Measuring AI in customer operations: the numbers I put on one page
The one-page scorecard I use for AI in customer operations: time back per agent, review rate, escalation time, CSAT, and how to read them together.
"The AI works" is not a result. A manager wants to know what changed for the team and for customers. I keep the answer to one page.
The numbers from my current role
2 h → 20 minmanual handling per agent per day
100–200messages routed daily
−25%escalation resolution time
89%CSAT across 60 tickets a day
95%VIP retention
+21%campaign engagement
How the scorecard is laid out
01Time backminutes of manual work per agent
02Safetyshare of AI outputs held for review
03Speedescalation resolution time
04QualityCSAT and the daily QA sample
05Revenue sideVIP retention and campaign engagement
Reading them together
One number alone
- Handling time falls
- Fewer items held for review
- Faster escalations
- More campaigns sent
The pair that keeps it honest
- CSAT holds or rises
- QA sample still clean
- Reopen rate does not climb
- Opt-outs stay flat
Where the data comes from
HubSpottickets, escalations, CSAT, campaign engagement
The workflow logevery classification, confidence and review decision
The time studythe "before" for handling time, from [the one-week map](/blog/map-manual-work-before-automating)