Deflection rate
Deflection rate is the share of support contacts resolved without a human agent. It is a common target and a misleading one: a deflected ticket and a resolved question are not the same event. A user who gives up counts as deflected. The better metric is resolution: whether the person actually got what they came for.
Why is deflection a flawed metric?
Because it measures the absence of a ticket, not the presence of a solution. A confusing bot that makes users abandon their question deflects beautifully and helps no one. Optimising for deflection can make a support experience worse while the dashboard turns green, which is why resolution and repeat-question rate matter more.
What should you measure instead?
Resolution rate (did the user complete the task) and repeat-question rate, which catches the deflections that didn't actually work. An honest system also reports sessions where it found nothing and said so, because a clean hand-off is a better outcome than a false resolution that sends someone in a circle.