Programme sizing · Jan 2026 ·
Sizing the prize before the build began
Before a company commits to an AI programme, someone should work out what it is worth. The useful version of that estimate has visible assumptions, arithmetic anyone can audit, and a range instead of a point.
Most AI business cases fail in one of two directions. Either the value is asserted without any arithmetic at all, as transformative impact, or it is computed to false precision, as an ROI of 312%, from assumptions the reader never sees. Both, in our experience, produce the same reaction in a serious buyer: polite disbelief.
A range built from four visible assumptions is worth more than a precise number nobody believes.
Here is what that looked like in practice, for a client firm weighing an AI programme. We started by writing down four assumptions, each one checkable against the firm’s own records. Everything downstream is multiplication, and that is by design.
The programme had two levers. The first was proposal automation. Twenty proposals a month at sixteen hours each, with an estimated 60 to 70% time reduction, frees somewhere between 2,300 and 2,700 hours a year. At the average rate, that is roughly €0.4 to 0.5M of capacity. The second lever, personal AI agents, was far larger. If a hundred specialists each recapture five to ten hours a week, that adds up to 23,000 to 46,000 hours a year, or €4.0 to 8.1M of potential billable capacity.
Caveats
The largest number in the model comes with the largest caveat, and the caveat belongs in the headline rather than the footnote. Recaptured hours only become value if there is demand to fill them. The estimate is capacity, not revenue. And that is exactly why the answer is a range. The low end and the high end describe different worlds of adoption and utilisation. Pretending to know which world you are in, before the pilot has even run, is precisely the false precision the model exists to avoid.
Implications
A range with visible assumptions supports decisions a point estimate cannot. Consider the small bar. Proposal automation, even at the very bottom of its range, was enough to cover the cost of the engagement several times over. That meant the go or no-go decision did not depend on the big bar at all. The big bar decided other things: the order of work, and what the pilot had to measure. Namely, hours actually recaptured, and hours actually redeployed into billable work.
And because every number sits just two multiplications away from an assumption, disagreement moves to where it is productive. Our billing rate is lower than that, is a conversation. I do not believe your ROI, is not. The whole model fits on one page. That, too, is a feature.