Deflection rate, with the formula and benchmarks

To work it out, divide the questions self-service answered by all the questions customers asked, including the ones that reached your team, then multiply by 100. If your help centre and chatbot answered 600 questions last month and 400 still landed in your inbox, your deflection rate was 60%.

It goes by other names. A helpdesk files each request as a ticket, so it’s often called ticket deflection rate. Salesforce calls it case deflection, and call centres call it call deflection.

For a founder who answers support alone, every deflected question is time back for building. Deacon is an AI customer support agent that answers your customers from your own docs in seconds, and it has a free plan.

How to calculate deflection rate

The bottom line is every question that would otherwise have reached you, which is the ones self-service answered plus the ones your team still got. Salesforce’s case deflection formula is the same sum, with successful deflections over successful deflections plus cases created.

The top line is the hard part. A question only counts if the customer would have written to you without the answer. A question typed into a chatbot passes that test more easily than a help centre visit, because the customer asked for help. What’s left to check is whether they wrote to you anyway.

Say your chatbot answered 700 questions in May, and 400 customers emailed you. Counting every answer, that’s 700 ÷ (700 + 400), or 64%. Then you match the emails to the chats, and 100 of the people the bot answered wrote in about the same thing within a week. They weren’t deflected, and they’re already among your 400 emails. That leaves 600 deflected, and a rate of 600 ÷ (600 + 400), or 60%.

Zendesk turns it the other way up, as a ratio. Its self-service score divides help centre sessions by the people who opened a ticket, and reads 4:1 as four customers trying to help themselves for every one who asked for help. It needs no matching, but it counts attempts, not answers.

What is a good deflection rate?

There’s no standard figure, because each study counts something different. The closest match for a small company running an AI agent is Freshworks’ 48%, for its customers with 1 to 250 employees in 2024. These are the published figures we found with a named sample, read on 22 September 2026.

The figures disagree because they measure different things. Freshworks counts its own AI agent on its own customers’ sites, and its report doesn’t say what makes a query deflected. Zendesk’s ratios count self-service visits and views, whether or not they helped. Gartner asked customers, and even for issues they called very simple, only 36% were fully resolved in self-service.

Vendor pages often quote a band instead, such as 40 to 60% for SaaS in Decagon’s glossary, without naming a study behind it. The Consortium for Service Innovation says “the most useful strategy is to trend against yourself”. Compare this month’s rate with the last few, and look for it rising as you fill the gaps in your docs.

Deflection rate vs containment rate and resolution rate

These rates get used as if they were one, but each counts a different success.

A customer who gives up on a chatbot and closes the tab was contained, and some tools count them as deflected too, but nothing was resolved. HubSpot’s customer agent report counts every conversation its bot handled without a person as a deflection, and warns that “Deflections do not always indicate resolution.” It’s one reason Decagon says most of its customers pay per conversation rather than per resolution, as Decagon pricing explains.

Why deflection rates often read too high

A deflection rate can look better than the support behind it, in four ways.

Counting visits counts people who were never going to write in. DB Kay’s paper on estimating deflection found that for every contact self-service saves, it also helps ten or more people who wouldn’t have got in touch. Its fix is to ask about 200 recent visitors, by phone, whether they found what they came for and whether they’d have contacted you otherwise, then multiply the two shares.

Giving up counts as success. A customer who abandons a contact form in frustration never reaches you, and Salesforce’s case deflection page calls those “unhappy visits”.

Coming back later isn’t caught unless you look for it. Gladly’s deflection glossary describes a stricter count that keeps a deflection only when the customer didn’t return about the same issue within a fixed window, such as 24 hours or seven days.

Hiding your contact details raises the rate and loses customers. Our guide to reducing support tickets covers the three kinds of message that should always reach a person.

What a deflected question is worth

A 2019 Gartner poll of customer service leaders put a contact by phone, live chat or email at $8.01 on average, against about $0.10 for self-service. At those prices, the 600 questions deflected in the example above saved about $4,750 in May. Our entry on cost per contact shows how to work out your own figure. For a founder answering support alone, the cost that matters is time, since every question that reaches you is time away from the product.

Raising your deflection rate with Deacon

In Gartner’s 2023 survey, the most common reason self-service failed, in 43% of cases, was that customers couldn’t find content relevant to their issue. Deacon answers them where they ask, in seconds, from your own docs and in whatever language they write in. When your docs don’t cover a question, it says it doesn’t know instead of guessing and asks for the visitor’s email, so you can reply yourself.

It also gets better the more you answer. Every question your docs didn’t cover has an Answer this button, and every answer you write becomes part of what Deacon knows, so it can answer the next person who asks. That raises deflection by answering more, not by hiding your email address.

And it tells you what customers need. Topics group similar questions, count how many people asked and label most groups as a how-to, a bug report, a feature ask and so on. You see which missing answers keep people stuck, and what customers want you to build next.

Deacon can’t see your inbox, so it doesn’t claim a deflection rate for you. The Answered figure on Analytics › Widget is the share of questions Deacon answered out of those it tried, leaving out any that failed with an error or arrived after your plan’s answers ran out. Conversations can be filtered by Answered, Unanswered, Wants a person, Failed, Over limit and Dismissed, so you can read the ones that didn’t go well. For a deflection rate, take the questions Deacon answered, leave out any tagged Wants a person and any whose visitor emailed you anyway, and put the rest into the formula above with the emails you received. Chatbot analytics covers the rest of the dashboard, and you can try it all on the free plan, which needs no card.