Deflection rate, as a ratio and as a percentage

Self-service customer service covers help centres, FAQ pages and chatbots. In a helpdesk, the software a support team answers email from, each request is a ticket, so the measure is often called ticket deflection rate. Salesforce calls it case deflection, and in call centres it is call deflection.

Working out deflection as a percentage or a ratio

The percentage divides the questions self-service answered by all the questions that would otherwise have reached your team. Salesforce counts those as the successful deflections plus the tickets that were opened.

The same idea also works as a ratio. Zendesk’s self-service score divides help centre visits by the number of people who opened a ticket. A score of 4:1 means four visits for every person who wrote in.

A project management app’s help centre had 2,400 visits in May, and 300 customers opened a ticket. That gives a self-service score of 8:1. The percentage needs a number the ratio leaves out, the visits that answered a question the customer would otherwise have sent to the team.

Estimating deflections with a short visitor survey

Most help centre readers never meant to write in. That is the finding of DB Kay’s paper on estimating deflection, from a consultancy that helps support teams with self-service. A deflection, in its method, is a visit where failing to find the answer would have led to a contact.

The method asks recent visitors two things. Did they find what they came for, and if they had not, would they have contacted you? It multiplies the two shares together and applies the result to every visit. DB Kay says to phone about 200 visitors, because the few who reply to a written survey are unlikely to be typical.

Say the app asks 240 recent visitors, and half say they found their answer. Of those 120, ten say they would have opened a ticket without it, one in twelve. Half of 2,400 visits is 1,200 that found an answer, and one in twelve of those is 100 tickets deflected in May. Salesforce’s percentage is then 100 ÷ (100 + 300), or 25%.

The ratio said 8:1. The percentage says the help centre answered one in four of the questions that would have reached the team. The other 1,100 visits that found an answer helped people who were never going to write in. DB Kay’s research found ten or more of those for every deflection.

An article’s feedback button cannot supply the first share. The Consortium for Service Innovation puts participation in that kind of feedback at about 1.5%. DB Kay says those who press it are probably not typical.

What a deflection is worth

A 2019 Gartner poll of customer service leaders put the cost per contact by phone, live chat or email at $8.01 on average. Self-service cost about $0.10. At those prices the app’s 100 deflections in May saved about $790, and counting all 2,400 visits would claim about $19,000.

DB Kay prices each deflection at the cost of a simple ticket, because self-service can only answer questions that already have a written answer.

Visitors who gave up and still counted

A visitor who reads an article, finds nothing and closes the tab still adds a visit to Zendesk’s ratio. So does a visit where the customer gives up halfway through the contact form. Salesforce’s case deflection guide calls those “unhappy visits”. The aim, it says, is to tell apart the customers who found their solution and the ones who gave up.

A customer who gives up on the chatbot and phones ten minutes later can count as deflected too. That happens when nothing links the two. Forrester’s Max Ball calls it “an epic customer-experience fail” in a post on chatbot business cases.

Gladly’s deflection glossary describes a stricter count. It keeps a deflection only if the customer did not return with the same question within a set window, such as seven days.

What HubSpot, Zendesk and Microsoft count as a deflection

Chatbot tools count deflection more loosely than DB Kay. HubSpot’s customer agent report counts every conversation its bot handled without a person, and says “Deflections do not always indicate resolution.” For chatbots, Zendesk’s blog on ticket deflection divides the people who used the bot by the people it sent to an agent. Microsoft’s Copilot Studio guidance gives no formula at all, because each organisation defines deflection differently.

Deflection figures from Zendesk, Freshworks and Gartner

These are the published figures we found with a named sample, read on 19 September 2026.

In Gartner’s survey, 73% of customers had used self-service at some point while getting help. Even the issues they called very simple were fully resolved there only 36% of the time.

Vendor pages such as Alhena’s guide to deflection rate call 40 to 60% good, and cite no study with a named sample. The Consortium for Service Innovation sees “little value in attempting to benchmark against other organizations”. Compare your own months instead. Zendesk’s self-service guide judges success by the ratio growing.

Deflection rate compared with three other rates

Calabrio’s page on chatbot containment says containment is often called deflection, but the four rates below count different things.

Raising the rate without hiding your team

Our guide to reducing support tickets covers what to write down first, and the three kinds of message that should always reach a person. Our own chatbot, Deacon, saves every conversation and marks the ones it could not answer, so you can see which visitors left without help.