What the metric counts
Deflection rate is one of the oldest numbers in contact centre reporting, and it has become one of the most contested since voice AI and chat automation started running the self-service layer. The formula is simple: self-service resolutions divided by total customer inquiries, expressed as a percentage. If 1,000 customers try to reach support and 400 never reach a human agent, the reported deflection rate is 40%.
What the formula does not do is check whether those 400 customers got an answer. It only checks whether they reached a person. A customer who reads an irrelevant help article and gives up looks identical in the data to a customer who found the answer and moved on. Both count as deflected.
Why the industry built it this way
Deflection rate exists because it is cheap to measure and easy to report upward. It requires no follow-up window, no customer satisfaction survey, no cross-channel tracking. A support platform can produce the number the same day a contact happens, which makes it attractive on a dashboard even when it says little about outcome.
Teneo, a conversational AI vendor, has argued the metric measures the wrong side of the transaction: did the company keep this call off the agent queue, rather than did the customer's problem get solved on this contact. Its point stands regardless of who is making it: deflection is a routing outcome, not a service outcome.
Deflection, containment and resolution are three different numbers
The industry has split the single idea of "kept off an agent" into three separate metrics, and buyers evaluating a vendor's automation claim need to know which one is being quoted.
Deflection rate counts any inquiry that avoided a live agent, regardless of what happened next.
Containment rate is narrower. It counts inquiries that entered an IVR or chatbot flow and were fully resolved inside that flow without escalating to a human, calculated as contained calls divided by total automated-channel calls. A vendor reporting 60% containment means 60 of every 100 automated interactions closed without a transfer, not that 60% of all inquiries were solved.
Net or "true" deflection goes a step further and subtracts customers who came back. One published formula defines it as self-service resolutions minus re-contacts within 48 hours, divided by total help-seeking attempts. This is the version that actually approximates "solved," because it accounts for the customer who clicked away and called back the next day with the same problem.
Re-contact rate is the number that catches the gap
The diagnostic that exposes an inflated deflection number is re-contact rate: the share of customers who reach out again about the same issue within a defined window, typically 48 to 72 hours. A high re-contact rate after an automated interaction means the system closed the conversation without closing the problem. If deflection is climbing and re-contact is climbing with it, the automation is not reducing work. It is postponing the same contact and making it harder to trace back to its origin.
This has a measurable effect on staffing math, and Gartner's most recent customer service leader survey shows why it matters beyond the metric itself. In a survey of 321 customer service and support leaders conducted in October 2025, only 20% reported reducing agent headcount because of AI, while 55% reported stable staffing despite handling higher contact volume. Volume is being absorbed somewhere. If a chunk of that "deflected" volume is quietly re-entering the queue through a different channel or a different day, stable headcount against rising volume is exactly the pattern that shows up.
What to check before trusting a deflection number
Ask which of the three metrics is being quoted, since vendors and internal teams sometimes use "deflection" loosely to mean any of the three. Ask for the re-contact window and rate alongside the headline number; a deflection or containment figure presented without a paired re-contact rate is incomplete by design, not by oversight. Ask whether the re-contact window covers cross-channel contact, since a customer who calls back after a failed chat interaction should count against the chat metric even though the second contact arrived through a different channel.
None of this makes deflection rate useless. It is a legitimate volume metric and a fair proxy for automation adoption. The failure mode is treating it as a resolution metric when it was never built to measure resolution at all.
Sources: Balto · TechSee · eesel AI · Teneo (vendor commentary) · TELUS Digital · Gartner press release, Dec 2 2025