IVR Containment Rate: Formula, What Good Looks Like, and Why It Misleads

IVR containment rate is the percentage of calls that enter your interactive voice response system and end there without being transferred to a live agent. The raw formula is simple: calls not transferred, divided by calls that entered the IVR. The problem is what that number includes. Callers who hung up in frustration and callers who rang back the next day count as contained, so a useful measurement removes both. That corrected figure is usually called true containment.

This post is the IVR-specific companion to our broader piece on containment rate vs resolution rate. That article argues why resolution should be the headline metric. This one is the working manual: how to calculate IVR containment, how to clean it, what the same metric means for chatbots and AI agents, and how to read a containment figure someone else hands you.

What does IVR containment rate mean?

An IVR “contains” a call when the caller finishes their journey in the menu and never speaks to an agent. The term comes from contact centre operations, where every call that reaches an agent has a labour cost, and every call the menu completes does not.

The definition in our glossary entry on containment rate applies across channels: the automated layer kept the interaction. The important word is “kept”. Containment describes the route a call took. It says nothing about whether the caller got what they rang for.

What is the IVR containment rate formula?

There are three versions worth knowing, from least to most honest.

Raw containment is what most IVR platforms report by default, because it is the easiest thing to count. A call either hit a transfer node or it did not.

Task containment requires your IVR to log a completion event: payment taken, balance read out, order status played, appointment confirmed. If your flows do not emit those events, adding them is the single most useful change you can make to IVR reporting.

True containment adds a time check. For each completed task, look for a second call (or chat, or email) from the same caller about the same reason within a fixed window. If there is one, the first call did not really contain anything. It postponed the contact.

Which repeat window should you use: 24, 48 or 72 hours?

Pick based on how long it takes a caller to discover the self-service did not work.

• 24 hours suits instant tasks where failure is obvious immediately, such as a payment that did not go through or a password reset that did not arrive.
• 72 hours suits tasks where the result shows up later, such as an order status that turned out to be wrong, or an address change that did not apply to the next delivery.
• Seven days or more suits billing and account changes, which our containment versus resolution post covers in more detail.

Whatever you pick, write it down and keep it fixed. A containment figure measured with a 24 hour window is not comparable to one measured with a 72 hour window.

A worked example (illustrative)

The numbers below are invented to show how the versions diverge. They are not a benchmark.

A contact centre receives 10,000 calls into its IVR in a month.

• 3,500 are transferred to an agent.
• 6,500 end without a transfer.

Of those 6,500:

• 900 hang up in the first menu, before choosing anything.
• 1,100 hang up partway through a flow without completing a task.
• 4,500 complete a self-service task.

Of the 4,500 completed tasks, 600 callers ring back about the same issue within 72 hours.

The same IVR, the same month, three numbers spanning 26 percentage points. The 65 percent figure goes on the vendor slide. The 39 percent figure is what the menu actually achieved.

Should first-menu hang-ups be in the denominator?

This is the most common argument in IVR reporting. Some teams remove very short calls, for example under ten seconds, on the basis that they are misdials. That is reasonable for genuine misdials. It becomes a problem when the cutoff is long enough to remove callers who heard the greeting, realised the menu would not help, and left. If you exclude short calls, report how many you excluded, so the reader can judge.

How to measure true IVR containment, step by step

1. Define the entry point. Count calls that reached the first IVR prompt, not calls that hit the carrier.
2. Tag every exit. Each call should end in one of: transferred, completed task, abandoned before task, abandoned during task, callback requested, or voicemail. If your platform cannot tell these apart, fix the logging first.
3. Decide how callbacks and voicemail count. A scheduled callback is a deferred agent contact. It should never count as contained, even though the caller technically left the IVR without a transfer.
4. Match repeat contacts. Join calls by caller ID or authenticated account number, and by call reason or menu path. Include other channels if you can, because a caller who gave up on the phone often emails.
5. Apply your window. Count a completed task as truly contained only if no matching repeat contact appears within the window.
6. Report the funnel, not one number. Raw, task and true containment side by side, plus the count of exclusions.

Why IVR containment misleads

Beyond hang-ups and repeat calls, a few design choices push containment up while making the experience worse.

Hiding the zero-out

If pressing zero or saying “agent” does nothing, more callers stay in the menu. Containment rises. So do complaints. Measuring how often callers attempt to reach a person, and fail, is a good counter-metric.

Menu loops

Flows that send unrecognised input back to the top of the menu keep callers in the IVR longer and raise abandonment, which raw containment then counts as success.

Forced deflection to another channel

Sending callers an SMS link to a web page ends the call without a transfer. Whether that helped depends on what happened next, which the IVR cannot see. Treat these as their own exit type rather than as contained.

Counting callbacks as contained

A “we will call you back” option removes a call from the queue and from the transfer count. It does not remove the work. Callbacks belong with agent-handled contacts.

IVR vs chatbot vs AI agent containment

Chatbots and AI agents inherited the metric directly. The mechanics differ, but the corrections are the same.

The AI agent row is the one to watch. An AI agent can answer a far wider range of questions than an IVR menu, which makes contained conversations more likely to be genuine. It also produces a failure the IVR does not: a fluent, plausible, wrong answer. The customer accepts it, the session ends, and containment records a win. That is why our post on measuring AI support quality recommends sampling transcripts against cited sources rather than trusting session-level metrics.

How containment differs from resolution

Containment answers “did the automation keep this interaction?” Resolution answers “was the customer’s problem solved?” Every genuine automated resolution is contained. Many contained calls resolved nothing.

True containment, as defined above, gets close to a resolution measure, because it requires a completed task and no repeat contact. Our resolution rate glossary entry and the longer argument in deflection vs resolution explain why the outcome measure is the better headline for leadership, for vendor evaluation, and for pricing.

How to read a containment rate someone else quotes

When a vendor, analyst or colleague gives you a containment figure, ask five questions before comparing it with yours:

• Are hang-ups and abandoned sessions in the numerator?
• Are short calls or sessions excluded from the denominator, and at what cutoff?
• Do callbacks, voicemail and SMS deflection count as contained?
• Is there a repeat contact check, and what is the window?
• Is the call or conversation mix similar to yours?

If the answer to the first question is yes and the fourth is no, you are looking at raw containment, and the true figure is lower by an unknown amount.

What to report instead of raw containment

A short monthly view that holds up in a leadership meeting:

1. Resolution rate across automated channels, verified by no repeat contact.
2. True containment for the IVR specifically, with the window stated.
3. Raw containment as context, so the gap is visible.
4. Zero-out attempts and failed transfer attempts.
5. Repeat contact rate by menu path or intent, which tells you exactly which flows to fix.

For the wider set of numbers worth tracking, see our support KPI dashboard.

Where Fidiora fits

Fidiora works on chat rather than voice, and it is built around the outcome rather than the route. It answers customer questions from your help docs, website and past tickets with cited sources, hands anything it cannot resolve to your team, and bills only for genuine resolutions: $0.69 each, dropping to $0.45 above 1,000 a month. Handoffs are always free and abandoned chats are never billed, so the number on the invoice is close to true containment by design rather than raw containment by default. There are no seat fees, you can set a monthly spend cap, and it works alongside your existing helpdesk. See pricing for how a resolution is counted.

The short version

Take your IVR’s reported containment. Remove calls that ended without a completed task. Remove completed tasks where the caller came back about the same thing within 72 hours. The number left is what your menu achieved, and it is the only containment figure worth comparing month to month.