What does a CFO own in the Digital FTE model?
A CFO owns the ceiling and the evidence. The ceiling is the plan block: on ftes.ai a plan buys a block of tokens for the whole organization, every Digital FTE draws from that one wallet, and at zero every role pauses. Plan changes and top-ups are permissioned to owner and admin roles and logged, so the block is a decision you make, not one that happens to you. The evidence is the standard you set for any claim of return: a baseline recorded before a pilot, two full cycles, outcomes valued conservatively at what they cost before, and human time counted.
You also own the per-role view. Usage is reported per Digital FTE, so a Finance Controller FTE's tokens and a Support FTE's tokens are never blended, and each role's cost per outcome can be read against its own baseline.
What should a CFO ask before approving a plan?
Ask what the pricing unit is and confirm it is the work, not seats. Ask what bounds spend and confirm there is a hard stop, per-role budgets, and daily visibility of spend against budget. Ask whether usage is visible per role, because without that the return cannot be computed per function. Ask what is not in the price: external platform spend a role controls, such as ads, and the human approval and setup time. Ask how plans change and on what cycle.
On ftes.ai the answers are: tokens; the wallet with its hard stop, per-FTE and per-mission budgets, and spend in every morning brief; usage per FTE in each role's Performance tab; external spend billed by the platform that spends it; plans changed from the workspace with the new block applying next cycle.
- Unit: work or seats?
- Bounds: hard stop, per-role budgets, daily visibility?
- Usage per role: yes or no?
- Not in the price: external spend, human time?
- Plan changes: how, and when do they apply?
What should a CFO measure?
Measure cost per outcome per role: tokens consumed in a cycle at the plan's rate divided by the outcomes recorded in that role's Targets tab, set against the baseline cost of the same outcome from the sheet recorded before the pilot. Measure the human hours on approval, setup, and governance, logged from the first week, and watch them fall as routine classes earn autonomy. Measure the block against usage each cycle and size the next block from the record.
Refuse any number that does not trace to the workspace, the baseline sheet, or a logged hour. A savings percentage from a slide is not an input; it is a forecast dressed as a fact.
What should a CFO refuse?
Refuse to budget assumed savings; they belong in the review after they are measured. Refuse a comparison without a baseline. Refuse to judge a role on its first cycle, which includes setup and close reading and is not representative. Refuse to blend roles into one cost figure, because a page and a reconciled transaction are not the same outcome. And refuse to raise a block mid-cycle to chase a result, because the comparison depends on the ceiling holding.
These refusals are what make the AI line defensible to a board, an auditor, or a sceptical peer.
How should a CFO frame the headcount question?
Frame it as allocation of work, not replacement of people. A role is a bundle: judgement and relationship work stays on the headcount line, and the platform gates decisions about people permanently anyway; repetitive evidence-based work can move to the metered line. The usual near-term result is a redeployed team plus a new line whose ceiling follows the work. Where a hire was planned for repetitive work, the AI line can absorb it instead, assessed on the same recorded basis.
The four-option framework in the library, hire, agency, AI employee, doing nothing, puts each on the same five dimensions; the AI employee is the only one with a hard ceiling, and it is not always the answer.
What do the first 30 days look like from the CFO's seat?
In the first week, confirm the baseline sheet exists for the pilot function with rates from your team, and approve the smallest plan block that fits one role. In weeks two to four, read the spend line in the morning brief occasionally rather than daily; the function's manager reads it daily. At the first monthly review, read the role's usage and its first-cycle cost per outcome, labelled first-cycle, and size the second block from usage rather than from the original guess.
By day 30 you have a measured line with a ceiling and a record; the honest comparison arrives at the second review, and it is one you can sign.
What does the platform not do for a CFO?
It does not forecast your return; it gives you the mechanisms to compute it from your own record. It does not measure what it cannot see; the performance view marks unmeasured items as unmeasured rather than estimating, which is the property that makes the numbers it does show usable. It does not decide accounting treatment, tax positions, or sign-off; a Finance Controller FTE proposes with evidence and a qualified person decides. And it does not hide human time; approval, setup, and governance hours are yours to log and belong in the total.
A CFO who wants a vendor to promise a number should look elsewhere. One who wants a line they can defend from the record has the mechanisms here.