Call deflection
In one sentence
Call deflection is the practice of moving a customer contact out of an expensive channel, usually the phone, into a cheaper one such as self-service, and it counts the contact that never arrived rather than the problem that got solved.
Not to be confused with Containment rate.
Definition
Call deflection counts the customers who would have phoned you and did not, because they got what they needed somewhere cheaper.
It records where the contact went. It does not record whether anybody was helped.
Deflection is a cost metric, and it is easier to game than anything else in the category. What it counts is contacts that never reached an expensive channel. That is channel avoidance, and channel avoidance is not the same thing as resolution. Holding those two apart is the entire discipline of reading a deflection number.
How it differs from containment
- Containment counts conversations resolved inside the channel they arrived in, with no human involved.
- Deflection counts contacts stopped before they reached a channel at all.
- A contact can be deflected without being resolved, and often is.
Three ways to produce the number, one of which is worth having
- Genuine self-service. The answer exists, the customer can find it, and it is enough. Nobody needs to call. This is the legitimate mechanism and it is real value.
- Redirection. The customer is handed a link or pointed at a page. They may find the answer and they may not. Either way the deflection is recorded.
- Friction. The number is buried, the queue runs long, the callback never arrives. Contacts fall and deflection climbs. The report reads it as success. The customer experiences it as contempt.
- A deflection target with no satisfaction guardrail and no repeat-contact guardrail actively rewards the third mechanism, because friction is the cheapest of the three to build.
The four numbers that keep it honest
- Repeat contact rate. Did the deflected customer come back through another channel within a few days.
- Task completion inside the deflected channel. Did they actually get the answer.
- Satisfaction measured on deflected contacts specifically, not on the overall base.
- Total contact volume across every channel. Real self-service brings that total down. Friction just relocates it.
Answering on the website is deflection in the legitimate sense, because the visitor gets the answer where they already are and no contact is ever created. The honest way to measure that is whether total inbound contacts fall while satisfaction holds, not how many conversations the voice widget handled.
Common misconception
That a high deflection number is straightforwardly good. Deflection bought with friction turns a support cost into a churn cost and a reputation cost. Both are larger than the cost avoided, and both land on a different line of the accounts, which is exactly why the trade looks attractive inside a quarterly review.
Why it matters commercially
Deflection is the number most likely to end up in a business case and least likely to be defined rigorously in one. Anyone selling voice automation should insist that the definition arrives with a satisfaction guardrail attached, because a deflection figure that later turns out to have been friction damages the vendor relationship as badly as it damages the customer one.
In voice specifically
A spoken answer on the page removes the reason to pick up the phone at the moment the reason exists. That is the mechanism working the way it is supposed to. It also makes the guardrails easier to check than in a contact center, because the conversation that replaced the call is right there to read.
Where AsqVox fits
On-site answering deflects the legitimate way, by answering the question rather than by hiding the route to a human. The transcript and sentiment dashboard behind the Orb is the mechanism for checking which of the three kinds of deflection is actually happening.
Visual
Three routes to the same deflection number
Deflection measures where the contact went. Only one of these paths solved anything.
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.
A call handled by a person costs roughly USD 7 to USD 12.
USD 7 to 12industry rangeWidely cited industry range, 2025 - Not an audited figure. It varies substantially by geography and by how complicated the call is, so a real business case should replace it with local labor cost.
A call handled by a voice agent costs roughly USD 0.40.
USD 0.40industry rangeWidely cited industry range, 2025 - Same caveat as the human figure. Treat the ratio between the two as directional rather than treating either absolute number as precise.
Automation containment benchmarks, useful for comparison: mature deployments run roughly 70 to 80 percent, average deployments 40 to 55 percent, and rule-based systems under 35 percent.
70 to 80% mature, 40 to 55% averageindustry rangeIndustry-reported ranges, 2026 - Industry-reported rather than audited. These are containment figures, not deflection figures, and they are quoted here because containment is the metric deflection is most often confused with.
Conversational AI deployments in contact centers were 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 of it. The one-in-ten figure is the more informative half, and it sits well below the containment rates vendors advertise. Surfacing it reads as more credible than leaving it out.
No published cross-industry benchmark for deflection rate exists. Organizations define the metric so differently that a comparison between them would be meaningless.
-no reliable figureIf a vendor offers you a deflection benchmark, ask what definition it was calculated under. The answer is usually the point at which the comparison collapses.
No established benchmark exists for deflection achieved by website voice agents specifically. The metric was built for contact centers and transfers imperfectly.
-no reliable figureThe nearest honest substitute is your own before-and-after: total inbound contacts across every channel, measured against a satisfaction guardrail, over a period long enough to see repeat contacts return.
Examples
In practice
A telecoms operator reports deflection up twelve points after a self-service push. Cross-channel analysis shows total contacts across phone, chat, email and social are unchanged, and the email queue has grown by roughly the volume the phone queue lost. Contacts moved, they did not disappear. Reporting is rebuilt around total contacts per thousand customers, which is much harder to game and immediately shows which initiatives reduce demand rather than redirect it.
The everyday version
Deflection is a customer who would have phoned you and did not. That is excellent if they did not phone because they found the answer on your website in ten seconds. It is a problem if they did not phone because they could not find your number, gave up, and bought from someone else. Both look identical on the report.
Usage
Who says it
- Contact center and CX leaders, who use it as a cost and efficiency metric.
- Finance, in business cases, usually without interrogating how it was defined.
- Vendors, frequently and enthusiastically, because it is the easiest number in the category to make look good.
Where it turns up
- Usually in the same clause as containment rate, cost per contact, self-service adoption, CSAT and total contact volume.
- Sometimes written into commercial terms with deflection targets attached, which turns the definition clause into a commercially significant one. Read it before signing it.
Common misuse
- Using deflection and containment as if they were the same word. They measure different things and they reward different behavior.
- Reporting deflection without total contact volume across all channels, which hides contacts shifting sideways rather than disappearing.
- Setting a deflection target with no satisfaction or repeat-contact guardrail, which is a direct incentive to build friction.
Questions people ask
What is the difference between call deflection and containment rate?
Containment counts conversations resolved inside the channel they arrived in, without a human. Deflection counts contacts that never reached the channel at all. A contact can be deflected without being resolved, and frequently is. Treating the two as interchangeable is the most common misuse of both.
Is a high call deflection rate good?
Not on its own. The same number can be produced by genuine self-service, by redirection that may or may not work, or by friction such as a buried phone number and a long queue. Friction converts a support cost into a churn cost and a reputation cost, which are larger and appear on a different line of the accounts.
How do you measure call deflection honestly?
Report it with four guardrails: repeat contact rate over the following few days, task completion inside the deflected channel, satisfaction measured on deflected contacts specifically, and total contact volume across every channel. Genuine self-service brings the total down. Friction only moves it somewhere else.
Is there a benchmark deflection rate?
No. There is no published cross-industry benchmark, partly because organizations define the metric so differently that comparing them would be meaningless. There is also no established benchmark for deflection by website voice agents specifically, because the metric was built for contact centers and transfers imperfectly.
Last reviewed 30 July 2026. Written and reviewed by Dhruv Dholakia, founder of AsqVox.