Support metrics glossary
Deflection rate. The most gamed number in support.
Deflection rate is the share of customer issues that get resolved without ever reaching a human agent. It drives your hiring plan, your tooling budget, and your cost per ticket. It is also the easiest support metric to fake, and most dashboards fake it.
Last updated 14 August 2026
What deflection rate means
Deflection rate is a queue metric. Take every issue customers raised in a period, across chat, email, phone, and your help center. Count the ones fully resolved without a human agent, whether by an article, a bot, or an AI agent. Divide the second number by the first, multiply by 100, and you have the metric.
Worked example. Your team sees 1,000 customer issues in June. 420 are fully resolved by self-service and never reach a person. Your deflection rate for June is 42%. The other 580 reached a human, and the metric says nothing about how those went. That is fine. Deflection measures one thing: demand that never hit your queue.
| Total issues in the period | Resolved without an agent | Reached a human | Deflection rate |
|---|---|---|---|
| 1,000 | 300 | 700 | 30% |
| 1,000 | 420 | 580 | 42% |
| 5,000 | 2,500 | 2,500 | 50% |
The most gamed metric in support
Here is the flat version. The gaming is not subtle. Vendors count every bot interaction as a deflection. A customer opens the chat widget, gets no useful answer, and gives up. Counted. A customer asks the bot a question, receives a link to a policy page, and emails you an hour later. Counted, because the chat never became a ticket. On that definition a bot that answers nothing still posts a beautiful number, because none of its failures create tickets inside the chat window.
Most teams inherit this definition without noticing. The fix is one rule. An issue is deflected only if it was resolved. If the customer comes back within 72 hours on any channel about the same issue, it was not deflected. It was delayed. Report the delayed number next to the deflected one for a month and watch your deflection rate drop. That drop is the truth arriving.
Deflection rate vs containment rate
These two get used interchangeably and they are not the same thing. Containment rate counts the conversations that started with your bot and never escalated, divided by all bot conversations. Deflection rate counts every customer issue, bot or not, that never needed a human. Containment is a bot metric. Deflection is a queue metric.
A bot can post a 90% containment rate while ticket volume never falls, if the bot spends its time on questions customers would never have asked a person. Containment tells you how the bot is doing. Deflection tells you how the queue is doing. The containment rate page goes deeper on the bot side. Read them together and you can tell when a dashboard is lying to you.
How to measure it honestly
Step one is the population. Deflection is only meaningful against total demand, not just the conversations your bot saw. You need the full count of issues customers tried to resolve, by every door they tried.
Step two is the numerator. Count a conversation as deflected only when the customer got the answer. Use the resolution signal your tooling gives you, or the simpler proxy: the same customer did not come back about the same issue within 72 hours. Everything else is an interaction, not a deflection.
Step three is the cut. A blended number hides everything. You might deflect 80% of order-status questions and 5% of refund disputes. Those need different fixes and different staffing, so track the metric weekly and by category, never as one company average.
What actually moves the number
The boring fixes do most of the work. Proactive notifications prevent the question entirely, which beats every kind of deflection because the issue never exists. A help center written in the customer's words instead of your internal vocabulary. Then automation that can genuinely resolve: an AI agent connected to your order system answers where is my order with the live carrier scan, and that conversation is a real deflection, not a counted abandonment. That is why WISMO is the first category most stores automate.
Deflection rate questions, answered first
What is a good deflection rate?
There is no honest universal benchmark, because the number swings with your issue mix, your channels, and your definition of resolved. Anyone quoting you a single industry figure is selling something. Compare against your own last quarter, by category, and treat any jump that arrives with a new vendor dashboard as a measurement change until proven otherwise.
Is deflection rate the same as containment rate?
No. Containment rate measures only the conversations your bot handled and asks how many never escalated to a person. Deflection measures all customer issues from every channel and asks how many were resolved without an agent at all. Containment is a bot metric. Deflection is a queue metric, and confusing them is how a 90% containment figure sits next to a queue that never shrinks.
How do I calculate deflection rate?
Divide issues resolved without a human agent by total issues in the same period, then multiply by 100. If 420 of 1,000 June issues were fully resolved by self-service, the answer is 42%. The hard part is the numerator: count only issues actually resolved, not every conversation your bot touched.
Why did my number jump the week we launched a chatbot?
Because the vendor counts interactions, not resolutions. A customer who opened the bot, got nothing useful, and gave up is counted as deflected on most dashboards. Add a comeback rule: if the same customer contacts you again within 72 hours about the same issue, subtract it. Recalculate and you have the real number.
Can deflection be too high?
Yes, when it is bought by blocking exits. A help center with no contact option and a bot with no human handoff both produce a beautiful number and furious customers. Deflection means the issue was resolved without an agent. If customers simply could not reach one, you did not deflect demand, you refused it, and it comes back angrier.