The economics of per-resolution pricing in AI customer support
Outcome pricing has become the standard model for AI support. It aligns incentives, but only if the buyer defines what a resolution is.
AI customer support vendors have largely converged on outcome pricing. Intercom charges $0.99 per resolution for Fin. Zendesk has moved its AI agents to outcome-based pricing. Sierra and Decagon negotiate per-conversation rates, often alongside substantial platform fees. The model is attractive in principle: the vendor is paid when the work is done, not when a licence is provisioned.
What a resolution is worth
A human-handled support ticket costs roughly $5 to $12 fully loaded in mid-market teams in the United States, accounting for salary, management, tooling, and shrinkage. In India the figure is lower but still material, typically Rs 80 to Rs 250 per ticket depending on channel and complexity. Against either baseline, a price of $0.49 to $0.99 per automated resolution leaves a wide margin, which is why the model spread quickly.
Deflection is not resolution
The central definitional question is the difference between deflection and resolution.
- Deflection means the bot produced an answer and the customer did not immediately continue. The customer may return tomorrow, on another channel, angrier.
- Resolution means the ticket was closed and stayed closed: no reopen within a defined window, and no follow-up ticket on the same issue.
Vendor marketing tends to quote the first number; case studies tend to reveal the second. Published case-study figures for end-to-end resolution generally fall between 40 and 55 percent, even where headline marketing claims 65 percent or more. Neither number is dishonest; they measure different things. The contract should measure the second.
How to evaluate an outcome-priced contract
| Item | What to require |
|---|---|
| Definition of resolved | Closed, no reopen within 7 days, same-topic follow-ups counted as reopens |
| Quality floor | CSAT on automated resolutions within a set range of human CSAT |
| Escalation handling | Escalated tickets are not billable, and arrive with case history attached |
| Platform fees | Quoted separately and compared across vendors; this is where headline prices diverge |
| Measurement access | Buyer can audit resolution counts from helpdesk data, not vendor dashboards alone |
A worked comparison
Consider a team handling 10,000 tickets a month at a loaded cost of Rs 150 per ticket, or Rs 15 lakh a month. An AI layer that genuinely resolves 45 percent of tickets at $0.69 (about Rs 58) per resolution costs about Rs 2.6 lakh for 4,500 resolutions, plus any platform fee. The human queue drops to 5,500 tickets. The arithmetic is favourable even at conservative resolution rates, provided the resolved tickets actually stay closed. If reopens run high, the same arithmetic quietly double-counts work, which is why the reopen window belongs in the contract.
Summary
Per-resolution pricing is a good model for buyers. It becomes a good deal when the contract defines resolution strictly, sets a quality floor, prices escalations at zero, and lets the buyer verify the count from their own helpdesk data.