AI Deflection Rate vs. Resolution Rate: What Gartner Actually Reports

Updated September 2026 with Gartner's 2025 and 2026 survey results.

Gartner's most-cited number on self-service is a sobering one: in a survey of 5,728 customers, only 14% of customer service issues were fully resolved in self-service. Even for issues customers called "very simple", just 36% were. Meanwhile 73% of customers use self-service at some point in their journey. Put those together and most people who try the bot or the help center still end up in another channel. A high deflection rate on a dashboard can hide exactly that: automation redistributing support load rather than reducing it.

Gartner deflection and resolution figures at a glance

None of these is a deflection rate. All five are about resolution or adoption.

Does Gartner publish an AI deflection rate for service desks?

Not as a benchmark. You will find posts saying "Gartner found AI deflects 45% of queries but only 14% are resolved, so 31% come back." The 14% is Gartner's. The 45% and 31% are not in the Gartner release, and we repeated that framing ourselves in an earlier version of this article before checking the source. What Gartner has published is closer to the opposite of a deflection score: it measures whether the customer's problem actually got solved without a person, and the answer in 2024 was one time in seven.

The newer Gartner surveys point the same way:

• A January–February 2025 survey of 5,801 customers found 60% of agents don't promote self-service at all, and only 35% of customers who last contacted support by phone were willing to try a GenAI assistant.
• An October 2025 survey of 321 service leaders found 91% are under executive pressure to implement AI, with "self-service success" among the top three priorities for 2026. 58% plan to turn agents into knowledge management specialists.
• Gartner's March 2025 prediction is that agentic AI will autonomously resolve 80% of common customer service issues by 2029. Note the verb: resolve, not deflect.

So if leadership asks for "the Gartner deflection rate", the honest answer is that Gartner tracks resolution, and the gap between 14% today and 80% by 2029 is the work.

Deflection and resolution are not the same number

Picture a chatbot reporting 90% deflection while only 40% of those conversations actually fixed anything. That isn't far-fetched, because deflection counts an abandoned conversation and a confidently wrong answer the same way it counts a genuine fix: the ticket never gets created, so it gets logged as deflected. A system with a lower deflection number but few repeat contacts is worth more than one with a high number and a phone queue that keeps growing. The second one moves the load somewhere less visible, usually a call or a re-opened ticket a few days later.

Why the knowledge base decides the outcome

An AI layer answers from whatever content it is given. If the article on VPN setup is from before the last firewall change, the bot will explain the old steps quickly and politely, and the user will call anyway. That is why Gartner's 2026 leaders are putting people on knowledge management rather than buying a bigger model. On an IT service desk this is very concrete: password resets and access requests are easy to automate, but only if the runbook behind them matches how the systems work this month.

What separates systems that reduce load from ones that just move it

• Track repeat contacts, not just closes. A request "deflected" today that comes back by email or phone within 48 hours wasn't resolved. If your reporting can't join those two events, the deflection number tells you very little.
• Give the knowledge base an owner. Every article the AI can quote needs a person responsible for it and a review date.
• Escalate a real miss right away. A bot that keeps retrying a request it can't handle is protecting its own score at the user's expense.
• Keep humans in the approval path for actions, not just answers. Once an agent can reset accounts or change access, the governance questions change. We cover that in agentic AI helpdesk governance.

The metric worth reporting to leadership

Not "we deflected 60% of tickets this month." Instead: of the requests handled without an agent, what share didn't come back within a week? That number is smaller and less flattering. It is also the only one that tells you whether support load went down or just went quiet for a few days. The UAE's August 2026 guide for government AI agents takes the same line, counting completed outcomes instead of processed transactions (more on what a company desk can take from it).

FanDesk is built around that idea: a searchable knowledge base that an agent, human or AI, can pull the right article from inside a ticket, SLA tracking that doesn't stop the clock when a bot replies, and reports built on resolution rather than ticket closure. If you are comparing tools on these points, our helpdesk selection guide lists what to check.

FAQ