What does cost per outcome measure?
What one unit of a function's output cost to produce in a cycle, with everything included, so that it can be compared with what the same unit cost before the Digital FTE owned the function. On ftes.ai the inputs are all visible: the role's usage in its Performance tab, the outcomes in its Targets tab with evidence, the human hours from the baseline sheet the team keeps, and external spend from the platform that billed it. The number is per role, because usage is reported per role and outcomes differ by role.
The glossary entry on cost per outcome gives the short definition; this piece gives the arithmetic.
How is the numerator built, with the inputs left blank?
Tokens: the role's usage in the cycle, [tokens], at the plan rate, which is the block price divided by the block's tokens, [rate], giving [figure]. Human hours: approval minutes from the Approvals tab or the manager's log, [hours], at the approver's rate, [rate]; review and send-back reading, [hours], at [rate]; setup in the first cycle only, [hours], at [rate]; total [figure]. External spend: any platform spend the role controls, ad budget for a Performance Ads FTE, [figure], billed by that platform. Numerator: the three added, [figure]. In the first cycle, mark setup separately so the second cycle's numerator excludes it.
If a human hour was not logged, it is not zero; estimate it conservatively and label the estimate.
- Tokens at the plan rate
- Approval, review, and setup hours at their rates
- External spend under its own cap
How is the denominator built, and what is excluded?
Outcomes: the units the role's Targets tab records with evidence in the cycle, [count], of the kind agreed at the first review: transactions reconciled and approved, tickets resolved at first line with a citation, shortlists delivered, pages published that passed the fact gate, qualified leads routed with evidence. Excluded: proposals nobody approved, which are tasks and not outcomes; outcomes the platform marks unmeasured, such as conversion or satisfaction without a survey; and anything counted by activity rather than result. The denominator is what a person confirmed happened, which is why the approval record and the Targets tab are the source.
A role with a large numerator and a small denominator in its first cycle is usually a role whose classes were all at Level 0 and read closely; label it and wait for the second cycle.
How is the baseline built to match?
Before the pilot, for a fortnight, record the same function the old way with the baseline time template: hours per outcome, [hours], at the rates of the people who did it, [rate], plus any external cost, [figure], plus tooling attributable to the function, [figure], divided by outcomes in the fortnight, [count], giving baseline cost per outcome, [figure]. The baseline must count the same kind of outcome the Targets tab will count, or the comparison is meaningless. Where the function was not done at all before, overnight replies, weekly citation sampling, there is no baseline for that outcome, and the honest statement is that the outcome is new rather than cheaper.
A baseline estimated after the pilot from memory is a guess dressed as a comparison; record it first.
How does the comparison read?
Second-cycle cost per outcome, [figure], against baseline cost per outcome, [figure], for [role] on [outcome], with human hours included in both, the first cycle labelled and set aside, and unmeasured outcomes excluded from both. State the difference as a figure and, if useful, as a share, and state what is not in it: the outcomes the platform could not evidence, the external spend that moved to a different line, and the redeployed hours the team now spends on judgement work, which are a benefit the number does not capture. The whitepaper on the economics of digital labor sets the same arithmetic in the context of a plan and a payback period.
The number is defensible because every input can be opened, and the caveats are stated rather than hidden.
How is it misread?
Four ways. Using the first cycle, which includes setup and close reading and is not representative. Omitting human hours, which flatters the FTE and is the omission a CFO will find first. Counting proposals as outcomes, which is activity. And blending roles, which produces a number that describes neither a reconciled transaction nor a resolved ticket. Each is avoided by taking the inputs from where they live: usage per role, outcomes from Targets with evidence, hours from the sheet, and the baseline from before the pilot.
The CFO persona piece sets out the standard of evidence a finance leader should hold this number to.