Why do most payback calculations fail?
Most payback calculations for AI fail because their inputs are assumed rather than measured. A percentage of hours saved, taken from a vendor slide, multiplied by a loaded salary, produces a number that looks precise and means nothing. The calculation is fine; the inputs are fiction.
The honest method replaces every assumed input with a measured one. It takes a pilot to get the measurements, which is why the pilot guide on ftes.ai insists on recording the baseline first.
What do you record before the pilot?
Before the pilot, record what the work costs today: the hours spent on the tasks the Digital FTE will own, by whom, and their cost; the outcomes produced, such as transactions reconciled or pages published; and the quality measures you have, such as days to close or first-line resolution rate.
This is the baseline. Without it there is no payback calculation, only a story. Two weeks of honest time recording by the people doing the work is usually enough.
What do you measure after?
After the pilot, measure the outcomes recorded in the FTE's Targets, the tokens consumed by the role from the workspace, and the human time spent on approval, setup, and send-backs. Measure the same quality metrics as the baseline, the same way.
Value the outcomes conservatively at what they cost to produce before the pilot, not at some higher notional value. If the work was not being done at all before, value it at what it would cost to have it done another way, or leave it unvalued and note it.
- Outcomes: from Targets, same units as the baseline
- Cost: plan share for the role from usage, plus human hours priced
- Quality: same metrics as the baseline, same method
- Unmeasured: listed, not estimated
How does the arithmetic work?
The arithmetic, as a labelled example with your numbers to fill in: monthly value equals the baseline cost of the outcomes the FTE now produces, from your recorded hours and rates. Monthly cost equals the role's share of the plan, from usage, plus the human hours on approval and governance at their rate. Setup cost equals month-one human hours at their rate.
Payback in months equals setup cost divided by the difference between monthly value and monthly cost, calculated once that difference is positive. If the difference is not positive after two full cycles, the role is not paying back on this remit, and the send-back reasons and usage per role are where to look.
What makes the result credible?
The result is credible when every input traces to a record: the baseline time log, the Targets outcomes, the usage per role, and the approval time. A finance leader can audit it, and the Digital FTE's own monthly review provides the outcome narrative alongside the numbers.
Report the misses too. A calculation that includes the outcomes that did not improve and the metrics that are unmeasured is the one that survives a sceptical reading, and it is the one the ftes.ai performance view is designed to support.
What does the payback record look like at the second review?
At the second monthly review the payback record has every input it needs, each traceable. From before the pilot: the baseline hours by task with their cost, the outcomes produced, and the quality measures. From the two cycles since: the outcomes in the Targets tab with evidence per item, the tokens consumed by the role from the workspace, the approval and setup hours the manager logged, and the same quality measures taken the same way. The Digital FTE's own report for each cycle sits beside them, stating what it delivered and missed.
The manager fills in the labelled arithmetic from those records: monthly value from the baseline cost of the outcomes now produced, monthly cost from the role's share of the plan plus logged hours, setup cost from month-one hours, and payback in months from the difference. If the difference is positive, the payback date is a calculation a finance leader can audit line by line. If it is not, the send-back reasons and usage per role say whether the remit, the model tier, or the choice of function is the cause, and that diagnosis is the review's real output.