Resolution rate is the headline number, but not the first one to check
Published benchmarks put early deployments around 30–50% resolution, rising to 50–70% once workflows and the knowledge base mature. Anyone promising 90% in week one is quoting a deflection number dressed up as resolution — worth knowing before you compare your own dashboard against a claim like that.
A resolution rate that's low in the first two weeks isn't unusual and isn't necessarily a problem on its own. What matters is whether it's trending up as the knowledge base improves, or flat because the same gaps keep recurring unaddressed.
The questions the assistant couldn't answer are the most useful number on the page
This list is a ranked view of what your knowledge base is missing, drawn directly from real questions rather than a guess at what to write next. A question that appears repeatedly in this list and has an obvious, single answer is the cheapest resolution-rate improvement available — one article, applied to every future instance of that question.
A question that appears once or twice and is genuinely unusual isn't worth an article; the report is a guide for prioritising, not a mandate to document everything.
Telling a knowledge gap from a threshold set too tight
Both show up as a lower resolution rate, but they mean different things and need different fixes. If the same topic keeps appearing in the unanswered-questions list with genuinely different phrasings each time, that's a knowledge gap — the article doesn't exist yet, or doesn't cover the actual question being asked.
If instead you see a topic escalating even when the assistant's own confidence score was high, that points the other way: the knowledge base has the answer, but the threshold for that topic is set stricter than the answer actually warrants. Check the confidence scores on those specific transcripts before assuming either cause — guessing between the two usually means fixing the wrong thing first.
First reply time and resolution rate move together, up to a point
A faster first reply generally correlates with a higher resolution rate, because most of what's being answered quickly is the repetitive, well-documented majority of the queue — order status, account questions, straightforward how-tos. Once that majority is well covered, first reply time mostly stops moving, because what's left in the queue is the harder, more varied minority that was never going to be a one-line answer.
If first reply time is fast but resolution rate stays flat, that's usually the unanswered-questions list telling you the easy majority is covered and the remaining gap is the genuinely hard, low-volume tail — which is a different problem to solve than "the knowledge base is incomplete."
CSAT is the number to weight against the other two
A high resolution rate with declining satisfaction on those same resolved conversations is a signal worth taking seriously — it usually means answers are technically correct but arriving too generically, or a threshold is letting through answers that are right on average but wrong often enough to notice. Read a sample of the actual transcripts behind a resolution-rate number before treating it as a finished success; the number alone doesn't say whether the person on the other end felt helped.