Deflection Rate

Deflection rate is the share of inbound support contacts resolved without a human agent — by a help centre article, a chatbot, or an AI assistant — and it is the headline number in almost every AI support tool's pitch. It is also one of the easiest metrics in SaaS to inflate, so a buyer comparing vendors is usually comparing four different definitions wearing the same name. The variations that matter. Some vendors count a session as deflected whenever it ends without a ticket being created, which counts every user who gave up and left as a success. Some count any session where the assistant produced an answer, regardless of whether it was right. Stricter definitions require an explicit resolution signal — the user marked it solved, or no follow-up contact arrived on the same issue within a window of a few days. The denominator moves too: deflection measured against all sessions including trivial ones looks far better than deflection against contacts that would otherwise have become tickets, and only the second is the number that shows up in your staffing plan. The way to make a vendor demo comparable is to ask for the exact numerator and denominator, the follow-up window, and how abandonment is classified — then ask to see the metric alongside customer satisfaction for deflected sessions and the reopen rate. Deflection that rises while satisfaction falls is not deflection; it is a queue of frustrated users taking a slower path to the same human. Paired with cost per resolution, deflection becomes a business case rather than a demo statistic, because the two together show whether the cheaper path is also an acceptable one.