Containment Rate vs Resolution Rate: What Containment Really Measures

Containment rate is the share of conversations that ended inside an automated channel, such as an IVR menu, a chatbot, or an AI agent, without reaching a human. It tells you where a conversation stopped. It does not tell you whether the customer got what they needed, and that gap is why containment so often looks better on a dashboard than it feels to customers.

This post covers what containment means, how to calculate it, how it differs across IVR, chatbots and AI agents, why it flatters, and what to put in front of leadership instead. If you have already read our piece on deflection vs resolution, this is the same argument applied to the metric most contact centres and chatbot vendors lead with.

What containment rate means

The term comes from call centres. An IVR (interactive voice response) system “contains” a call when the caller completes their journey in the menu and never speaks to an agent. Checking a balance, paying a bill, or hearing opening hours are the classic contained calls.

Chatbots inherited the metric, and AI agents inherited it from chatbots. In every case the meaning is the same: the automated layer kept the conversation, and a human did not have to pick it up. Our glossary entry on containment rate has the short version.

The important word is “kept”. Containment is a statement about routing, not about outcome.

The containment rate formula

The basic formula is simple:

Containment rate = automated sessions with no transfer to a human / total automated sessions x 100

If 1,000 people start a chat with your bot and 300 are transferred to an agent, containment is 70 percent.

The trouble is in the numerator. Those 700 contained sessions include several very different outcomes:

• The customer got a correct answer and left happy.
• The customer got a wrong answer, believed it, and will find out later.
• The customer could not find a way to reach a person and closed the window.
• The customer gave up on chat and emailed or phoned instead.
• The session timed out because the customer walked away.

Only the first is a success. The formula treats all five identically. Before you calculate anything, write down whether abandoned and timed-out sessions are in or out, because that one decision moves the number more than any improvement you will make to the bot.

IVR vs chatbot vs AI agent containment

The same word covers three quite different systems, and the ways containment misleads differ between them.

IVR containment rate

IVR containment is the oldest version and the most mechanical. Callers press keys or speak short phrases, and the menu either completes the task or transfers. The main distortion is hang-ups: a caller who gives up in the third menu layer is often recorded as contained. Call centres have long paired IVR containment with repeat call rate for exactly this reason.

Chatbot containment rate

Rule-based and decision-tree chatbots behave like IVR in text. They answer a fixed set of questions well and everything else badly. The distortion here is usually design: bots that hide the “talk to a person” option, or loop customers back to the menu, raise containment by making escape harder. That improves the metric and damages the experience, which we cover in why customers hate your chatbot.

AI agent containment rate

AI agents built on large language models can answer a much wider range of questions from your documentation and take actions in connected systems. That makes containment more meaningful, because more contained sessions are genuine answers. It also introduces a new failure: a confident, fluent, wrong answer. A customer who accepts an incorrect answer is contained and satisfied in the moment, then returns days later with a bigger problem. Containment cannot see that at all.

Why containment flatters

Two effects inflate containment in almost every deployment.

Abandonment counts as success

A customer who closes the window has, by definition, not been transferred. Unless you explicitly exclude abandoned and timed-out sessions, they sit in the numerator alongside genuinely helped customers. The worse the automation is, the more people abandon, and the higher containment can climb. That is a metric moving in the opposite direction to the thing it is supposed to describe.

Channel switching is invisible

Containment is measured per channel. A customer contained in chat who then emails, phones, or opens a ticket through a web form has not been helped. They have been moved. The chat dashboard records a win, the email queue records a new ticket, and unless someone matches the two by customer and topic, nobody notices.

This is the same leakage that undermines ticket deflection, and it is why any containment number reported without a cross-channel check should be treated as an upper bound.

Containment vs deflection vs resolution

These three terms are often used interchangeably. They are not the same, and the differences matter when you are comparing vendors or reporting upwards.

Every genuine automated resolution is also contained and deflected. The reverse is not true. That asymmetry is the whole reason to prefer resolution rate as the headline number. We go into how to define a resolution in what counts as a resolution.

Containment benchmarks, and why we do not quote one

You will find plenty of “good containment rate” figures online. We have not found one from a reputable, recent, methodologically transparent source that we would stand behind, so we are not going to give you a number.

The reasons benchmarks vary so widely are worth understanding, because they apply to any figure a vendor quotes you:

• Definitions differ. Some include abandonment, some exclude it, some exclude sessions under a few seconds.
• Question mix differs. A business whose contacts are mostly order status will contain far more than one whose contacts are mostly complex technical faults.
• Human access differs. A bot with a visible “talk to a person” button will contain less than one without, and will often serve customers better.
• Channel differs. IVR, chat and messaging have different abandonment behaviour.

The useful comparison is your own containment over time, measured the same way, alongside the quality checks below.

How to measure good containment

Containment becomes a useful number once you attach two checks to it. Together they separate customers who were helped from customers who were merely contained.

Repeat contact within seven days

For every contained session, check whether the same customer contacted you again about the same topic within seven days, on any channel. Match by customer identity (email, account ID, or phone number) and by topic or intent, not by ticket thread.

A contained session with no repeat contact is strong evidence of a genuine outcome. A contained session followed by an email the next morning is a failure that containment scored as a success. Seven days is a sensible default. Some teams use fourteen for billing and account topics where problems surface later.

CSAT on contained sessions

Ask a single satisfaction question at the end of automated sessions and report CSAT for contained sessions separately from escalated ones. Blended CSAT hides the thing you want to know. Response rates on bot surveys are usually low and skew towards the very happy and the very annoyed, so treat this as a directional signal and read the comments, not just the score.

Put them together

A practical definition you can adopt this week:

Verified containment = contained sessions that were not abandoned, did not time out, and had no repeat contact on the same topic within seven days.

That is very close to a genuine resolution rate, and it cannot be improved by hiding the human option.

What to report to leadership

Leadership wants to know three things: are customers being helped, what does it cost, and is it getting better. Raw containment answers none of them reliably.

A short monthly report that does:

1. Verified resolution rate. The share of all contacts genuinely solved by automation, with abandonment, timeouts and handoffs excluded, verified by the absence of repeat contact.
2. Raw containment rate, as context. Shown next to verified resolution. The gap between the two is the size of the problem nobody sees.
3. Repeat contact rate. Across channels, by topic. If this is rising while containment rises, customers are giving up.
4. CSAT split by contained and escalated. Both should be healthy. A strong escalated score and a weak contained score means the automation is the problem.
5. Cost per resolution. Total automation and staffing cost divided by genuine resolutions. See cost per resolution.

If you only have space for one number, make it verified resolution rate. If you report containment at all, never report it alone. Our support KPI dashboard post has a fuller layout.

Where Fidiora fits

Fidiora is built around resolution rather than containment. It answers customer questions with AI, hands anything it cannot resolve to your team for free, and bills only for genuine resolutions: $0.69 each, dropping to $0.45 above 1,000 a month. Abandoned chats, silent timeouts and handoffs are never billed, which means the number on your invoice is the verified number, not the flattering one. There are no seat fees, and you can set a monthly spend cap. The full definition is on our resolution page.

The one change worth making

Take your current containment rate and remove abandoned and timed-out sessions. Then remove any customer who contacted you again about the same topic within seven days.

The number that remains is what your automation actually achieved. It will be lower, and it will be the first containment figure you can defend in a leadership meeting.