A board should judge AI investment on the same discipline as any other capital decision, however fast others are spending. That means a named value lever, a baseline, a hurdle the return must clear and funding released in stages as evidence arrives. Market enthusiasm is a reason to move with purpose, not a reason to lower the bar.
AI investors now debate how AI will be funded at the scale of markets and nations. This article is AIPEX’s own guidance on the question that reaches every boardroom: how much should we spend on AI, and how do we know it is working?
Why does everyone else’s spending make the decision harder?
When competitors, suppliers and investors are all talking about AI, a board comes under two opposite pressures. One says: spend now or be left behind. The other says: wait until the returns are proven. Both are reasonable instincts, and both lead to poor decisions when they replace analysis. We look at the pacing question on its own in Should Your Business Slow Down on AI?
Spending because others are spending tends to produce a portfolio of pilots with no common measure of success. Waiting for proof tends to leave the business without the data, skills and experience it needs when the case finally becomes clear. The way through is not a different appetite for risk. It is a better way of releasing money. The same discipline protects the business if the market turns, as we set out in If the AI Bubble Bursts, Will Your AI Investments Still Pay?
What should the hurdle be?
AI investment does not need a special hurdle rate. It needs the organisation’s normal one, applied honestly. Two adjustments make that possible.
- Count the full cost. Licences and model usage are often the smallest part. Data preparation, integration, change management, training, monitoring and the time of senior people all belong in the case. So does the running cost after launch, which can grow with usage.
- Count only money. Hours saved, tasks automated and staff satisfaction are useful early signals, but they are not returns until they become revenue, cost removed, spend avoided or risk reduced. A case that relies on time saved has not yet reached the hurdle.
Where the business case is genuinely uncertain, the answer is not to lower the hurdle. It is to buy information cheaply before buying scale. A small, well-designed pilot with a clear baseline is a way of paying for evidence. Its purpose is to answer the question the business case could not, and it should be judged on whether it did, not on whether it produced a pleasing demonstration. Boards can approve these small bets readily, provided each has a defined question, a budget ceiling and an end date.
How does stage-gated funding work for AI?
Stage-gated funding treats each AI initiative as a series of small commitments, each released only when the previous stage has produced evidence. It lets a board back more ideas at lower risk, and stop the ones that are not working before they become expensive.
| Stage | Money released for | Evidence needed to move on |
|---|---|---|
| 1. Case | Defining the value lever, the baseline and the owner | A baseline taken, a named owner, a hurdle the case must clear |
| 2. Proof | A time-limited pilot in real conditions | Early measures moving: adoption, time, errors, compared with the baseline |
| 3. Scale | Roll-out across the business, with integration and training | Outcome measures reaching the P&L, signed off by finance |
| 4. Run | Ongoing operation and improvement | Return tracked quarterly against the hurdle, with a review date |
Every stage should have three possible outcomes: proceed, fix or stop. A portfolio in which nothing is ever stopped is not being governed.
How should the board read the portfolio as a whole?
Individual cases matter, but boards should also look at the shape of the portfolio. How much money is in each stage? How long have initiatives sat in proof without moving to scale or stopping? How much of the claimed benefit has finance actually signed off? A portfolio heavy with pilots and light on signed-off returns is telling the board something important, whatever the individual reports say.
A common pattern is heavy spending on AI licences that staff do not use enough to justify. Our article on AI licence ROI and adoption looks at that problem in detail. The same discipline applies: measure use in the workflow first, then decide whether to expand, fix or cut.
What should the board ask before approving AI spend?
- Which value lever does this investment move: revenue, cost, margin, risk, speed or service?
- What is the baseline, and when was it taken?
- What is the full cost, including data, integration, change and running costs after launch?
- What hurdle must the return clear, and is it the same one we use for other investments?
- What evidence releases the next stage of funding, and what would make us stop?
- Who owns the return, and when will finance sign it off?
If management can answer these, the board can approve with confidence, whatever the market is doing. If it cannot, approve the first stage only.
Where to start
Our briefing on how to measure AI ROI and why productivity numbers mislead sets out the five steps to a measure the CFO will accept, from a named value lever to a named owner of the return. Its recommended starting point is to score your largest AI investment with the AI ROI Board Readiness Scorecard below. If the result is Amber or Red, the next step is an Executive AI Strategy Day (£1,499 + VAT), which builds the baseline, measures and owners for every AI priority in one working day.
Frequently asked questions
Does AI investment need a different hurdle rate?
No. It needs the organisation's normal hurdle, applied honestly: count the full cost, including data, integration, change and running costs, and count only returns that reach money, such as revenue, cost removed, spend avoided or risk reduced.
What is stage-gated funding for AI?
Releasing money in stages, from business case to pilot, scale and run, with each stage funded only when the previous one has produced evidence. Every gate should allow three outcomes: proceed, fix or stop. It lets boards back more ideas at lower risk.
How should a board read its AI portfolio?
Look at how much money sits in each stage, how long pilots have run without scaling or stopping, and how much claimed benefit finance has signed off. See How to Measure AI ROI →