In Fini (usefini.com), AI resolution rate counts only conversations with the status Resolved by AI, while Deflection rate is 100% - Human escalation rate and therefore also counts conversations still Waiting for Customer as deflected. Fini treats resolution rate as the headline metric and prices on resolutions, because a deflected conversation is not necessarily a solved one. Both metrics exist in the product and both are useful. This page gives the exact definitions, shows how far apart they can drift with a worked example, and explains when each one is the right number to look at.

​
The exact definitions

Every conversation the agent handles ends in exactly one of three statuses, and the three sum to 100% of conversations in the selected window:
  • Resolved by AI: the agent resolved the conversation without a human.
  • Escalated to Human Team: the agent handed the conversation to a teammate.
  • Waiting for Customer: the agent replied and is waiting on the customer.
Put another way:
The same split as a picture. Deflection is the outer box; resolution is only the part of it that actually ended resolved (example figures, matching Agent A in the worked example below): The definitions come straight from the Analytics page, where the Deflection rate and Human escalation rate KPI cards sit at the top, the Resolution rate trend plots daily resolution alongside the Waiting for Customer share, and the Conversation Status doughnut shows the three-way split.

​
A worked example

Two agents each handle 1,000 conversations in a week. Both escalate 200 of them to humans. On deflection, the two agents look identical. On resolution, Agent A is 15 points ahead. Agent B is leaving a quarter of its conversations open: customers who were asked a clarifying question and never answered, or who got an answer that didn’t settle the issue and went quiet. Some of those customers will come back through another channel, which deflection never sees. Try it with your own numbers. The defaults are Agent A; enter Agent B (550, 200, 250) to watch deflection stay at 80% while resolution drops:
When deflection rate and AI resolution rate drift apart, open Inbox and filter Conversation status to Waiting for Customer. Read ten of those conversations. A pattern of unnecessary clarifying questions usually points to a Rulebook intent gate that is too strict or a knowledge gap the agent is covering with questions.

​
Why Fini leads with resolution

A support team’s job is to solve the customer’s problem, not to keep it away from a human. Deflection measures the second thing. It goes up when the agent answers correctly, and it also goes up when the agent asks a question the customer abandons, gives an answer the customer doesn’t trust, or simply keeps the customer from reaching a person. Resolution rate only goes up when the conversation actually ends resolved. This is also why Fini’s pricing is tied to resolutions rather than conversations or seats. Fini’s pricing is usage-based, per resolution, where a resolution is an issue solved end to end with no human handover. If you pay for resolutions, the vendor’s incentive and yours point the same way: more conversations genuinely solved. See What counts as a resolution for the billing definition, including where the Resolved by AI count and the bill can differ, and Pricing for current rates.
To confirm (internal, remove before publish): Is the billed resolution count the same as the count of conversations with status Resolved by AI in Analytics for the same period? If there are differences (for example, reopened conversations, conversations that later escalate, or test and replay traffic), list them so this page and the billing page say the same thing.
For a longer discussion of the two metrics outside the product, see Fini’s article Deflection rate vs resolution rate.

​
When deflection is still useful

Deflection is not a wrong number, it answers a narrower question: how much load reached your human team? That is the right lens in a few situations. The rule of thumb: report AI resolution rate to stakeholders as the measure of how well the agent works, and use deflection and Human escalation rate to plan your team’s workload.

​
Reading the gap over time

The gap between deflection and resolution is your Waiting for Customer share. Track it as its own number:
  • A stable, small gap is normal. Some customers always go quiet after getting their answer.
  • A growing gap after a change often means the agent started asking more questions or giving less complete answers. Compare the period before and after the change with the steps in Measure your resolution rate.
  • A large gap on one intent rule shows up in the Intent rule breakdown, where AI Resolve Rate is low but Escalated Rate is also low. That rule is leaving conversations open rather than resolving or handing off.

How accuracy is measured

What the benchmark numbers mean and how to grade accuracy on your own traffic.

Measure your resolution rate

Step-by-step in Analytics and through the API.

What counts as a resolution

The billing definition of a resolution.

Analytics

KPI cards, the Conversation Status doughnut, and escalation reasons.