Deflection rate (ticket deflection)

The simplest formula for an AI assistant is: conversations resolved without a human ÷ total conversations, over a period. For a help centre, teams often estimate it as the number of people who viewed help content and did not open a ticket afterwards, divided by all people who viewed help content.

Both versions rely on a definition of “resolved”. The strict one requires a positive signal — a thumbs-up, an explicit confirmation, no repeat contact on the same topic within a few days. The loose one counts any conversation that did not end in a ticket, which overstates success.

“Deflecting” a customer sounds like pushing them away, and a deflection target can encourage exactly that: hiding the contact link, making handoff hard, closing conversations early. The number goes up while satisfaction goes down, and customers come back through another channel.

That is why many teams now talk about resolution rate or containment rate instead, and pair it with customer satisfaction and the rate of repeat contacts. A request is only deflected well if the customer's problem was actually solved.

Read the conversations that were not resolved and group them by reason: missing information in the knowledge base, a tool the assistant lacks (order lookup, returns), a policy question that needs a person, a misunderstanding. Fix the most frequent reason first. Adding the missing answer or the missing tool raises resolution honestly; making the human route harder only moves the problem.

AI SmartTalk does not use a single “deflection” score. The dashboard breaks questions down by outcome — answered by the AI, escalated or handed over to a person — and shows who resolved them, the AI, a human or no one. It also shows the satisfaction rate from customers' thumbs-up and thumbs-down on answers, the thumbs-down still to review, and the questions the assistant missed, which are the list of gaps to fill.