GlossaryFloor 4 · The Organisationa mosaic: AI is only one tile among manyFloor 4 · The Organisation
return on investment
No. 079 · v2026-08FR: retour sur investissementReturn on investment sets what a use brings in against what it costs, everything included: licences, calls, integration, review and support. Like the calculation for a machine tool, where the purchase price is rarely the decisive item.
What it is not
Return on investment is not time saved. An hour saved has value only if it is reallocated to something, failing which it turns up in no account. Nor is it the cost of the calls: that is generally the smallest line, far behind integration and human review time. And it is not a measure that can be made after the fact: without a baseline before deployment, all that is left is statements of satisfaction.
In depth
Three kinds of gain
Three natures of gain exist, and they are not proved in the same way. Cost reduction is the most verifiable: a task took so much time, it takes less, at constant volume. Capacity increase is real but indirect: more cases are handled with the same headcount, which shows only if volume is tracked. Quality gain is the most difficult: fewer errors, better turnaround, higher satisfaction, all things that have to have been measured before in order to be compared afterwards. Confusing these three natures produces flattering presentations and results nobody can find.
The full cost
The full cost has four items, and the most visible is the smallest. Licences and calls are easily quantified. Integration is heavier: connecting the data, building the guardrails, writing the evaluations. Human review is the item systematically forgotten, although it conditions use on everything that matters: a text produced in ten seconds and reviewed in five minutes has not saved ten minutes. And then there is support, without which the tool stays used by a minority of volunteers.
The method that holds
The method that holds in practice is modest. You choose two or three precise tasks, not a domain. You measure their current duration and quality for a few weeks, before any deployment. You tool up, you measure again, and you generalise only on what has moved. This discipline looks slow against the surrounding enthusiasm, but it is the only one that produces a figure defensible before a committee, and it also protects against abandonments decided on an impression.
Relations where the neighbours live
Check 3 questions · click your answer
Level 1 · Recognise
In the cost of an AI project, which item is most often underestimated?
Level 2 · Distinguish
A team reports saving two hours a week per person. Is that a gain?
Level 2 · Distinguish
Why do so many projects fail to demonstrate an effect?
Lexigraph, "Return on investment", v2026-08, https://www.lexigraph.org/en/return-on-investment/, CC BY 4.0.