Containment rate
In one sentence
Containment rate is the share of conversations an automated system finishes on its own without handing the person to a human, calculated as conversations resolved without human involvement divided by total conversations handled.
Not to be confused with Call deflection.
Definition
Containment rate is the share of conversations an automated system finishes on its own, without passing the person to a human. It is the headline efficiency metric for voice automation.
If a hundred people get in touch and seventy are finished with by the machine, containment is 70 percent. What that number does not tell you is whether those seventy got what they came for.
The formula is not the hard part. The numerator is conversations resolved without human involvement. The denominator is total conversations handled. Every argument about containment is really an argument about what counts as resolved, which is why the definition has to be pinned down before it goes anywhere near a contract.
Three ways containment gets inflated
- Counting abandonment as containment. The caller gave up and hung up. No human was involved, so a naive metric writes it down as a success.
- Counting deflection as resolution. The agent sent a link and closed the conversation. Nothing was resolved, and the customer may well be back.
- Excluding conversations that were routed to a human before the agent ever engaged, which quietly shrinks the denominator.
The metrics containment has to be read alongside
- Resolution rate. Did the problem the person had actually get solved.
- Repeat contact rate. Did they come back within a few days. This is the honest correction to an inflated containment number.
- CSAT on contained conversations specifically, not overall CSAT. High containment paired with low satisfaction among contained conversations means you are trapping people rather than helping them.
- Escalation rate, which is the inverse, and more importantly escalation quality: did the human arrive with context or start cold.
Containment against deflection
- Deflection moves a contact out of an expensive channel.
- Containment resolves it in the channel it arrived in.
- Deflection is the more dangerous one to optimize for, because you can achieve it simply by making a channel harder to reach.
Common misconception
Almost everyone assumes higher containment is always better. Containment has an optimum, not a maximum. Past a certain point the extra containment is bought by frustrating people who needed a human, and a support cost quietly becomes a churn cost. Where that optimum sits depends on how much of the incoming volume is genuinely automatable, and that share swings wildly between one industry and the next.
Why it matters commercially
Containment is the number that lands in the business case and on the board slide. It translates straight into money saved, which is why a procurement team puts it in the service level agreement rather than leaving it in the pitch deck. Anyone selling voice automation should be able to state the definition precisely and should push to have it agreed before anyone signs. A vague clause reads as generous early and costs both sides later.
In voice specifically
Containment transfers imperfectly to voice on a website. Most website deployments have no queue behind them and nobody to escalate to, so the closest honest analogue is question resolution rate: did the visitor get their answer and stop hunting around the site. Lead capture takes the place of escalation, which makes a captured lead the path working rather than containment failing.
Where AsqVox fits
The Orb dashboard surfaces transcripts and sentiment for every conversation. That is the raw material for checking whether a contained conversation was genuinely resolved or merely ended.
Visual
Containment, honestly measured
Containment counts conversations that ended. Resolution counts problems that ended. Only one of those is worth paying for.
Statistics
Every figure carries its source and year. Vendor numbers are labelled as vendor numbers, and where no reliable figure exists this page says so rather than borrowing one.
Mature deployments report containment of roughly 70 to 80 percent.
70 to 80%industry rangeIndustry-reported range, 2025 - Not an audited figure. Containment benchmarks vary substantially by vertical and by how each operator defines resolution, so two published numbers are rarely measuring the same thing.
Average deployments report containment of roughly 40 to 55 percent.
40 to 55%industry rangeIndustry-reported range, 2025 - Same caveat. Ask any vendor quoting a number in this band whether abandoned conversations are excluded before comparing it with anything.
Rule-based systems report containment below 35 percent.
under 35%industry rangeIndustry-reported range, 2025 - Same caveat. The gap between this band and the mature band is the size of the prize being sold, which is exactly why the definition clause matters.
A call handled by a person costs roughly USD 7 to USD 12.
USD 7 to 12industry rangeWidely cited industry range, 2025 - The spread is driven by geography and call complexity. A business case should run on local labor cost rather than the published range.
A call handled by a voice agent costs roughly USD 0.40.
USD 0.40industry rangeWidely cited industry range, 2025 - Same caveat. Treat the ratio between the two figures as directional rather than treating either absolute number as precise.
Conversational AI in contact centers was forecast to cut agent labor costs by USD 80 billion by 2026, with one in ten agent interactions automated by 2026.
USD 80bnanalyst forecastGartner press release, attributed to VP analyst Daniel O'Connell, 2022 - Dated 31 August 2022, and the date belongs in any citation. The one-in-ten figure is the half worth quoting: a ceiling that low on near-term automation sits well under the containment rates vendors advertise, and that mismatch is worth putting in front of a reader rather than skipping past.
There is no published benchmark for containment or resolution rate for website voice agents specifically. The metric was built for contact centers, and the honest move is to state the transfer problem rather than borrow a contact-center number and imply it applies.
-no reliable figureIf a website voice figure is quoted at you, ask what it was measured against. Anything in the 70 to 80 percent band almost certainly came from a phone deployment.
Examples
In practice
A utility reports 74 percent containment in month one and celebrates. Repeat contact analysis then shows that 21 percent of contained conversations produced a follow-up contact within 72 hours, which puts adjusted resolution at roughly 58 percent. The team redefines the metric to exclude any conversation followed by a repeat contact, publishes the lower honest number, and turns the gap into the roadmap: the top three repeat-contact reasons become the next three automations.
The everyday version
Say a hundred people asked your voice agent something this week. Seventy got their answer and carried on with their day. Thirty needed you. The question worth asking is not how to push seventy up to ninety. It is whether those thirty reached you without a fight, and whether the seventy really got what they came for or just gave up politely.
Usage
Who says it
- Contact center and CX leaders treat containment as a primary operational KPI and will already have a house definition, which you should ask for before quoting a number back at them.
- Finance uses it to model cost savings and will not interrogate the definition unless someone prompts them to.
- Vendors quote it in marketing, frequently without stating whether abandoned conversations were excluded.
Where it turns up
- Next to first contact resolution, average handle time, escalation rate, CSAT, service level and cost per resolution.
- Often written into a service level agreement with money attached to missing it, which is why the definition clause deserves more scrutiny than the target number.
Common misuse
- Reporting containment without stripping out abandoned conversations. It is the most common inflation of the lot, and it is almost always an accident rather than a trick.
- Using containment and deflection interchangeably. They measure different things and reward different behavior.
- Setting a containment target with no paired satisfaction or repeat-contact guardrail, which incentivizes trapping customers.
Questions people ask
How is containment rate calculated?
Divide conversations resolved without any human involvement by total conversations handled. The formula is the easy part; the argument is always about what counts as resolved. A definition that does not exclude abandoned conversations will overstate containment, and usually nobody meant it to.
What is a good containment rate?
Mature deployments are reported at roughly 70 to 80 percent, average deployments at 40 to 55 percent, and rule-based systems below 35 percent. Treat those as industry-reported ranges rather than audited figures, because they move with the vertical and with how resolution is defined. No published benchmark exists for website voice agents specifically.
Is a higher containment rate always better?
No. Containment has an optimum, not a maximum. Past a certain point the extra containment comes from frustrating people who needed a human, which turns a support cost into a churn cost. The right target depends on how much of your incoming volume is genuinely automatable, and that varies enormously by industry.
What is the difference between containment rate and deflection rate?
Containment resolves the contact in the channel it arrived in. Deflection moves it out of an expensive channel into a cheaper one. Deflection is the more dangerous metric to optimize for, because you can hit the target simply by making a channel harder to reach.
Last reviewed 30 July 2026. Written and reviewed by Dhruv Dholakia, founder of AsqVox.