Direct answer: Most organisations can't say what they spend on AI because the spend is split across five budgets: software licences, pilots, supplier contracts, project budgets and expenses. To fix it, add those five together into one view, then put every pilot on evidence: success criteria written before it started, adoption measured by real use rather than licence counts, an exit date, and a named owner who decides to scale or stop. Do it before Q4 budgets close, and next year's AI money goes to the bets that are working.
Ask a board what the organisation spends on AI and the answer is usually a pause. Not because nobody knows, but because nobody knows all of it.
The spend sits in five places:
Each line looks reasonable on its own. Nobody adds them up, so nobody can say whether the total is producing anything.
Q4 is when this year's AI pilots either get funded properly for next year or quietly fade out. Too often the decision is made on enthusiasm rather than evidence: the most visible pilot gets the money, and the quiet one that was actually working gets cut.
Pilots that can't show a credible path to measurable value by the budget conversation rarely survive into next year. The evidence needs to be ready now, not assembled in the week the budget is due.
Without one view of AI spend across all five budgets, the board can't weigh AI against anything else, and can't see which bets are paying back.
Pilots often start without written success criteria. When the measure is decided at the end, every pilot can be declared a success, and the word stops meaning anything.
A licence count says how many people could use a tool, not how many do. Report who uses each AI tool, how often and for what, and let adoption decide the next renewal.
Pilots without exit criteria don't end. They drift, consuming budget and attention, neither scaling nor stopping. Before a pilot starts, write down the measure, the threshold to scale, the threshold to stop, and who decides on what date.
AI projects often lose their owner at launch. Someone has to own accuracy, retraining, prompt changes, errors and the benefits after go-live. Name that person before launch, not after the first problem.
Staff adopt AI tools before policy is written. That spend, and the data risk that comes with it, sits outside any budget line. Find what's actually in use, set rules by type of data, and give people an approved option that's as easy as the one they found themselves.
You don't need a new framework to do this. You need one page per AI initiative, and one summary page for the board.
For each initiative, record:
| Field | What to capture |
|---|---|
| Spend | Total across licences, pilot, contracts, project and expenses |
| Purpose | The business problem it was meant to solve |
| Success criteria | The measure agreed before it started, and the target |
| Evidence so far | What the measure shows today |
| Adoption | Who uses it, how often, for what |
| Owner | The named person accountable after go-live |
| Exit date | When the scale-or-stop decision is due |
| Recommendation | Scale, fix or stop |
The summary page is just the totals and the recommendations. It's the page that should be on the table when next year's AI budget is set.
With that evidence, every pilot falls into one of three groups:
Stopping a pilot isn't a failure. Funding one that isn't working is.
We help organisations move AI from scattered pilots to funded services, using the same Sync, Align, Optimise approach we apply to every IT estate.
Our AI readiness whitepaper sets out the questions to answer before choosing a model.
Request a 30-minute discovery conversation with our team. We'll help you see what AI is really costing, and which pilots have earned next year's budget.