Start from the labour it replaces
Don't price from your cost. Price from the function: a loaded support agent or reviewer costs $35k–$80k a year. An outcome price is a fraction of that and still far above your model cost.
Pick the unit the buyer already counts
Support counts resolved tickets. Insurance counts claims. IT counts tickets. Legal counts documents reviewed. Charge in the unit they already track — it removes the maths argument.
Set the base plus usage
A base platform fee ($199–$999/mo) plus per-outcome usage is the lowest-risk structure. The base covers the relationship; usage scales with value.
Meter it properly
Only bill outcomes that meet a success bar: no error, confidence above threshold, not routed to a human. Our outcome_meter.py does exactly this and emits invoice lines.
Want the full playbook?
The free 2026 AI Revenue Playbook turns these shifts into a plan with costs, scripts and pricing templates.