Containment Is Table Stakes. Here Is What CX Leaders Actually Ask For

Containment rate measures how many callers never reach a human. Resolution rate measures how many callers got what they called for. For a decade the industry optimized the first number and called it success. The leaders I talk to every week have stopped doing that.

What are containment rate and resolution rate?

Containment is the share of calls an automated system handles without transferring to a person. Resolution is the share of calls where the caller's actual need was met, whether by the automation, a person, or both. A call that ends with the caller hanging up in frustration is contained. It is not resolved.

That gap is the whole story. Containment is a cost metric dressed up as an experience metric. It tells you how many people you kept away from your agents. It does not tell you whether that was good for them.

What do CX leaders say when you ask what they actually care about?

Two different customers said a version of the same thing to us this month.

• A roadside assistance executive at an auto club told us headcount savings were already banked and no longer the point. "This project is about member satisfaction," she said. Her team's target is a satisfaction score, and she would trade a few points of containment to hit it.
• A hospitality leader put it most bluntly. She is not concerned with containment, "because the guest could have hung up and not gotten what they needed."

None of them were rejecting AI. All of them were rejecting the way AI has been graded.

Consumers agree with them. In a September 2026 survey of just over 1,000 U.S. consumers, 69% said good service means their problem gets solved; only 12% named speed. And 57% said they do not care whether a human or an AI solves it, as long as it is solved. The same survey found 90% assume they will need a human before the issue is resolved.

Why did containment get gamed?

Because it was easy to measure and easy to improve. An IVR with no transfer option has 100% containment. A chatbot that says "I did not understand that" five times before offering an agent has excellent containment. Vendors priced on it, buyers reported it upward, and everyone agreed not to look too closely at what happened to the callers inside the number.

The cost shows up later and somewhere else: repeat contact rate, CSAT, churn, the agent queue that is somehow still full. By the time it is visible, the containment dashboard has already declared victory.

What should you ask a vendor instead?

Three questions, in order.

• What is your resolution rate by workflow, not blended? Blended numbers hide the one flow that is failing.
• What happens to the caller when the agent fails? A graceful handoff with context is a resolution path; a loop is not.
• Show me the calls you count as contained that did not resolve. If they cannot, they are not measuring it.

Why does this matter for your operation?

Every unresolved call that counts as contained tends to come back later as a repeat call, an escalation, or a complaint. Those costs land on the same team that was promised savings. Customer service works when the question the caller asked gets answered. Replicant builds each AI agent around completing the task, reports resolution per workflow, and shows what happened to every call that did not resolve. If you want to see resolution by workflow in your own operation, our team can walk you through it.

FAQ

What is a good containment rate for an AI agent?
It is the wrong question on its own. Containment ranges widely by use case (simple lookups can run above 70%, complex scheduling closer to 30%), but any containment figure should be read next to resolution for the same workflow. A high containment rate with low resolution means callers are being kept out, not helped.

Is resolution rate harder to measure than containment?
Slightly, because it requires defining what "resolved" means per workflow (a payment posted, a ride booked, an appointment confirmed). That definition is also what makes it useful. Modern conversation intelligence can score resolution on every call automatically.

Can an AI agent have high containment and low resolution?
Yes, and it is the most common failure pattern in the industry. An agent that never transfers but frequently fails to complete the task will show strong containment and weak resolution. The fix is to measure the second number and design the agent to hand off well when it needs to.

Does focusing on resolution mean more transfers to humans?
Sometimes, and that is fine. A transfer with full context that leads to a resolved call is a better outcome than a contained call that did not solve the problem. Over time, resolution-first design raises containment anyway, because the agent is built around completing tasks rather than avoiding handoffs.