A bot quoted the wrong return window to forty customers and nobody noticed, because confident answers look identical to correct ones. Citations are how you tell them apart.
AI support isn't about replacing humans. It's about filtering what humans shouldn't have to touch — and handing off cleanly when it can't. Four patterns, where each one earns its keep, and where each one breaks.
"2-minute average first response" sounds great but correlates weakly with satisfaction. Swap in these three metrics and you will see a very different picture.
A 4.6/5 CSAT feels great. It's also telling you almost nothing. Three structural problems — survivor bias, rating inflation, agents begging for 5-stars — have broken the metric. Here's what to measure instead.