Chatbot Resolution Rate· Customer-service metric
Chatbot Resolution Rate — Definition, Formula, and Healthy Ranges (2026)
Quick answer: Resolution rate is the percentage of conversations the bot fully solved on its own — not just the ones it managed to keep away from a human. That distinction is the whole point. A containment rate counts every chat that did not escalate, including the ones where the customer gave up and walked away; a true resolution rate counts only the chats where the issue was actually settled, which means it needs evidence of a solved problem rather than the mere absence of an escalation. Read it as the most demanding number in your self-service stack: if your platform calls a silent quit a "resolution," your resolution rate is really a containment rate wearing a more flattering label.
What it is
Resolution rate measures how often a customer-service chatbot brings a conversation to a genuine close — the user's question was answered, the task was completed, and they did not need a person to finish the job. Written out:
Resolution rate = bot-resolved conversations / total bot conversations
The arithmetic is trivial. The entire difficulty lives in the numerator: what earns a conversation the label "resolved." Done rigorously, a resolution requires a positive signal — the user confirmed the answer helped, the order actually shipped, the flow reached a real end state, or a post-chat rating came back positive. Done loosely, "resolved" collapses into "did not escalate and did not reopen within a few days," which is a very different and much weaker claim. The gap between those two definitions is exactly where resolution rate either becomes your most valuable metric or quietly turns into vanity.
How "resolved" gets defined (and mis-defined)
There is no universal definition of a resolved chatbot conversation, so every platform picks one, and the choice decides what the number means. Three common definitions, from strongest to weakest:
- Confirmed resolution. The user explicitly says the answer worked, or a downstream system confirms the outcome (payment captured, ticket closed by the customer, "yes that solved it" tapped). This is the honest definition and the hardest to collect.
- Outcome-inferred resolution. No explicit confirmation, but the flow reached a terminal success node and the user did not reopen or escalate inside a defined window. Reasonable, but it credits some chats the customer abandoned.
- Absence-of-escalation resolution. Anything that did not hand off to a human counts. This is not really a resolution rate at all — it is a containment rate, and treating it as resolution is the single most common way the metric lies.
The practical rule: pin down which definition your tool uses before you trust the figure, and prefer the strongest one your data supports. A resolution rate built on confirmed or outcome-inferred signals is a real quality metric; one built on absence-of-escalation is just containment with better marketing.
Resolution rate versus containment and deflection
These three numbers get used interchangeably and they are not the same, which is why a dashboard can show a glowing self-service story while customers quietly leave unhelped. The cleanest way to hold them apart is by what each one actually requires:
| Metric | What it counts | What it ignores |
|---|---|---|
| Deflection rate | Conversations kept off human channels entirely | Whether the question was answered at all |
| Containment rate | Conversations that did not escalate to a human | Whether the user solved their problem or just gave up |
| Resolution rate | Conversations the bot genuinely solved, with evidence | Nothing it should — it is the strict one |
The relationship is nested. Deflection is the loosest: a chat can be "deflected" simply because the user never reached a human, even if they rage-quit. Containment is tighter but still credits the abandoned chats — the ones where someone closed the tab mid-flow counts as "contained" because no escalation fired. Resolution is meant to strip those out and keep only the genuine wins. So in an honestly measured stack, resolution rate sits below containment, which sits below deflection. If your resolution rate equals your containment rate, you are almost certainly not subtracting the abandoned chats, and the number is inflated by exactly the failures it is supposed to expose. This is the same blind spot the abandonment rate exists to catch, read from the other direction.
Resolution rate versus resolution time
Resolution rate and resolution time sound like a pair and measure different things. Resolution rate asks how often the bot solves the problem; resolution time asks how long it takes when it does. A bot can post a strong resolution rate and a terrible resolution time — it gets there eventually, but only after eight frustrating turns — or a fast resolution time on the small slice of chats it resolves while failing most of the rest. Read them together. Rate without time describes how often you win; time without rate describes how fast you win the cases you do. The healthy target is a resolution rate high enough to justify the deployment and a resolution time short enough that the wins do not feel like a slog. Neither number alone tells you whether the bot is pulling its weight; that is what the full metrics guide is for.
What counts as healthy (2026)
There is no single published resolution-rate benchmark, partly because vendors define "resolved" so differently that the figures are not comparable across tools. The bands below are editorial working figures, deliberately set a few points under the containment-rate ranges we use elsewhere — because an honest resolution rate excludes the abandoned chats that containment counts. They assume an outcome-inferred-or-stronger definition, and they are directional, not guarantees:
| Bot type | Healthy resolution rate | Reading |
|---|---|---|
| Rule-based / decision-tree FAQ bot | 10-16% | Narrow scope; resolves only the exact questions it was scripted for |
| NLU intent bot | 18-30% | Handles a real intent library; resolution depends on knowledge coverage |
| LLM + retrieval (RAG) bot | 30-45% | Grounded answers lift genuine resolution, provided the knowledge base is good |
| Mature, well-tuned LLM deployment | 40-55% | Premium ceiling for general support; higher usually means a loose definition |
Two forces move these bands more than the platform brand does. The first is knowledge coverage: resolution is gated by whether the answer actually exists in the bot's knowledge, so a thin or stale knowledge base caps resolution no matter how capable the model. The second is scope honesty: a bot pointed only at questions it can answer will post a higher resolution rate than one asked to field everything, so a rising resolution rate can mean better coverage or a narrower remit — check which before celebrating. And treat any general-support resolution rate above the high-fifties with suspicion: it usually signals an absence-of-escalation definition quietly counting give-ups as wins, not a bot that solves four out of five problems.
How platforms expose it
Whether you can even measure resolution honestly depends on the platform class. Support-desk products such as Intercom and Tidio report a bot resolution (or automated-resolution) figure natively and let you separate bot-resolved from agent-resolved conversations, which is what makes resolution distinguishable from containment without exporting transcripts. The thing to check is their definition: confirm whether a "resolution" requires a positive user signal or just the absence of a reopen, because the marketing headline rarely says. Flow-first builders like Manychat and SendPulse treat a solved request as a flow reaching its end node, so "resolution" is whatever terminal state you wired — you decide what counts, and you assemble the reporting around it. Developer-grade builders such as Botpress let you fire an explicit resolution event with your own success criteria, which is the clean way to build a confirmed-resolution metric rather than inferring one.
Whatever the surface, the question to put to a platform is not "what is your resolution rate" — every vendor will quote a flattering one — but "how do you define a resolved conversation, and can I see resolution separately from containment and from handoff?" A tool that reports one blended self-service number is telling you chats did not escalate, which you already suspected. A tool that reports resolution against containment, escalation, and abandonment is telling you whether customers were actually helped, and that is the number worth defending on a dashboard.
Related terms
- Chatbot containment rate — the looser sibling that counts non-escalation; resolution is what is left after you subtract the give-ups.
- Chatbot deflection rate — the loosest self-service number, measuring chats kept off human channels regardless of outcome.
- Chatbot abandonment rate — the failures resolution must exclude and containment wrongly credits.
- Human handoff — the escape hatch a resolution avoids; a clean handoff is a non-resolution that still ends well.
- Customer service chatbot — the bot category resolution rate applies to.
FAQ
What is a good chatbot resolution rate?
It depends on the bot type and, more than anything, on how "resolved" is defined. As directional working figures for an honest, outcome-based definition: a rule-based FAQ bot lands around 10-16%, an NLU intent bot around 18-30%, an LLM bot with retrieval around 30-45%, and a mature, well-tuned LLM deployment around 40-55% for general support. Treat anything much higher with suspicion — it usually means the metric is counting non-escalation as resolution, which makes it a containment rate, not a resolution rate.
Is resolution rate the same as containment rate?
No, and conflating them is the most common measurement error here. Containment counts every conversation that did not escalate to a human, including the ones the customer abandoned. Resolution counts only the conversations the bot genuinely solved, with some positive evidence. In an honest stack resolution sits below containment, because containment credits the abandoned chats that resolution strips out. If the two numbers are equal, your resolution rate is almost certainly inflated.
How is resolution rate calculated?
The formula is bot-resolved conversations divided by total bot conversations. The arithmetic is easy; the judgment is in the numerator. A confirmed-resolution definition requires a positive signal (the user said it helped, or a downstream system confirmed the outcome). An outcome-inferred definition accepts a terminal success node with no reopen or escalation in a set window. An absence-of-escalation definition counts anything that did not hand off — which is really containment. Pick the strongest definition your data supports and state which one you are using.
Why is my chatbot's resolution rate lower than its containment rate?
That is the healthy and expected relationship — it means your measurement is honest. Containment counts non-escalated chats including give-ups; resolution excludes the give-ups and keeps only genuine solves, so it should sit a few points lower. If your resolution rate instead matches or exceeds containment, the definition is too loose and is folding abandoned conversations into the win column.
How do I improve a low chatbot resolution rate?
Resolution is gated by knowledge coverage, so the fastest lever is usually content: find the topics where the bot reaches the right intent but cannot produce a correct answer, and close those gaps. After coverage, check scope (is the bot being asked questions it was never meant to handle?) and the quality of the handoff for the cases it should escalate rather than guess. The metrics guide and the companion improve-resolution playbook walk the full diagnostic loop.
Sources
- Intercom. Documentation — automated resolutions and bot performance reporting. intercom.com/help (verified 18 June 2026).
- Tidio. Help center — bot resolution and analytics. tidio.com/help (verified 18 June 2026).
- Botpress. Documentation — analytics and custom events. botpress.com/docs (verified 18 June 2026).
- Chatbotscape Glossary. Chatbot containment rate. /glossary/chatbot-containment-rate (verified 18 June 2026).
- Chatbotscape evaluation methodology. /methodology (continuously updated).