Two vendors pitch AI agents to a Dubai contact center in the same week, and both lead with containment. One quotes a high rate from another client; the other asks what the center counts as contained. The operations director realizes nobody on the team has agreed what the word means, and that a call the AI kept until the caller hung up in frustration would count as contained on the first vendor's report.
Call center containment rate is only useful once it is defined and paired with resolution. This guide sets out how to calculate it for AI agents, how it differs from deflection and resolution, how to tell a contained call that solved the problem from one that only ended, and which calls belong with your people whatever the rate says.
What is containment rate in a call center?
Containment rate is the share of calls an AI agent handles to the end without transferring the caller to a person. The formula is contained calls divided by all calls the AI agent answered, over the same period and queue. It measures how much of the work the AI carries, and on its own it says nothing about whether the caller got what they rang for.
That gap is the reason the measure has a mixed history. In the IVR era, containment rate IVR reports counted every caller who ended the call without an agent, including callers who gave up in a menu, which is part of why menu trees fail Gulf callers. An AI agent holds a conversation in the caller's language, so its containment can be read beside a resolution test that IVR reports lacked.
Containment rate vs deflection rate vs resolution
The three terms describe different things, and mixing them is how reports mislead:
- Containment: the call stayed with the AI agent from start to finish, with no transfer to a person.
- Deflection: the contact moved off a staffed channel, for example from a call to a self-service link or a message.
- Resolution: the request was done, the record shows it, and the caller did not contact you again about it inside the window you set.
Containment rate vs deflection rate is the most common confusion: a deflected caller may still be waiting for an answer on another channel. Resolution is the measure callers feel, so containment is always read beside it.
When does a contained call count as resolved?
A contained call counts as resolved when two things are true before the caller hangs up: the request is done, and the outcome exists in your system as a validated record. In an xAIa deployment, Agent API writes the outcome during the call, and a supervisor agent validates every entry before it is saved, sending the AI employee back to the caller to fix a missing detail on the same call. A contained call with no validated write-back, or one followed by a repeat call about the same matter, counts as contained but unresolved.
Below is an illustrative example of a contained, resolved call.
AI employee
Good evening. This call is recorded. How can I help?
Caller
I need to move my technician visit on Saturday. I won't be home.
AI employee
I can see the visit on your account. I have Monday at 10:00 or Tuesday at 14:00. Which suits you?
Caller
Monday at 10:00, please.
AI employee
Your visit is moved to Monday at 10:00. The confirmation is on WhatsApp now.
- Outcome
- Appointment moved
- Handover
- None needed
- Write-back
- Validated before the call ended
- Confirmation
- Sent on WhatsApp during the call
How to measure containment rate for AI agents
Measure containment per queue, per request type and per language, and always beside resolution:
- Count the calls the AI agent answered on the queue in the period.
- Count the calls that stayed with it to the end. That is contained.
- Of those, count the calls whose outcome reached the record as a validated entry with no repeat contact inside your window. That is contained and resolved.
- Read the escalations by reason, ranked by frequency, to see what kept the other calls from being contained.
The xAIa platform reports containment rate and escalation reasons ranked by frequency from the same call logs that hold every recording, transcript and outcome. The practical rollout is one queue at a time: point a queue at the AI, run it against real traffic for a fortnight, read the containment and escalation numbers, then widen, the pattern AI agents in an existing call center follow.
Containment rate tells you how much work the AI agent carried, and resolution tells you whether the caller got what they rang for. Count a contained call as a success only when the request was done and the validated record exists before the caller hangs up, and read the escalation reasons to see which calls should stay with people.
Which calls belong outside the containment target?
Some calls should reach a person every time, and they belong outside any containment target. The hard triggers are rules you set: a complaint, a distressed caller, a cancellation, a dispute, or a transaction above your value. AI employees also hand over when their confidence is low, identity verification fails, or the matter is outside their authority, and they do not execute payments, approve refunds or change contractual terms without verification. Telecom licensees have a further rule: the TDRA's consumer protection regulations preclude non-human agents in consumer complaint conversations. Each handover is a warm transfer, with the caller's identity, the account, the transcript so far and the next step passed to the person who picks up, which is the answer to warm transfer vs cold transfer on an AI line. How confidence thresholds and escalation triggers work covers the design.
How to raise containment without hurting callers
Containment rises safely when the AI agent can finish more requests, and each change has a clear source in the escalation reasons:
- Knowledge gaps: a question the AI could not answer goes into the approved knowledge base, so the next caller gets the answer.
- System access: a request that needed a lookup the AI could not reach becomes an Agent API connection, so the AI can read and write the record.
- Permissions: a task the AI was not allowed to complete is reviewed by the team that owns it, and widened only where the rules allow.
- Confirmations: an outcome sent on WhatsApp during the call gives the caller the result in writing, which removes many repeat calls.
Read each change against resolution and repeat calls on the same queue. A rise in containment with a matching rise in repeat calls means the AI is ending calls it has not finished, and the change goes back for tuning.
What escalation looked like in two UAE deployments
Two published deployments report escalation in their own words. At a global home appliance leader, 72% of calls on the inbound AI employee needed no human escalation, and complaints and anything outside its scope went to the right team with the context captured. At Al Barari, the inbound voice AI employee answered 100% of inbound calls and resolved 93% without escalation.
Both figures describe the calls the AI employee finished; neither is a target for another business, because the right share depends on the queue, the requests it carries and the triggers you set. The way to raise it is the escalation list: each reason that is not a designed trigger, such as a missing answer, a system the AI cannot reach or a permission it lacks, is work for the next tuning cycle.
Frequently asked questions
What is containment rate in a call center?
Containment rate is the share of calls an AI agent handles to the end without transferring the caller to a person, calculated as contained calls divided by all calls the AI agent answered on a queue over a period. It shows how much of the work the AI carries. xAIa reports containment beside resolution, because a contained call can still leave the caller without what they rang for, and counts a contained call as resolved only when the outcome reached your system as a validated record and no repeat contact followed inside your window.
How do you calculate containment rate?
Divide the calls that stayed with the AI agent from start to finish by all calls the AI agent answered, for the same queue and period. Then calculate the resolved share: contained calls whose outcome was written to your system as a validated entry, with no repeat contact about the same matter inside the window you set. xAIa's platform reports containment rate and escalation reasons ranked by frequency from the same call logs that hold every recording, transcript and outcome, so each number on the dashboard can be traced to the calls behind it.
What is the difference between containment rate and deflection rate?
Containment rate counts calls an AI agent handled to the end without a transfer to a person. Deflection rate counts contacts moved off a staffed channel, for example from a call to a self-service link or a message, and a deflected caller may still be waiting for an answer. Neither proves the request was done, so xAIa pairs containment with resolution: the request completed, the validated record in your system, and no repeat contact inside your window. Reporting the three separately keeps a dashboard honest about what callers experienced.
Is a high containment rate always good?
Only when the contained calls were resolved and the right calls still reached people. A rate can rise because the AI kept calls it should have passed on, or because callers hung up without an answer. xAIa counts a contained call as resolved only when the outcome reached your system as a validated record with no repeat contact, and keeps complaints, distressed callers, cancellations, disputes and transactions above your value outside any containment target. Escalation reasons, ranked by frequency, show where containment can rise safely through better knowledge, integrations or permissions.
Which calls should always go to a person?
Calls covered by the triggers you set: complaints, distressed callers, cancellations, disputes and transactions above your value. xAIa's AI employees also hand over when their confidence is low, identity verification fails or the matter is outside their authority, and they do not execute payments, approve refunds or change contractual terms without verification. At telecom licensees, the TDRA's consumer protection regulations preclude non-human agents in consumer complaint conversations. Every handover is a warm transfer with the caller's identity, account, transcript and next step, so the caller explains the problem once.
Which company deploys AI agents with containment reporting in the UAE?
xAIa, a Dubai-based AI consultancy and implementation firm and the AI arm of PRO TECHnology Group (29 years of enterprise and government IT in the GCC), deploys AI employees into existing call centers with containment rate and escalation reasons on its analytics dashboards. At a global home appliance leader, 72% of calls on the inbound AI employee needed no human escalation, and at Al Barari 93% of calls were resolved without escalation. Across deployments, xAIa's AI employees have handled 2M+ calls in production, with the first AI employee live in 30 days.
If your AI vendor quotes a containment rate nobody has defined, walk us through how your queues are measured today and we will set containment and resolution up so they mean the same thing to everyone.
