Units of work · outcome pricing
Buy work in units. Pay on the outcome.
No seats. No token meters. No hourly billing. Every unit of work has a fixed price and a guaranteed result — and you pay on that result. If a unit misses, we pay you.
UNITS OF WORK
Everything the team does is a priced unit of work.
A unit of work is the smallest piece of your operation with a definable outcome. Four shapes cover how work arrives. Figures live on your signed price list. Price placeholder
Unit of work · Monitoring
Continuous work on a defined scope — a category, a lane, a clinic schedule.
- Example
- e.g. a category plan watched 24/7, exceptions surfaced and proposed
- Trigger
- Scoped once, runs continuously
- Guaranteed outcome
- Stated coverage and response time, every day of the month
Unit of work · Incident
A discrete job triggered by a signal, worked to a close.
- Example
- e.g. root-cause on a quality defect, supplier miss, or margin leak
- Trigger
- Fired by a threshold breach or a request
- Guaranteed outcome
- Root cause found, fix proposed, or action taken inside the band
Unit of work · Session
Live work in a room where a decision gets made.
- Example
- e.g. scenario defense during a negotiation or an S&OP review
- Trigger
- Booked to a meeting on your calendar
- Guaranteed outcome
- The room leaves with a defensible decision and the reasoning on file
Unit of work · Cycle
Work tied to your operating calendar, delivered at its deadline.
- Example
- e.g. a planning cycle, a quarter, a month-end close
- Trigger
- Repeats on your calendar
- Guaranteed outcome
- Delivered complete and on time, graded against the stated result
The x100 problem
A chat costs pennies. An agent costs a hundred to two thousand times that.
Everyone priced AI like a chatbox: send a question, get an answer, pay for the words. Agentic teams do not work that way. A single unit of work runs reasoning loops, calls tools, retrieves documents, and hands context between members — and every step is metered in tokens you pay for. By the time the job is done, the token count is not ten times a chat. It is one hundred to two thousand times a chat. Token billing was built for the chat. It breaks on the team.
| Where the tokens go | What happens | Cost effect |
|---|---|---|
| Reasoning models think out loud | A model reasons internally before it answers, spending thousands of thinking tokens before one word of output. | 5–20x a standard model on the same task |
| Context is re-sent every call | LLMs are stateless. The whole conversation, system prompt, and retrieved documents are shipped again on every turn. | Compounds with every step |
| RAG stuffs the prompt | A five-token question is answered by injecting thousands of tokens of background documents into the prompt. | Input tokens dwarf the answer |
| Multi-agent handoffs | Each member passes its full context to the next. The same payload is paid for again at every handoff and every retry. | Token spend multiplies across members |
| Loops and retries | A planning member may call a specialist five times while iterating. Each call is a fresh, full-priced model run. | 3–10x prototype estimates in production |
The invoice that arrives is not ten dollars. It is four figures — and nobody can tell you which tokens produced the result and which produced heat. That is the x100 problem: token meters measure effort, not outcomes, and effort is exactly what agentic teams spend the most of. This is why we do not bill on tokens. We bill on the graded outcome.
Pay on outcome
How the money moves.
- STEP 01
Agree the unit and the result
Each unit of work is defined before it starts: scope, the guaranteed result, the price, and the remedy. One line per unit on the price list.
- STEP 02
The team works
Named members, bounded authority, every action in the ledger. You see the job run — you are not paying for a black box.
- STEP 03
The outcome is graded
Did the unit achieve the stated result? Yes or no, against criteria both parties agreed to before the work started.
- STEP 04
You pay on the outcome
Result achieved — the fixed price is due. Result missed — the remedy on the price list applies: we pay. No hourly meters. No token counting. No seats.
| Term | What it means |
|---|---|
| Unit of work | A named job with a defined scope, performed by named members of a bonded team. |
| Guaranteed outcome | The stated result of the unit, written so that both parties can tell whether it happened. |
| Outcome-based payment | Payment is due on the graded result, not on effort, hours, tokens, or seats. |
| Remedy | If the unit misses its stated result, we pay. The remedy is on the price list, not in a negotiation. |
Why we can sign
The books.
Everyone logs activity. We keep the books. Logs can't set an outcome price — books can.
Graded outcomes
Every unit is graded: did it achieve the stated result, or not.
Real money attached
Each graded outcome carries what it cost us and what we owed on a miss.
Standardized units
The same unit means the same thing across customers, so the numbers add up.
Consented pooling
Rates are pooled across organizations with consent — thousands of them. Only combined rates leave the pool.
Pool size: thousands of organizations — count TBC
Autonomy ladder
Authority is earned in the ledger.
- 01
Recommendations only
The team proposes. Every action needs a human yes.
- 02
Bounded autonomy
The team acts inside stated bands and escalates above them.
- 03
Expanded autonomy
Bands widen where the ledger shows the team has earned it. They narrow if it does not.
FAQ
Straight answers.
The guarantee
Everyone logs agent actions. We keep the books. Logs can't set a guaranteed price — books can.
Every competitor asks you to trust a demo. We ask you to read a contract.
Every digital worker is bonded
Each member of each team carries a bond. The team is accountable for the work, not just for the effort.
Fixed price, guaranteed result
Every job sits on a price list with a stated result. No seats. No token meters. If a job fails, we pay.
We keep the books
For every job we have ever run we hold what it cost, which model ran it, and whether it worked — pooled across thousands of organizations.