A resolution is a customer conversation the AI agent closes without a human. That sounds simple, but the definition decides both your bill and the resolution rate a vendor advertises, and no two vendors define it the same way.
The short version
- Most vendors count a conversation as resolved when the customer stops replying, not when the customer says the problem is fixed.
- Some bill for handoffs to a human or for every conversation, resolved or not.
- The party counting resolutions is usually the vendor's own software.
- Fix the definition in the contract, and ask how disputed resolutions are credited.
Four ways vendors define it
| Definition | How it works | What to watch |
|---|---|---|
| Confirmed | The customer says the answer helped, or clicks a button. | The strictest test, and the rarest. Few customers bother to confirm. |
| Assumed | The customer does not reply within a set window, often 24 to 72 hours. | A customer who gave up is counted the same as one who was helped. |
| Verified by a model | A second AI model judges whether the conversation was resolved. | Better than silence alone, but the vendor is grading its own work. |
| Not escalated | Anything that did not reach a human counts. | This is really a containment measure. See resolution vs deflection. |
What it looks like in real price lists
- Intercom Fin charges US$0.99 per outcome. An outcome includes a resolution the customer confirms, a resolution assumed after the customer goes quiet, and a handoff to a human that you configured as part of a procedure.
- Zendesk changed its model in May 2026 so that only resolutions verified by a model count against your allowance; contained and escalated conversations do not. It does not publish the price of a resolution.
- Salesforce Agentforce sells a plan at US$2 per conversation, which bills whether or not the issue is resolved, alongside newer per-resolution and credit-based options.
- Freshworks Freddy bills US$0.49 per session. A session is consumed whether or not the customer's problem is solved.
- Ada sells prepaid automated resolutions that its own AI classifies, and its terms allow the count to be calculated from a sampled rate.
Why this matters more than the price per resolution
Two vendors can both charge a dollar per resolution and produce very different bills. If one counts silence after 24 hours and the other requires a model to verify the outcome, the first will report a higher resolution rate and charge you for more conversations on the same traffic. The headline rate and the unit price are both downstream of the definition.
What to put in the contract
- The written definition of a billable unit, including the silence window if there is one.
- Whether handoffs, abandoned chats, spam and test conversations are billable.
- What happens when a customer reopens a conversation that was billed as resolved.
- Your right to audit a sample of billed resolutions, and how errors are credited.
- Overage rules: what happens when you exceed the committed volume, and at what rate.
Frequently asked questions
Is an assumed resolution a bad thing?
Not by itself. Many customers who get a good answer simply leave. The problem is that you cannot tell them apart from customers who gave up, so measure customer satisfaction and repeat contacts next to the resolution rate.
Which definition is fairest to the buyer?
Confirmed resolution is the strictest, but it undercounts real successes. A model-verified resolution with a right to audit a sample is a reasonable middle ground. What matters most is that the definition is written down and you can check it.
Keep reading
- Resolution rate vs deflection rateguide
- Per-resolution vs per-seat pricingguide
- How to run an AI agent pilotguide
- Agentic customer service, explainedguide
- The best AI customer service agents in 2026best list
This page was researched and drafted with AI assistance from the sources listed on it. We have not run hands-on tests of these products. Method: How we review