Containment rate

Containment rate is the percentage of conversations that an automated channel, such as an IVR, chatbot or AI agent, finishes without transferring the customer to a person.

How it is calculated. Containment rate is contained conversations divided by all conversations the automated channel handled. Microsoft’s Dynamics 365 documentation shows the split in its bot metrics: total bot conversations equal deflected conversations, resolved by the bot without escalating to a service representative, plus escalated conversations. Microsoft’s documentation also calls the same idea deflection. Copilot Studio’s metrics reference defines deflection rate as the share of incoming requests resolved through self-service instead of escalation to a person.

Contained and solved differ. A customer who gives up is not escalated, so a naive count treats them as contained. Copilot Studio tracks abandonment as its own outcome: engaged sessions that end after 60 minutes of inactivity without resolution or escalation. Keeping abandoned sessions out of the numerator stops the metric from rewarding dead ends. The CFPB’s 2023 report on chatbots in consumer finance describes customers caught in loops with no way to reach a person, which a pure containment target can encourage.

Read it with other metrics. Microsoft’s guidance for customer-facing agents groups containment with resolution rate and first contact resolution, and lists “no human handoff” as an anti-pattern: every agent should be able to pass the customer to a person with full context. A useful dashboard shows containment next to escalation rate, abandonment, first contact resolution and customer satisfaction.

Illustrative example. An AI agent handles 1,000 chats in a week. 700 end with the customer’s issue resolved, 200 are escalated to representatives and 100 are abandoned. Counting everything not escalated gives 80% containment; counting only resolved sessions gives 70%. The ten-point gap is the abandoned chats.

When the customer is an agent. If a customer’s AI agent sends a structured request to a business’s agent, less of the exchange depends on interpreting speech or free text. The requests that still need a person, such as policy exceptions, should escalate with the full exchange attached.

Neighbouring terms. Escalation is the complement of containment. First contact resolution checks whether a contained issue stayed solved.