The resolution rate measures the share of conversations in which the visitor's request was settled. For an AI assistant, a distinction is usually made between the share settled by the assistant alone, without the team stepping in, and the share settled after a person took over. It is the most widely used indicator for judging whether an assistant really helps.
How it is calculated
The basic formula is simple: the number of resolved conversations divided by the number of closed conversations, over a given period. Everything depends on the definition of "resolved", which varies between tools:
- Closed without human handover: the conversation ended without the team stepping in. This is the broadest definition, and the most flattering, because a visitor who gives up also counts as "resolved"
- No return within a set period: the visitor did not come back with the same question in the following days
- Confirmed by the visitor: they indicated that the answer helped, with a thumbs up or a closing question
- Judged resolved on analysis: an automatic summary or a human review considers that the request received a useful answer
When a tool bills "per resolution", the definition used directly determines the amount of the bill. Always ask for it in writing.
Resolved by AI, by the team, or unresolved
A useful reading distinguishes three outcomes for each closed conversation:
| Outcome | What it means | What it says about the assistant |
|---|---|---|
| Resolved by the assistant | The documented answer was enough | The documents and the prompt cover this need |
| Resolved by a person | The team took over and settled the request | The human handover worked |
| Unresolved | No one settled the request | A document is missing, a handover failed or the visitor left |
The third row is the most instructive. It points to unanswered questions, handovers left waiting and abandoned conversations. That is where the improvements lie.
What rate should you aim for?
There is no universal right figure. The rate depends on your business, the scope given to the assistant and your handover choices. An assistant that mainly answers practical questions can settle a large majority of conversations on its own. An assistant for an accounting firm, which must pass on every question about an individual file, will have a lower rate, and that is normal.
What matters is the trend. After launch, the assistant's resolution rate should rise as the documents are completed, then level off. A sudden drop often signals a change in the offer that the documents did not follow.
The pitfalls of a rate that looks too good
A high resolution rate is not always good news:
- The assistant no longer offers a handover. Visitors give up instead of being passed on, and the conversation counts as "resolved"
- The answers are wrong but confident. A convincing hallucination closes a conversation just as well as a correct answer
- Sensitive topics are handled by the assistant. A complaint "resolved" with a generic answer will come back later, often more tense
That is why the resolution rate is always read alongside other indicators: a sample of answers reviewed for accuracy, the share of conversations with a negative tone, the number of unanswered questions and the team's handover time.
How to improve it honestly
- Read the unresolved conversations every week and note the cause
- Complete the documents when information was missing, after approval by a person in charge
- Clarify ambiguous documents when the assistant answered beside the point
- Adjust the prompt when the assistant declined to answer when it could have, or the reverse
- Keep the human handover on reserved topics, even if it lowers the figure
The goal is not a maximum rate, but a rate that reflects correct answers on the topics the assistant should handle, and clean handovers on the others.
Resolution rate and first contact resolution
Contact centres have long tracked first contact resolution: the share of requests settled at the first exchange, without call-backs or reminders. For an assistant, the notion is similar: a request settled in the first conversation, without the visitor having to come back. An assistant that passes on a complete request, with all the useful information, also helps the team settle it at the first contact.
The resolution rate in KernChat
In KernChat, each closed conversation gets one of three outcomes: resolved by the assistant, resolved by a human or unresolved. The outcome takes into account any handover and the automatic summary written at closing. The workspace dashboards show the assistant's resolution rate, calculated on closed conversations, and let you find the unresolved conversations to review them.