Step 1: Read usage per role for the last cycles
Open each Digital FTE's Performance tab and read its usage per cycle for as many cycles as exist, with the outcomes in its Targets tab beside it. Usage is reported per role on ftes.ai, so each role's line is its own; never blend them. Note the trend: usage usually rises in the first cycle, settles in the second, and moves with volume after that. Note which classes earned autonomy and when, because promoted classes change the human hours more than the tokens. Put the numbers in the budget workbook with the cycle dates.
The metric explainer on cost per outcome gives the arithmetic the budget line will be judged by.
Step 2: Size next year's block from the record
Take the settled usage per role, apply the volume you expect for each function next year from the function's own plan, and add the roles the Second-Role Test says the team will be ready to add, each estimated from the closest existing role's usage. Sum to the organization's expected usage, add room for peaks and for the first cycle of each new role, and choose the plan block that covers it. The block is a ceiling with a hard stop at zero, so the maximum exposure in any period is the block bought, which is the sentence the budget line should carry.
Plan changes are made from the workspace by an owner or admin and apply next cycle; the budget notes the cycle the change lands.
- Settled usage per role, scaled by each function's plan
- Planned roles estimated from the closest existing role
- Room for peaks and first cycles; a block that covers the sum
Step 3: Put the human hours on the same line
From the baseline sheets the function managers keep, take approval minutes per day per role, review hours per month, and governance hours per quarter, at the rates of the people who spend them, and put them on the AI line beside the block. They fall as classes earn autonomy, and the budget can show that trend from the record, but they are never zero. A budget that omits them is the one a CFO will send back first, and rightly. Setup hours for planned roles go in the cycle each role is added.
The CFO persona piece sets the evidence standard for this line.
Step 4: Keep external spend on its own line
Any spend a role controls on another platform, ad budgets for a Performance Ads FTE above all, is billed by that platform under its own cap and belongs on that platform's line, not on the AI line. The AI line carries the role's tokens and hours; the ads line carries the media. Blending them makes the AI line look large and the media line look small, and neither is true. The same applies to any tool subscription a role uses through a connection.
The hidden-costs piece lists what else to look for around any vendor's number.
Step 5: Budget nothing for assumed savings
Savings belong in the review after they are measured, not in the budget before they exist. The budget carries the block, the hours, and the external spend as costs, and it carries the baseline cost per outcome for each function as the comparison the review will make. Where a role has two clean cycles, the record's cost per outcome can be shown beside the baseline as evidence for the line, labelled as last year's measurement. Where a planned hire for repetitive work is being absorbed by a role instead, say so in the headcount line, plainly.
A budget built on assumed savings is a forecast dressed as a fact, and it fails the first review.
Step 6: Defend the line from the record
At the budget review, the line is defended with three things a finance leader can open: usage per role and its trend, cost per outcome per role against the recorded baseline for the roles with two cycles, and the hours log. The questions to expect are the CFO's five: unit, bounds, usage per role, what is not in the price, and how plans change; the answers are tokens, the wallet with its hard stop and per-role budgets, yes, external spend and human time, and from the workspace next cycle. The four-option framework in the library covers the question of why a role rather than a hire, agency, or nothing, for any function the review challenges.
A line built this way survives a sceptical review because nothing in it came from a vendor.
What does done look like?
A budget workbook with usage per role by cycle, next year's block sized from it with the plan tier named and the cycle it applies, per-role budgets inside the block, human hours on the AI line at their rates with the trend from the record, external spend on its own lines under its own caps, planned roles with their estimated usage and the cycle they start, baseline cost per outcome per function as the review's comparison, no line for assumed savings, and any absorbed hire stated in the headcount line. The review reads it and signs it.
The whitepaper on the economics of digital labor gives the model the workbook follows.