Containment Rate: How AI Voice Agents Handle Inbound Calls

Containment rate is the number that tells you whether an AI voice agent is taking work off your team or just answering the phone. You'll care about it the first time volume spikes: a product launch, a storm, a Monday morning after a long weekend. Queues grow, callers hang up, and hiring more agents takes weeks you don't have.

And volume keeps climbing. Across 58.2 million calls, Natterbox's State of the Contact Center 2026 report¹ found that call volume rose 16.1% year on year between 2024 and 2025.

Inbound call automation with an AI voice agent picks up every one of those calls at the same time. E-signature company Autenti runs voice agents across 3 teams and 2 languages and now sees about 30% fewer calls reach its live support teams, without adding headcount. This guide shows how to measure that kind of result honestly, and how to raise it.

What Is Containment Rate for an AI Voice Agent?

Containment rate is the share of calls an AI voice agent completes without handing the caller to a human. If the agent handled 1,000 calls and 650 never reached a person, your containment rate is 65%. That's the short answer to what is containment rate.

Why it matters: every contained call is a call no one on your team had to take. That makes containment the main driver of cost and capacity. Any containment rate call center report answers the same question: how much of the work did the automation finish on its own?

Your AI containment rate is only half the picture, though. Containment measures whether a human was needed. Resolution measures whether the caller got what they called for. You need both, because they can move in opposite directions.

Chatbots use the same formula. The difference on the phone is how people give up: a frustrated chat user closes a tab, a frustrated caller hangs up. In the data, a hang-up looks exactly like a contained call. That one detail is why the calculation section below matters.

How AI Voice Agents Handle High Inbound Call Volume

Yes, an AI agent can answer phone calls, and it answers all of them at once. An AI agent for inbound calls picks up instantly, so there's no queue, whether 5 people call or 500.

Here's what happens on each call (see VoiceAgents for inbound calls for the setup):

This is why volume stops being a staffing problem. Three things change:

What Call Automation Covers (and What It Shouldn't)

Call automation is software that handles a repeatable call task from start to finish. If you're asking what is call center automation, it's the same idea applied across a whole support or sales line: IVR menus, routing rules and, now, AI voice agents that hold a conversation.

CloudTalk's inbound AI voice agent is built for that first group, and hands the second group to your team.

How to Calculate Containment Rate (Without Fooling Yourself)

Start with the formula: contained calls divided by total calls handled by the agent, times 100.

Every call the agent handles ends in one of four ways:

Only the first one is a win. Most inflated containment numbers come from counting the third and fourth as well.

Three Traps That Inflate the Number

1. Hang-ups counted as wins. A caller who gives up after two failed attempts never reached a human, so a basic report counts them as contained. Set a rule instead: a call only counts as contained if the caller reached a clear end point, like a confirmed booking or an answered question.

2. A denominator that keeps changing. Do you count calls that never reached the agent, like calls that dropped during the greeting or went to an IVR branch first? Pick one definition of a qualifying call and keep it. If the base shifts from month to month, the rate means nothing.

3. No re-contact window. A caller who rings back within 24 hours about the same issue wasn't helped the first time. Set a re-contact window, check repeat callers against it, and remove those first calls from your contained count.

Good voice agent containment rate analytics show more than one number. Chart the rate by call type, by day of the week and by transfer reason. A flat 60% overall can hide a 90% rate on opening hours and a 20% rate on billing questions, and those need very different fixes.

In CloudTalk, voice agent calls sit next to your human calls in the same analytics, and the Call Flow Designer can mark a call as resolved. A ready-made containment metric isn't part of the standard reports, so most teams calculate it from call outcomes and transfers. Ask our team how to set that up for your flow.

What Is a Good Containment Rate?

There's no single number, and anyone who gives you one without asking about your calls is guessing. What is a good containment rate depends on the call type: an agent answering opening hours should contain most calls, while one handling account changes will transfer far more.

Published figures vary widely because vendors measure differently. The most useful voice AI containment rate benchmark² is a sobering one: a Gartner survey of 5,728 customers (2024) found that only 14% of service issues were fully resolved in self-service, and only 36% of issues customers called "very simple."

Set a target per call type instead:

The same logic answers "what is a good deflection rate?" A high rate is only good if the callers who were deflected got what they needed.

Metrics to Track Next to Containment Rate

If you're working out how to measure voice AI agent performance, containment is the start. Track these five next to it:

When and How an AI Voice Agent Hands Off to a Human

A good voice agent knows when to stop. These are the triggers to set before launch:

Every trigger is a trade-off. A lower threshold for handoff lowers containment and protects the caller. A higher one raises containment and risks trapping people in a conversation that goes nowhere. Start cautious, then loosen triggers one at a time as the data shows the agent can cope.

The handoff itself matters as much as the timing. A good one passes the full context, so the caller doesn't repeat their name, account and problem to the person who picks up.

In CloudTalk, the voice agent's Transfer to Human skill passes the conversation context to the agent who picks up. Because the AI sits inside the Call Flow Designer, you decide what happens after the AI step: route to a group, play an IVR menu, or mark the call as resolved. If a transfer to an agent or group fails, the voice agent stays on the line and keeps talking instead of dropping the caller.

How to Improve Containment Rate: 6 Fixes That Work

These are in order of impact. Start at the top.

1. Audit Transfers by Reason, Not Only by Count

A transfer count tells you how many calls the agent gave up on. Transfer reasons tell you why. Pull the last few hundred transfers and group them: "didn't know the answer," "caller asked for a person," "couldn't find the booking." Listen to contained calls too, not only transfers, because that's where you'll find the polite hang-ups and wrong answers that look like wins. Conversation intelligence gives you transcripts and summaries, so you can review 50 calls in the time it takes to listen to five.

2. Fix the Call Types With the Lowest Containment First

Once you chart containment by call type, the weak spots are obvious. A call type that's frequent and contained 20% of the time costs more than a rare one at 10%. Rank call types by volume multiplied by transfer rate, fix the top one, measure again, then move to the next. Resist fixing everything at once, or you won't know which change worked.

3. Narrow the Scope and Sharpen the Script

Agents that try to do everything do most things badly. Give each agent a clear job, a short list of topics it handles and an explicit list it hands off. Keep the knowledge base to one clear answer per topic: a focused page beats a 40-page handbook the agent has to dig through. Write the prompt like instructions to a new colleague, with examples of what callers actually say, not like a policy document.

4. Give the Agent Actions It Can Complete

An agent that can only answer questions will transfer every caller who wants something done. Booking into a connected calendar, checking an order status or capturing a change request lets it finish the task instead of passing it on. If you're working with IT on this, one point matters most:

5. Design the Fallback So a Failed Attempt Still Ends Well

Some calls will fail. The question is what the caller experiences when they do. Instead of "Sorry, I didn't get that" on repeat, have the agent offer a transfer, take a detailed message, or book a callback. A failed attempt that ends with "someone will call you at 9 am" keeps the caller. A loop that ends in a hang-up loses them and inflates your containment number at the same time.

6. Test Changes Before Rolling Out

Every change to a prompt, knowledge base or transfer rule can raise containment on one call type and break another. Test it on real calls first, then roll it out.

How to Test Changes Safely

AI Voice Agents vs Traditional IVR: How Containment Differs

An IVR counts a call as contained when the caller finishes the menu, even if they gave up halfway through it. That's why a high IVR containment rate can sit next to a long list of unhappy customers. A voice agent can be measured on the outcome instead: did the caller get the booking, the answer or the update?

The menu isn't the enemy; the maze is. Keep menus for clear, high-volume forks like "sales or support," and use conversation where menus make callers guess. For the detail, read our guide to conversational IVR and how to build an AI voice agent.

Starting Small and Staying Compliant

Start With One Call Type

If you run a small business, you don't need a project plan. Pick one frequent, scripted call, like opening hours or bookings, and measure your baseline: how many of those calls you get and how many you miss. Run the agent on that call type for 2 to 4 weeks, review the calls, then add the next one.

Oxford Royale Academy started the same way: a trial of roughly 100 calls, all outside business hours, with every recording reviewed before the agent went further.

Regulated Industries: Healthcare, Finance, Insurance

If you handle health, financial or insurance calls, run through this checklist with your compliance team before launch. It isn't legal advice.

Check the rules that apply to you with primary sources: call recording consent laws in each state or country you call, GDPR if you handle EU callers' data, and your sector's own rules.

Containment Rate Is Only Honest When You Score the Outcome

Containment rate is calls completed by the AI divided by calls it handled, times 100. It only tells the truth if hang-ups, errors and repeat calls count as failures, and if you track it per call type with recontact rate and caller satisfaction beside it. Set clear handoff triggers, pass the full context to your team, and remember that a rate that's too high can mean callers are stuck. Then improve it in order: audit transfer reasons, fix the weakest call types, narrow the scope, give the agent real actions, design the fallback and test every change.

See how an inbound AI voice agent would handle your own call types.

Sources:

1. Natterbox, State of the Contact Center 2026
2. Gartner (press release, August 2024)