What's the difference between containment rate and deflection rate?

They're closely related and often confused. Deflection rate usually measures the percentage of customers who solve their problem through self-service, including knowledge base, FAQ pages and search, before they reach an agent. Containment rate is more specific: it measures conversations that started in an AI agent or chatbot and got resolved there. All contained conversations are deflected, but not all deflected conversations are contained.

What containment rate should I aim for?

Depends on what you're automating. For focused, high-volume use cases like order tracking or password resets, a well-built AI agent can reach 70 to 80%. For broad, open-ended support, 30 to 50% is more realistic. Set the target based on the specific topic you're automating, not as a single company-wide number.

Can a high containment rate hurt customer experience?

Yes, if the rate is high because the bot is refusing to escalate, not because it's actually solving problems. The metric to watch alongside containment is post-conversation customer happiness. If containment is high but customer happiness is dropping, your bot is winning the metric and losing the customer.

Containment rate

Containment rate is the percentage of customer queries an AI agent or chatbot resolves end-to-end, without escalating to a human. It's the cleanest metric for whether your automation is actually doing its job.

What is containment rate?

Containment rate is the percentage of customer queries an AI agent or chatbot fully handles, without escalating to a human. It's the metric that tells you whether your automation is doing the job, or just adding another step before the human takes over.

If 100 customers start a conversation with your bot and 65 walk away with their question answered, your containment rate is 65%. The other 35 hit human support.

How to calculate containment rate

The simple version: total resolved bot conversations divided by total bot conversations.

The more honest version: only count a conversation as "contained" if the customer didn't come back within 24 hours with the same question. Bots that close conversations early but generate repeat tickets the next day are not actually deflecting anything, just delaying.

What's a good containment rate?

It depends on the use case. For high-volume, predictable queries (WISMO, returns, FAQs) a well-built AI agent can hit 70 to 80%. For broader support, 30 to 50% is more realistic. Anyone quoting 95%+ is either lying or counting differently.

How to improve containment rate

Pick the right use cases. Repetitive, predictable, low-stakes. Not emotional, not exception-heavy.
Train on real conversation data. Generic models with no exposure to your domain will underperform.
Build a clean escalation path. Counterintuitive: knowing when to hand over to a human improves containment, because customers who can escalate easily try the bot first.
Iterate on what didn't work. Look at the conversations that escalated. Most patterns repeat.

Join the Joy Lab

Neople's Substack all about AI and the teams that use it

Check it out

Share this post

https://www.neople.io/glossary/containment-rate

FAQs

Accuracy rate

Accuracy rate measures how often a system, such as a digital coworker, provides correct responses or completes tasks without errors.

Agentic AI

Agentic AI refers to AI systems capable of autonomous decision-making and taking actions to achieve specific goals without human intervention.

Agent utilization rate

Agent utilization rate measures the percentage of time customer service agents spend actively engaged in handling customer inquiries relative to their total available work time.