A support manager checks the dashboard on Monday morning and sees an 80% containment rate. That sounds promising until three customers contact the team again about the same issue. Chatbot containment rate tells you how many conversations stayed in the automated channel. It does not prove that customers got a useful answer, solved their issue, or felt satisfied.
What does chatbot containment rate mean?
Chatbot containment rate is the percentage of customer conversations completed in the automated channel without transfer to a human agent. It is also called a self-service rate in some reports. The related handoff rate shows the share transferred to a person. Escalation describes the transfer event or the reason for it.
The limit matters: containment measures channel ownership, not outcome quality. A conversation can remain with an AI agent because the customer got a complete answer. It can also remain contained because the customer stopped replying, left the site, or did not know how to request help.
Use containment as a channel-volume measure. Then check outcome metrics before judging support quality. A high percentage can mean your agent handles common questions well. It can also hide unanswered questions and abandoned conversations.
Aide makes the same distinction in its containment rate definition, while Fini describes the metric as a time-bound measure of conversations handled without a human. The practical difference is what your team counts as a conversation and when it decides that one is complete.
How to calculate chatbot containment rate
Use this formula:
Chatbot containment rate =
conversations completed without human involvement ÷ total chatbot conversations × 100
If 80 of 100 chatbot conversations end without a handoff, the containment rate is 80%.
That calculation is simple. The reporting rules are not. Define the denominator before you publish the number. One reasonable scope is every conversation that entered the website widget during the selected time window. A narrower scope can include only eligible support conversations, such as product, billing, or setup questions. Either choice can work. Mixing them from one report to the next cannot.
Write the definition beside the dashboard metric. Different transfer and abandonment rules can produce different rates from the same conversations.
| Conversation type | Reporting decision to document |
|---|---|
| Human handoff | Count as not contained when a team member takes over the conversation. |
| Abandoned conversation | Decide whether leaving before an answer counts in the denominator and apply the rule every time. |
| Duplicate sessions | Decide whether duplicate sessions are merged or counted separately. |
| Test chats | Exclude internal tests from production reporting. Keep the exclusion rule fixed. |
| Open conversations | Decide whether conversations still open at the end of the window remain pending or count as not contained. |
Use a fixed time window, such as a calendar week or month. Keep the conversation rules unchanged so you can compare reports. If you change the definition in April, label the break in the trend instead of presenting March and April as directly comparable.
A useful report might state: “Containment rate is the percentage of website-widget conversations during the month that ended without a human reply. Internal tests are excluded. Duplicate sessions are counted separately. Conversations open at month end are reported as pending.” The exact wording depends on your operation. The documented rule is what makes the number usable.
Do not change the target first when the rate moves sharply. Check the conversation logs. A sudden increase can come from a traffic mix change, a widget event, an inactivity rule, or a reporting change rather than better answers.
What containment rate tells you, and what it hides
A high containment rate tells you that more conversations stayed in the automated channel. That can reduce the volume reaching live chat. It does not tell you why those conversations ended.
Customers may leave after receiving a complete answer. They may also stop replying because the answer missed the question. Some may give up after finding no path to a person. If your report treats all three outcomes as contained, the rate combines success and failure.
Consider a shipping question. An agent gives a visitor the general shipping policy. The visitor leaves without asking for a human. The conversation is contained. If the visitor needed the status of a specific order, the shipping policy did not solve the issue. The channel stayed automated, but the outcome remains unclear.
A low rate is not automatically a failure. Billing disputes, account access problems, and emotionally charged complaints often need a person. A customer arguing about a charge may require judgment and context even when the knowledge base contains the refund policy. Routing that conversation to a human can be the correct result.
Do not set 100% containment as the goal. The target is appropriate automation with a safe path to a person when the agent cannot finish the job. Blocking handoff requests to protect a percentage is a poor support decision.
Training sources affect the rate because they affect answer coverage. A knowledge base with current product documentation can help an agent handle setup questions. Exact Q&A pairs can keep a refund rule or other sensitive wording from being paraphrased into something misleading. Better coverage may increase genuine self-service, but the percentage still needs an outcome check.
The metrics to read beside containment
Human handoff rate shows how often conversations require a person. Read it by topic. A high handoff rate for account access may be expected if the agent cannot verify identity. A high rate for basic setup questions may point to missing training content or unclear answers.
Repeat contacts show whether customers came back about the same issue within a defined follow-up window. Set the window before reviewing the trend. For example, you might connect a second conversation to the first when it comes from the same identified user within a stated period. Without a fixed rule, repeat-contact data becomes another subjective count.
CSAT adds the customer’s own view after a contained conversation. Use the same question, timing, and response rule each time. A CSAT score from a small or changing group should not be treated as a direct replacement for conversation review.
Resolution-quality review checks the conversations themselves. Sample contained conversations by topic, such as pricing, setup, refunds, account access, and order status. Ask whether the response addressed the customer’s actual request, used the right source, and gave the customer a clear next step. Record the result using the same review standard.
Containment by intent is more useful than one site-wide percentage. Break the report into the questions your customers ask. A single 80% rate can hide 95% containment for opening-hours questions and 35% for refunds. Those groups need different training and handoff rules.
The metric set should distinguish genuine self-service success from conversations that ended without a clear outcome. Containment tells you where the conversation ended. Repeat contacts, CSAT, and review tell you what that ending meant.
Where the measurement breaks
Denominator drift makes the trend unreliable. If abandoned sessions count in January but not February, the rate changed partly because the formula changed. The same problem appears when a team starts excluding duplicate sessions or open conversations without marking the reporting change.
Premature session closure can raise containment without proving resolution. An inactivity timer may close a conversation after the visitor stops typing. That event tells you the session ended. It does not tell you that the customer accepted the answer.
Hidden human involvement also distorts the number. A conversation should not count as contained if a team member answered through live chat, a shared inbox, or another support channel during the same issue. Decide how your systems connect these events. If they cannot connect them, say that the metric covers the widget channel only instead of calling it a resolution rate.
Unsafe automation produces a flattering number at the customer’s expense. An agent that avoids handoff requests or answers beyond its knowledge base may increase containment while increasing repeat contacts and complaints. The right response is to review those conversations and adjust the handoff rule or training source.
Missing identity and context limits what an agent can safely resolve. An account-specific question needs an approved way to identify the user before the agent can claim an outcome tied to that account. A general answer about shipping is not the same as checking one customer’s order.
When containment moves sharply, review conversation logs before changing the target. Look for a change in traffic, training content, handoff behavior, session closure, or reporting rules. The percentage is a signal to investigate, not a score to protect.
How the workflow looks with AssistLoop
AssistLoop gives you a concrete automated channel to measure. You train an AI support agent on company content, embed it in a website widget, let it answer customer questions, and configure human handoff for conversations that need a person.
Start with training sources and knowledge base setup. AssistLoop can use uploaded PDF, DOCX, and TXT files, website content, pasted text, and exact Q&A pairs. These sources give you different ways to cover common questions and pin down answers that should not be loosely rewritten.
Then configure human handoff. When a team member takes over, they can inspect the full conversation history. That context helps you separate a conversation that stayed automated from one that was properly transferred, and it gives you material for resolution-quality review.
The wider AssistLoop feature set includes conversation logs and support automation capabilities. Use those logs to inspect contained conversations by topic instead of relying only on the site-wide percentage.
The measurement boundary is simple. AssistLoop can show which conversations stayed automated and which were handed off. Containment still needs to sit beside repeat contacts, CSAT, and resolution review before you judge the outcome.
Keep usage volume separate from containment quality. Each user message and each AI response counts as one message credit. A busy period can use more credits without indicating better or worse support. Review AssistLoop pricing and message credits separately from the containment report.
Last verified: September 2026
AssistLoop is not the right fit if you need a full ticketing system or a ticket queue as the primary workflow. It is built around AI conversations in a website widget, human handoff, and actions connected to the support experience.
The useful report is not “Our chatbot contained 80%.” It is “Our chatbot contained 80%, repeat contacts fell for setup questions, and reviewed answers met our quality standard.” That sentence connects channel volume to customer outcome.
Create your AI agent and test the definition against your own conversation logs before setting a target.
FAQ
What does chatbot containment rate mean?
Chatbot containment rate is the percentage of customer conversations completed without transfer to a human agent. It measures where the conversation stayed, so it does not by itself prove customer satisfaction or issue resolution.
How do you calculate chatbot containment rate?
Divide conversations completed without human involvement by total chatbot conversations, then multiply by 100. If 80 of 100 conversations end without a handoff, the containment rate is 80%.
Is chatbot containment rate the same as chatbot resolution rate?
No. Containment records whether the conversation stayed with the chatbot, while resolution asks whether the customer’s issue was solved. A conversation can be contained because the customer left, stopped replying, or did not request a human.
What should you measure with chatbot containment rate?
Pair it with human handoff rate, repeat contacts, CSAT, and resolution-quality review. Together, these measures show whether containment represents genuine self-service or simply a conversation that ended without a clear outcome.
Written by