scattered. Some sit in software licences, some in consulting, some in cloud bills that grow with use, and much of it in the time of people already on the payroll. The benefits are scattered too. A tool that saves each person an hour a week is real, but it only becomes money if the work is done with fewer people, more output or better service. If nobody decided in advance which of those was expected, nobody can say later whether it happened.
That is why the first useful step is not a new measure but an old discipline: a business case for each material use, with a baseline, a target and an owner.
What might the board be getting wrong?
Boards often approved the spending under pressure to be seen acting on AI, without asking for measures at the start. It is not fair to demand a return now on terms that were never set. Boards also sometimes expect every pilot to pay back. Some learning is worth paying for, as long as it is labelled as learning and has a limit. And a board that punishes stopped projects will get fewer of them stopped.
What should the board ask for instead?
Ask for one page listing the AI uses that make up most of the cost. For each, show what it cost, what it was meant to achieve, what it has achieved, and what management recommends: continue, change or stop. Ask for the total cost of AI in one number, including internal time where it is significant. Agree how often the board will see this, and keep the format the same each time so the trend is visible.