Metrics and terms
The difference between finished and resolved
Every support team wants this number and it is the hardest one in the category to produce honestly. It requires a judgement about a person's problem, made after the conversation, by somebody with the standing to make it. A widget observes none of that. This product does not compute a resolution rate.
not measured here
This product does not compute a resolution rate. No conversation is marked resolved, no visitor is followed up, and nothing links a later conversation to an earlier one. The insights page reports messages, refusals, cache share, latency, spend and a list of unanswered questions, and none of those is a claim that a problem ended.
What it means
A resolution rate is the share of contacts where the customer's problem was actually solved. It is the only metric in the support vocabulary that points directly at the thing anybody cares about, and that is exactly why it resists measurement: solving a problem is an outcome in the customer's life, not an event in a system. Ticketing tools approximate it by letting an agent mark something resolved, which measures an agent's opinion, or by asking the customer, which measures the small unrepresentative population who answer. Both approximations are visible and disclosed. A widget has neither. What it can see is that a conversation stopped, and a conversation stops for success, for abandonment, for a phone call, and for a bus arriving.
How it is actually calculated
The three approximations, and what each really measures
An agent marks it resolved: this measures agent judgement and closure discipline, and it drifts with workload, because a busy queue produces optimistic closures.
The customer confirms: this measures the answering population, which is small and self selected, and it is the only version with any claim to be about the outcome.
No reopening within a period: this measures silence, which conflates a solved problem with a customer who gave up and went elsewhere. It is the most common definition and the weakest.
Why none of the three works in a widget
There is no agent in the loop to make a judgement, no closure event to record one against, and no identity that persists so a reopening could be detected. A visitor with the same problem tomorrow starts a new conversation nobody can link to the old one.
Asking the visitor is possible in principle and is what the rating buttons do, and a rating is a feeling about a conversation rather than a verdict on whether a problem ended. Treating one as the other is how most published resolution rates in this category get built.
What would be needed to compute it here
Three additions, each with a real cost. A persistent identity for visitors so a return could be recognised, which the widget deliberately avoids. A closure step somebody actually completes. And a follow up asked late enough to be about the outcome rather than about the reply.
The follow up is the honest route and it is also the one nobody answers. That is not a reason to fake the number; it is the reason resolution is checked by sampling rather than reported as a rate.
How the number gets moved without anything improving
How a resolution figure gets built out of nothing
Define resolution as the absence of a follow up and the number becomes excellent as soon as your visitors stop bothering. Silence is scored as success, and the more discouraging the experience the more silence there is.
Define it as a positive rating and you have measured the delighted minority, then presented their opinion as a rate over everybody.
The accidental version happens inside honest teams: a proxy is adopted with a caption explaining its limits, the caption is dropped on the second slide, and by the quarterly review the proxy is being discussed as though somebody checked. Proxies survive; their disclaimers do not.
What to look at instead, or alongside
- A sample of conversations read against the source material, which answers the underlying question directly for a fixed number of cases.
- Whether a visitor asked the same thing twice in one conversation, which is a genuine signal that the first answer did not work.
- The unanswered questions list, which records what definitively was not resolved and what to do about it.
- Repeat contacts in your own email or phone channel about topics the assistant covers, which is where an unresolved problem actually shows up.
Questions
- Could you infer resolution from the conversation ending well?
- Only by guessing. A conversation that ends after a good answer looks identical to one that ends after a bad answer the visitor did not challenge. Anything inferred from an ending is a model of the visitor's mood, and putting a percentage on it would give a guess the appearance of a measurement.
- What do we report to somebody who insists on a resolution number?
- Report the sample. Say you checked forty conversations against the material, how many were right, and what the failures had in common. It is a smaller claim, it is defensible, and it is the only version anybody can act on.
- Does a low refusal rate mean things were resolved?
- No, and this is the substitution to watch for. A low refusal rate means the assistant answered. Whether those answers ended anyone's problem is a separate question, and the two come apart most sharply on exactly the material where accuracy matters.
Keep reading
- Deflection rate: our figure counts refusals, not resolutionsThe Deflection tile is the share of assistant messages that were not refusals. Support teams usually mean something wider by the word.
- Fallback rate: how often the assistant declines to answerThe refusal share is the other half of the Deflection tile. What triggers a fallback, what it costs, and why a low one is not automatically good.
- Coverage gaps: questions your material does not answerEvery refused question is a documented hole in your knowledge base, written by a real visitor. How the list is built and how to work through it.
- Every metric and termWhat each one means, and which of them this product actually computes.
Try it on your own material
Upload a document or point it at your site, paste one line of HTML, then ask it something only your business could answer.