Resolution rate for support teams and AI agents

To work it out, divide the problems solved by the problems customers raised, then multiply by 100. If your team got 600 tickets in March and solved 510 of them, your resolution rate was 85%.

For an AI agent, the same name means the share of conversations it solved without a person, often called AI resolution rate or automated resolution rate. Vendors that charge per resolution each decide what counts, and the prices on this page were read on 23 September 2026.

For a founder, what matters is how many customers get their problem solved without waiting on you. Deacon is an AI customer support agent that answers your customers from your own docs in seconds. Its plans charge a flat price for a set number of answers, never a fee per resolution, and there’s a free plan.

How to calculate resolution rate

The choices hide in the bottom line. Of the 600 tickets, 60 were still open at the end of March and 30 were waiting on the customer. Leave the 30 out and the rate is 510 ÷ 570, or 89%. Then in April, 25 of the solved tickets came back when the customer wrote again, so only 485 stayed solved, 81%.

In a helpdesk, someone on your team decides what’s solved. Zendesk’s help says agents set a ticket to Solved once they’ve resolved the issue. By default it closes 4 days later, and a reply after that opens a follow-up ticket instead. NiCE’s glossary calls the agent’s word “the simplest but least accurate” test. It rates asking the customer as better, and checking whether they came back about the same issue as “the most objective”.

Without a helpdesk, Gmail does the job. Give each customer thread a Solved label when you finish, take it off if the customer writes back, and at the end of the month divide the labelled threads by the threads that came in.

A ticket resolution rate barely separates one team from another, because almost every ticket gets marked solved in the end. In Freshworks’ 2025 benchmark report, covering more than 32,000 companies in 2024, the median company resolved almost 100% of its tickets. The reopen rate says more. It was 5% at the median company and 1% at the best 20%.

What an AI agent’s resolution rate is divided by

Intercom’s own example has 1,000 conversations in a month. Fin, its AI agent, was active in 750, got the chance to answer 500 and resolved 250. Fin’s resolution rate used to divide by the 750, for 33%. In July 2026, Intercom changed it to divide by the 500, for 50%. The share of all conversations Fin resolved, which Intercom calls its automation rate, stayed at 25%.

So the same 250 resolutions read as 25%, 33% or 50%. Ask which bottom line a vendor’s figure uses before you compare it with anything.

How Intercom, HubSpot, Zendesk and Ada decide a bot resolved it

Intercom calls a customer who leaves without asking for more help an assumed resolution, and bills it. So an answer that didn’t help, or a hallucination the bot made up, can count as resolved if the customer believes it and leaves. Ada’s guide to automated resolution sets the opposite rule. “A conversation that avoids a handoff but gives a wrong answer doesn’t count.”

What is a good resolution rate?

Vendors report 52 to 76% for their AI agents, each by its own rules. Asked about self-service in general, customers say far less gets solved. These are the published figures we found with a named sample, read on 23 September 2026.

We found no independent study that compares the vendors’ figures. Level AI’s glossary calls 70 to 79% solid and 80% or more world class, with no source. Those are SQM Group’s bands for first contact resolution, a different measure.

Resolution rate vs containment rate, deflection rate and first contact resolution

Take a customer who asks a bot how to get a refund, gets an answer that doesn’t fit their order, and closes the window. Nobody on your team hears about it. Here’s how each measure counts that chat.

Average resolution time is a different kind of measure, the hours from the customer’s first message to the fix.

Checking a resolution rate before you pay for it

Three questions tell you what a vendor’s resolution rate means, and with per-resolution pricing they also set your bill.

• What counts as resolved, and does a customer who simply leaves count?
• Is it divided by every conversation, or only the ones the bot got to answer?
• Does anything check back later for a customer who returns with the same problem?

These were the prices on 23 September 2026. Intercom charges $0.99 per outcome, one per conversation at most, and an outcome can be a resolution or a handoff Fin carries out through a Procedure you’ve set up. With Fin for Sales, a qualified prospect costs $9.99, or $0.99 when Fin routes them to an outcome you’ve marked self-serve. HubSpot’s customer agent costs $0.50 per resolved conversation from 14 April 2026, paid in HubSpot credits at 50 a resolution, on its Pro and Enterprise plans. Since 18 May 2026, Zendesk has counted only resolutions its language model verified against your allowance.

Resolving more questions with Deacon

Deacon charges for answers, not resolutions. An answer is a reply Deacon writes on your live site, and an honest “I don’t know” counts as one, so there’s no argument about what counted. Every plan answers with the same model, nothing is billed past your plan’s answers, and no plan upgrades itself. If a month’s answers run out, visitors get a short reply saying Deacon can’t look into it right now and are asked for their email, unless they’ve already left it in that conversation, and you can still read and reply to every conversation.

It answers from your own docs, in seconds and in the customer’s language. When your docs don’t cover a question, Deacon says it doesn’t know instead of guessing, and asks for the visitor’s email. When a visitor leaves their email or asks for a person, the owners on your team get an email about it, and you reply from the dashboard.

Then you close the gap. Every question your docs didn’t cover has an Answer this button. Your answer becomes part of what Deacon knows, so the next person who asks can get it in seconds, and Save & check shows you the reply Deacon now gives and how many other open questions your answer covers. The more you answer, the less Deacon needs you.

It also tells you what customers need. Topics group similar questions, count how many people asked and label most groups as a how-to, a bug report, a feature ask and so on, so you can see what to fix and what to build next.

Answered on Analytics › Widget isn’t a resolution rate either. It’s the share of questions Deacon answered out of those it tried, it counts a greeting or a request for a person as answered, and it leaves out replies that failed or arrived after your answers ran out. Deacon doesn’t ask customers whether their problem was solved, so it has no resolution rate to give you. Filter conversations by Answered, Unanswered or Wants a person to read how they went, and see chatbot analytics for the rest of the dashboard. You can try it all on the free plan, which needs no card.