Every AI conversation with a board eventually arrives at the same question: "What's this going to cost us?" It's a fair question, and it's the wrong one to lead with - not because cost doesn't matter, but because asked first and on its own, it quietly frames AI as an expense to be minimised rather than an opportunity to be sized. And the framing decides the answer before anyone's done the maths.
The short version: Don't open with "what will it cost?" Open with "what's it worth if it works, and what's the cheapest way to find out whether it will?" Price is what you pay. Cost is what you forgo by getting it wrong - or by doing nothing. Stage your spending so you buy confidence before you commit capital.
Price is not cost
Price is the number on the invoice - the licences, the build, the consultants. Cost is the bigger, quieter figure: what you give up by choosing this over that, what you lose if it doesn't work, and what it costs you to not act while a competitor does. A project can have a low price and a ruinous cost, and a high price and a trivial one. Confusing the two is the most expensive accounting error in technology.
The classic example is the cheap pilot that "fails." A $25,000 experiment that tells you an idea won't work is often described as money wasted. It isn't. It's money that bought you certainty - and it's far cheaper than the alternative most organisations actually run: a $100,000, three-month project that arrives at the same conclusion the slow, painful way. Same lesson, four times the price, plus the quarter you'll never get back. The cheap experiment had the lower cost even though it had a price and "failed."
Invest in confidence, not in projects
The way to act on this is to stop funding AI as all-or-nothing projects and start funding it the way a sensible investor funds anything uncertain: in stages, buying down the risk before committing the big money. You don't need to be certain to start. You need to know what the next increment of certainty costs, and whether it's worth buying.
Concretely, it looks like this:
| Stage | Spend | What it buys you |
|---|---|---|
| Probe | ~$10k | Moves your confidence from 20% to 60%. Is this even feasible, and does it move a metric we care about? |
| Prove | ~$40k | Takes you from 60% to ~95%. A time-boxed proof against a real hypothesis, with real data. |
| Commit | The big build | Only now, once the risk is largely gone and the value is no longer a guess. |
The numbers are illustrative, but the logic is the point. You're not avoiding investment; you're sequencing it so the largest cheque is the last one, written when you already know it will land. A board that funds AI this way is doing what it would expect any prudent operator to do with an uncertain bet - not betting the farm on a slide deck.
Why cheap experiments beat expensive certainty
Underneath this is a principle Jeff Bezos has spent decades on: the lower the cost of an experiment, the more experiments you can run, and the more you run the more things you find that work.
"We've tried to reduce the cost of doing experiments so that we can do more of them. If you can increase the number of experiments you try from a hundred to a thousand, you dramatically increase the number of innovations you produce."
- Jeff Bezos
For a board, this reframes the budget question entirely. The goal isn't to approve the one big AI project with the best business case. It's to fund a portfolio of cheap, fast experiments, expect a good share of them to disprove their own hypothesis, and let the winners more than pay for the lot. That's how the value pipeline and experimentation pillars are meant to work - we go into the mechanics in the GIVE framework.
The question worth asking instead
So the next time AI comes to the board, try replacing "what will it cost?" with three better questions. What is this worth to us if it works? What's the cheapest experiment that would tell us whether it will? And what is it costing us, right now, to keep deferring the decision? Answer those, and the price - the number everyone started with - usually turns out to be the least interesting figure on the page.
Sizing the opportunity, not just the invoice? Our Accelerator is built around staged, time-boxed experiments - so you spend to learn before you spend to build. Get in touch to talk it through.