What is a token?
A token is the unit in which language models measure work. Text the model reads and text it writes are both counted in tokens, and a piece of work that involves reading many documents and writing a long draft consumes more tokens than a short reply from a short prompt.
On ftes.ai, tokens are the unit of the plan: a plan buys a block of tokens for the organization, and every Digital FTE draws from that block as it works. The price is the block, not a number of seats or roles.
What consumes tokens?
Work consumes tokens. When a Digital FTE drafts a page, screens applications, reconciles a ledger, checks facts against sources, measures visibility, runs a mission, or writes the morning brief, it is reading and writing, and that consumes tokens from the wallet.
The amount depends on the size of the task and the model tier chosen in the FTE's Setup tab. A deep read of a large document set for a Research Analyst brief consumes more than a routine reconciliation match; a higher model tier consumes more per unit of work than a lower one.
- Drafting and rewriting content, replies, and proposals
- Fact-checking against sources
- Measuring, such as visibility sweeps
- Missions: production, diagnostic, and strategy runs
- The morning brief and monthly review
What does not consume tokens?
Using the workspace does not consume tokens. Opening the morning brief, reviewing and approving proposals, browsing the pipeline and library, reading performance reports, and changing settings are all free of charge, because no model work happens when a human reads what has already been produced.
This matters for the economics: the human side of the loop, which is approval, costs human minutes but not tokens. The tokens pay for the FTE's work, and the manager's review of it is not metered.
How do you read usage?
Usage is reported per Digital FTE in the workspace, so the organization can see which roles consume what. Spend against each FTE's budget appears in the morning brief, with a warning before the budget is reached, and mission spend is visible against the mission budget.
Reading usage per role is what makes cost per outcome possible: tokens consumed by a role in a period, divided by the outcomes it delivered. That comparison, role by role, is more useful than the total.
What happens when tokens run out?
When the wallet is empty, the Digital FTEs pause and tell you in the workspace. Nothing runs on credit. Top up, or wait for the next plan cycle, and they resume where they stopped, with no work lost.
Plans can be changed from the workspace, and the new block applies from the next cycle. The pricing page on ftes.ai shows the current blocks, and the workspace's usage per FTE is the guide to which block fits.
How does token usage differ between roles?
Token usage differs between roles because the work differs. A Finance Controller FTE reconciling transactions does many small pieces of work, each reading a bank line and a few candidate invoices and writing a short proposal, so its usage is steady and low per outcome. A Content FTE drafting a long page reads the Brand Brain and the approved sources, writes a draft, runs the fact gate, and possibly revises, so each outcome costs more tokens. A Research Analyst FTE reading a large document set for one brief may consume in a day what a reconciliation role consumes in a week.
Model tier multiplies the difference: a role on a higher tier consumes more per unit of work, which is worth it for synthesis and long-form drafting and wasteful for exact-match reconciliation. The workspace reports usage per FTE precisely so these differences are visible rather than blended into a total, and the right comparison is always cost per outcome for that role against the baseline cost of the same outcome, never total tokens across roles that do different work.
A practical habit for a new organization is to read usage per role at the end of the first cycle before drawing any conclusion about cost, because the first cycle includes setup, remit tuning, and closer reading of proposals than the steady state will need. The second cycle's usage is the one to size the next block from and to divide into cost per outcome.