The denominator problem in AI return figures
Two organisations can report the same return on the same deployment and mean entirely different things, because almost nobody publishes what went under the line.
- Return figures are not comparable between organisations, because almost nobody publishes what went under the line.
- A licence-only denominator is not a conservative estimate. It is a different measurement, and licences are usually a minority of first-year cost.
- Time saved is not money until the time is removed or resold. In year one it is normally reabsorbed, which is a real outcome and not a saving.
- Publish the denominator beside the number, in the same table, every time.
Reported return figures for enterprise AI are not comparable with each other, and often not comparable with themselves quarter to quarter. The reason is boring. The numerator gets scrutiny and the denominator does not.
What gets left out
A licence-only denominator counts seats and nothing else. It is the most commonly quoted and the least informative. A fuller denominator includes integration engineering, the data work that preceded it, security and privacy review, the internal time spent in enablement, the cost of the evaluation harness if one exists, and the ongoing inference bill, which moves with usage and therefore moves with success.
The gap between those two denominators is not marginal. In deployments where we have seen both numbers, the licence line is frequently a minority of total first-year cost. A figure built on the licence line alone is not a conservative estimate. It is a different measurement.
The numerator has the same disease
Time saved is the usual numerator, and time saved is not money until something happens to the time. Three things can happen. The time is removed, which means a role changed or ended. The time is resold, which means the same people now bill or produce more. Or the time is reabsorbed, which means it went into other work and the financial effect is zero this year.
Reabsorption is the normal case in year one, and it is a real outcome worth having. It is not a saving, and reporting it as one is the single fastest way to lose a finance partner who will otherwise be useful to you for years.
The rule
Publish the denominator next to the number, every time, in the same table. An internal figure without its denominator should not clear review. If that feels like an unreasonable standard, note that it is the standard applied to every other capital decision the organisation makes.
A return figure whose denominator is unstated is not a conservative claim or an optimistic one. It is an unfalsifiable one, and it will be treated as such the first time somebody with an incentive to check actually checks.
A usable format
State it as a sentence with four slots: over period, against a denominator of total cost, itemised, the deployment produced effect, of which portion has been booked. The last slot is the one that makes the sentence honest, and it is the one people leave off.