Why does this line behave differently?
A Digital FTE budget line behaves differently from a software licence because spend follows work rather than seats, and differently from a salary because the ceiling is hard. The plan block is the maximum that can be consumed in a cycle; an empty wallet pauses work rather than overspending.
That makes the line easier to defend than either comparison, as long as it is planned as what it is. The mistake is to budget it as a licence, which ignores usage, or as a salary, which ignores the ceiling.
What goes in the line?
Four things go in the line. The plan block, chosen from expected roles and output and priced on the pricing page. The human approval time, which is minutes per proposal and adds up to hours per month for an active role. The setup time in the first month: connecting tools, writing the remit, reviewing early proposals. And external platform spend that a role controls, such as an ad budget, which is billed by that platform.
Leave out assumed savings. They belong in the review after they are measured, not in the budget before they exist.
- Plan block: the ceiling, from the pricing page
- Approval time: minutes per proposal, hours per month
- Setup time: first month, human hours
- External spend: ads or other platform budgets, separate line
How do you size the block?
Size the block from the roles you will run and the output you expect from each. A single role publishing a few pages a month needs less than three roles working daily. If unsure, start with a smaller block; the workspace reports usage per FTE, and the block can be changed from the next cycle.
The first cycle is a measurement cycle. Its usage is the basis for the second cycle's block, which is how the budget becomes evidence-based within two months.
How do you review the line?
Review the line each cycle on two numbers per role: usage in tokens and cost per outcome, meaning tokens divided by the outcomes recorded in the role's Targets. A role whose cost per outcome is falling is getting more efficient or earning autonomy; one whose cost is rising with flat outcomes needs its remit or model tier examined.
The monthly review each Digital FTE writes in the first person is part of the evidence: it states what was delivered, what was missed, and what it recommends, and the finance leader can read it alongside the usage.
What about the year-end view?
At year end, the line shows the blocks purchased, usage per role per cycle, cost per outcome per role, and the human hours spent on approval and setup. Against that sits the outcome record from Targets and any measured changes in the baseline metrics recorded before each pilot.
That is a defensible view because every number in it was measured. The comparison with the alternative, whether headcount, agency, or leaving the work undone, uses the same recorded baselines, as the headcount-versus-AI-budget guide describes.
What does a labelled first-year budget look like?
As a labelled outline with your own figures to fill in: quarter one carries the smallest plan block that fits one role, the manager's setup hours in month one, approval hours at the Level 0 rate for the first two cycles, and no assumed savings. Quarter two carries the block resized from the first quarter's usage per role, approval hours at the lower rate the record shows once routine classes are promoted, and, if a second role is added, a shorter setup line because the platform pre-fills what it already knows.
Quarters three and four carry the blocks sized from the record each cycle, the small recurring lines for integration upkeep and governance reviews, and any external platform spend a role controls on its own line. The review column alongside each quarter carries the measured outcomes from Targets and the cost per outcome per role, and the year-end view compares them with the baseline recorded before the first pilot. A budget built this way has no invented number in it, and every line can be traced to a workspace figure or a logged hour.