Outcomes-Based Pricing
Outcomes-based pricing is a business model for AI agents in which customers pay per achieved outcome rather than per unit of compute consumed (tokens), per user (seat), or per time period (subscription).
The canonical implementation at Sierra: pricing per resolved customer interaction — an agent that closes a customer support case without human escalation. Containment rates of 50–90% are cited as the commercial range.
The incentive alignment argument
| Pricing model | What the vendor is incentivised to do |
|---|---|
| Token-based | Generate more tokens |
| Seat-based | Keep more seats occupied |
| Subscription | Minimise cost to serve |
| Outcomes-based | Achieve the customer’s actual goal |
Bret Taylor‘s argument: tokens are inputs, not value. The customer cares about resolved issues, booked appointments, completed transactions — not about how many tokens were consumed to produce them. Pricing per resolution creates direct incentive alignment between vendor and customer.
Practical requirements
Outcomes-based pricing is only viable when:
- Outcomes are measurable. “Resolution” must be operationally defined and verifiable (e.g., issue closed without escalation; transaction completed; appointment confirmed).
- Containment rates are high enough. If the agent only resolves 20% of interactions, per-resolution pricing makes the unit economics unstable.
- Edge cases can be priced separately. Interactions that escalate to humans require a separate pricing treatment (typically not charged, or charged at a lower rate).
Sierra’s practical threshold: below ~50% containment, outcomes-based pricing is not commercially viable for the vendor. Above 90%, the customer may question whether the pricing has become too expensive relative to human labour.
Relationship to SaaS transitions
Outcomes-based pricing for agents represents a structural shift analogous to SaaS replacing on-premises licensing:
- On-premises charged for software installation and maintenance, independent of usage.
- SaaS charged for access, closer to value but still not directly tied to outcomes.
- Outcomes-based charges for value delivered.
Each transition moved the pricing model closer to actual customer value. Outcomes-based pricing is the natural end state for agents because an agent’s value is entirely in what it accomplishes.
Challenges and critiques
- Measurement gaming. If “resolution” is self-reported by the agent or the vendor, there is an incentive to over-claim.
- Attribution complexity. Some outcomes require both the agent and a human — splitting credit is non-trivial.
- Adverse selection. Vendors may optimise for easy-to-resolve interactions (to maximise containment rate) rather than the full distribution of customer needs.
- Contract negotiation complexity. “Outcome” must be defined contractually; this adds friction to enterprise sales cycles.
See also
- Agentic Engineering
- Tool Use
- Bret Taylor on Sierra
- Evals — measuring outcomes is an evals problem