Resolution rate in helpdesks and AI support tools

The name covers two numbers. In a helpdesk such as Zendesk or Freshdesk, it is the share of tickets your team marked as solved. For a chatbot it is usually called AI resolution rate or automated resolution rate, the share of conversations the bot solved alone.

Resolution rate for a month of tickets

A software company’s team received 600 tickets in March. It solved 480 of them that month, so its resolution rate was 80%. Of the other 120, 70 were still being worked on and 50 were waiting for the customer to reply. Leave the 50 out and the rate is 480 ÷ 550, or 87%. In April, 30 of the solved tickets came back when the customer wrote again, so only 450 stayed solved, 75%.

In a helpdesk, someone on your team decides. Zendesk’s help article on solved tickets says agents set a ticket to Solved after they resolve the issue. By default the ticket closes for good four days later, and a reply after that opens a new ticket. Deciding by the agent’s word is the simplest method and the least accurate, according to NiCE’s resolution rate glossary. It rates two methods above it, asking the customer and checking whether they came back.

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

What a bot’s resolution rate is divided by

An online shop’s chat widget had 1,200 conversations in March, and its bot got to answer 900 of them and resolved 540. Divided by the 900 it got to answer, its resolution rate is 60%, and divided by all 1,200 it is 45%.

Intercom calls the first Fin’s resolution rate. The second is its automation rate. Intercom’s update to its Fin metrics in July 2026 left out the conversations its bot never got to answer. The worked example in the update went from 250 ÷ 750, or 33%, to 250 ÷ 500, or 50%, with the same 250 resolutions.

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

Intercom charges for both kinds of resolution in its row, including a customer who leaves after an answer that did not help. A hallucination, which is an answer the bot makes up, can pass the Intercom and HubSpot tests if the customer believes it and leaves.

Resolution rates published by Intercom, HubSpot and Ada

For AI agents, each figure below is the vendor’s own number, worked out by its own rules. We found no independent study that compares them, and we read these on 19 September 2026.

Customers asked about self-service in general report a much lower number. In Gartner’s survey of 5,728 customers in December 2023, 14% of issues were fully resolved there.

We found no benchmark for ticket resolution rates. Level AI’s glossary calls 70 to 79% solid and 80% or higher world-class. Those are SQM Group’s bands for first contact resolution, a different measure.

Working through the unresolved conversations

Sort the conversations that ended unresolved by why they ended that way. For the online shop in March, that means the 360 conversations its bot answered without resolving.

• The bot had no answer in your documentation.
• The customer needed something only a person can do, such as a refund.
• The customer stopped replying.
• The problem was a bug in your product.

Only the first is fixed by writing. The second is a handover done right, even though Ada’s reports mark every handover as not resolved. Intercom’s July 2026 change raised a resolution rate by 17 points without solving one more problem.

Deacon marks each question it could not answer in your conversations, and an answer you write there joins the documentation it answers from.

One refund question, counted four ways

Take a customer who asks the bot how to get a refund, gets an answer that does not fit their case, and closes the window. Nobody on your team sees the conversation.

Under the looser deflection definitions, that is a deflection. The bot never handed over, so the containment rate counts it as contained, and Intercom would bill it as an assumed resolution. A resolution rate that checks whether the problem was solved should fail it.

Resolution time is a different measure, the hours between the customer’s first message and the fix.

Checking a resolution rate you are quoted

Three questions tell you what a vendor’s resolution rate means.

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

With per-resolution pricing, the answers set your bill. These were the prices on 19 September 2026. Intercom charges 99 cents per outcome, and an outcome is a resolution or, in some set-ups, a handover. Since May 2026 Zendesk’s AI agents bill only for resolutions its language model has verified. HubSpot’s customer agent costs 50 cents per resolved conversation from April 2026.

Deacon’s plans are a flat monthly price with a set number of answers, whatever share of conversations end up resolved.