A containment rate is the share of conversations a chatbot finishes without handing them to a person. It does not tell you whether the customer's problem was solved. Resolution rate tries to measure that, and deflection measures something wider again. Vendors use all three words, each with its own definition, so a headline rate from one vendor cannot be compared with another's.
The short version
- Containment: the bot handled the conversation and no person joined.
- Resolution: the customer's problem was solved, either confirmed by the customer or assumed from silence.
- Deflection: the customer did not contact your team at all, often because a help article or bot answered first.
- Vendor rates are vendor claims with vendor definitions; several count silence as success.
- Measure your own rate on your own traffic, and count unhappy abandonment as a failure, not a win.
Three terms, three questions
Containment answers "did the bot keep this conversation away from my team?" A customer who gives up and closes the chat is contained. So is one who got a wrong answer and left. That is why containment on its own flatters a bot.
Resolution answers "was the customer's problem solved?" Some resolutions are confirmed, when the customer says the answer helped. Most are assumed, when the customer stops replying for a set time. An assumed resolution is a guess, and the length of the waiting window changes the number.
Deflection answers "did this person avoid contacting us?" It usually covers the help centre as well as the bot, and it is the hardest of the three to measure, because you are counting contacts that did not happen.
Why vendor rates are not comparable
The AI agent vendors that bill per resolution publish their own definitions, and they differ in ways that move the number.
- Intercom Fin bills per "outcome". An outcome includes a confirmed resolution and an assumed one, where the customer is silent for 24 hours after Fin's last answer. It also includes a handoff you configured inside a Procedure, which a person then finishes. Fin's headline claim is an average resolution rate of 76%. See the Intercom Fin review.
- HubSpot's Breeze Customer Agent counts a conversation as resolved when the agent shared a content source or performed an action and the visitor did not ask for a person within 72 hours; the status is judged 72 hours after the visitor's last message. HubSpot claims 70% on its product page, while its April 2026 press release said 65%. See the HubSpot review.
- Tidio quotes 67% for Lyro on its product page, but its own pricing-page calculator assumes a 55% "response rate". See the Tidio Lyro review.
- Voiceflow claims 70% on its pricing page. Botpress publishes no headline rate, though a customer story on its site reports 75% containment for hostifAI. Manychat and Chatfuel publish no resolution or containment rate.
Every figure above is a vendor claim or a customer story the vendor published. None is an independent measurement, and each uses a different window, a different denominator or a different word. A longer silence window, or counting configured handoffs, can make one product's rate look higher with no difference in how customers are served. Our how we review page explains how we grade this kind of evidence.
Why the definition also changes your bill
On per-resolution pricing, the definition is the billing unit. A customer who reads an answer and leaves without a word is billed as resolved by both Intercom Fin and HubSpot's agent, whether or not the answer helped. Our guide to what counts as a conversation compares the billing units side by side, and chatbot pricing models explains what each model means for a monthly bill. Head to head, the HubSpot vs Intercom Fin comparison shows how the two definitions play out.
How to measure your own rate
- Fix the denominator. Decide which conversations count. Leave out spam, test chats and greetings with no question, and write the rule down so it does not drift.
- Measure containment and handoff together. Containment is bot-only conversations divided by the denominator; handoff rate is the share that reached a person. They should add up to the whole.
- Split "contained" into three. Confirmed helped (the customer said so or clicked a positive rating), silent, and abandoned unhappily (negative rating, repeated question, rage words). Only the first is a proven resolution.
- Check for repeat contact. A customer who comes back about the same issue within a few days, by any channel, was not resolved the first time. Pick a window, such as a week, and keep it fixed.
- Read transcripts. Review a random sample every week and grade the answers as correct, partly correct, wrong or correctly handed off. This catches confident wrong answers that no rate will show.
- Compare with a baseline. Measure ticket volume and customer satisfaction before launch, so you can see whether the bot changed them.
Use the vendor's dashboard as a starting point, but export the raw conversations if the product allows it, so your definition is applied the same way every month and across any product you switch to. When the rate is low, the fix is usually content: see how to write a bot knowledge base.
Frequently asked questions
What is the difference between containment rate and resolution rate?
Containment rate is the share of chatbot conversations that ended without a person joining. Resolution rate is the share where the customer's problem was solved, confirmed by the customer or assumed from silence. A customer who gives up counts as contained but should not count as resolved.
What is a good chatbot containment rate?
There is no benchmark that holds across vendors, because each vendor defines its rate differently and the mix of questions differs by business. Measure your own rate with a fixed definition, compare it with your numbers before launch, and read transcripts to check that contained conversations were actually answered well.
Does Intercom Fin count silence as a resolution?
Yes. Intercom Fin's outcomes documentation treats a conversation as an assumed resolution when the customer stays silent for 24 hours after Fin's last answer, and it is billed. A handoff you configured inside a Procedure is also a billable outcome, even though a person finishes the job.
How does HubSpot decide a chatbot conversation is resolved?
HubSpot's help centre counts a Breeze Customer Agent conversation as resolved when the agent shared a content source or took an action and the visitor did not ask for a person within 72 hours of their last message. Each resolution costs US$0.50, so a visitor who reads a sourced answer and leaves is billed.
Keep reading
- What is the difference between a chatbot and an AI agent?guide
- Rule-based or AI chatbot: which should you build?guide
- How is chatbot software priced?guide
- What counts as a conversation in chatbot pricing?guide
- Best chatbot buildersbest list
Researched and drafted with AI assistance from the sources listed on this page. We have not built or tested bots on these platforms yet. Method: How we review