Customer service KPIs that matter when AI handles support
The metrics that show whether AI support is working: resolution rate, reopen rate, handoff quality, CSAT, first response time and cost per resolution — with how to calculate each.
Why old support metrics mislead with AI
Tickets per agent and handle time were built for teams where people answered everything. Once AI resolves part of the volume, those numbers can rise or fall for reasons that have nothing to do with quality. Measure outcomes instead.
The six KPIs to track
Track these weekly, split by AI and human conversations:
- AI resolution rate = conversations closed by AI without a human ÷ all conversations
- Reopen rate = resolved conversations the customer reopened within 7 days ÷ resolved conversations
- Handoff quality = share of handoffs where the agent didn’t need to re-ask for information
- CSAT = satisfied responses ÷ all survey responses, for AI and human separately
- First response time = median time to first meaningful reply
- Cost per resolution = total support cost ÷ resolved conversations
How to read them together
A rising AI resolution rate is only good news if reopen rate and CSAT hold steady. If resolution climbs while reopens climb too, the AI is closing conversations it hasn’t really solved — tighten confidence thresholds or improve the source content.
Review a sample of 20 AI conversations each week. When an answer fails, fix the knowledge article it came from; that improvement carries to every channel.
Frequently asked questions
What is a good AI resolution rate?
It depends on your mix of questions. Teams with many repetitive questions see higher rates than teams with complex, account-specific work. Track the trend and the reopen rate rather than chasing a fixed number.
How do I calculate cost per resolution?
Divide total monthly support cost — seats, AI usage, tools and staff time — by the number of conversations resolved that month.