Common definitions include: the customer did not reply within a set time, the customer clicked “this helped”, or the conversation closed without being handed to a person. Each of these can be true while the customer’s problem is not solved.
Why it matters
A customer who got a wrong answer and gave up looks exactly like a satisfied one in most resolution counts. Because the vendor sets the definition and does the counting, two tools’ rates cannot be compared, and neither can be checked from your side. For a small business this matters when the rate is used to justify switching off human review, or when pricing is charged per resolution.
A more checkable question is what happened to each reply: did a person send it unchanged, edit it, or throw it away? That is the sent-unchanged rate.
An example
A chat tool reports that it resolved most conversations last month. A look at the transcripts shows many ended after an answer that did not fit, with the customer later writing to the email inbox instead.
In detail: why AI resolution-rate claims mislead small businesses.