Time the Reply, Not the Draft

Say a service rep used to spend ten minutes drafting a reply. With AI, it takes three. The customer still waits until tomorrow for an account detail and a supervisor’s sign-off.

I’d take twenty actual requests and record when each arrived, when a complete reply went out, and whether the customer had to write back. Keep the awkward cases in the sample. Then ask where those seven minutes went. Did the queue shrink? Did staff answer another request? Or was the day simply less punishing?

That last outcome matters. I wouldn’t put it on a spreadsheet as earnings, though. If faster service is the goal, time the request until the customer gets an answer, including the wait after the draft is done.