Most businesses measure support with one number - response time - and it's the least interesting one. Once an AI agent answers in under two seconds, response time stops discriminating between good and great. Here are the five metrics that still do, and what each one tells you to fix.
1. Deflection rate: what never needed you
Deflection is the share of conversations resolved without a human. Healthy grounded-AI setups sit around 80–90%. If yours is lower, it's almost never the AI - it's coverage: the answers customers want aren't in your documents yet. Chase deflection by writing, not by tuning.
2. Knowledge coverage: the gap list
Every question the bot couldn't answer is a logged gap, and the gap list is your editorial calendar. A shrinking gap list means your knowledge base is converging on reality. A gap that repeats daily is a document you owe your customers by Friday.
3. Escalation quality: the right conversations, not fewer
Don't celebrate a falling escalation count by itself - celebrate what's in it. The escalations reaching you should be genuinely human-worthy: disputes, bulk orders, edge cases. If "what's your return window?" is still reaching a human, that's a coverage bug wearing an escalation costume.
The goal isn't zero escalations. It's zero boring escalations.
4 & 5. After-hours share and conversation-to-order rate
After-hours share tells you how much of your demand lives outside business hours - for most Indian SMBs it's more than half, which is the revenue you were silently forfeiting before automation. And conversation-to-order rate closes the loop: customers who get an instant, accurate answer buy more often than those who wait. Track it, and support stops being a cost line and starts reading like a sales channel.
Review the five together monthly. Any one of them alone can be gamed; together they describe whether your support is actually compounding.