Containment Rate: Definition and Formula

Containment rate is the share of support conversations that an automated system handles entirely on its own, without ever transferring to a human. A conversation is "contained" if it starts and ends within automation; it is not contained if it escalates to a person at any point. The term comes from voice IVR systems and has carried over to chatbots and AI agents.

Containment is closely related to deflection, and the two are often confused. Deflection usually asks whether a question was answered before a ticket was ever created; containment asks whether an active conversation the automation was already handling reached the end without a handoff. In practice both try to measure the same thing (how much work stayed with the machine) from slightly different angles.

The trap in the number

Containment rate has a well-known failure mode: a conversation can be "contained" because the customer gave up, not because they were helped. If someone abandons a frustrating chat, the system never escalated, so the interaction still counts as contained, while the customer walks away unhappy, or comes back through another channel. A high containment rate driven by dead ends is worse than a lower one driven by real resolutions, which is why the metric is dangerous read on its own. The tell is often in the data next door: a spike in repeat contacts or channel-switching shortly after "contained" chats is a sign the containment was hollow.

Reading it honestly

Containment should never be optimized in isolation. Push it too hard (by making escalation difficult or hiding the path to a human) and you inflate the number while degrading the experience. The metric earns its meaning only next to satisfaction and repeat-contact rate: containment that holds while CSAT holds is real; containment that rises while customers grow unhappy or contact you again is a false positive. There is no credible universal benchmark, so the honest figure is your own, tracked with those guardrails in view.

Containment rate in Evoriqa

Evoriqa is designed so that containment reflects resolution rather than abandonment. The agent answers from your knowledge and, crucially, hands off to a human the moment it cannot help rather than stonewalling the customer — so a contained conversation is one the AI genuinely finished, not one the customer abandoned. The analytics dashboard shows resolution and deflection alongside CSAT, and knowledge-gap detection reveals where conversations are failing, so you raise containment by giving the agent real answers, not by making the exit harder to find.