8 customer care AI assessment indicators include self-processing rate, accuracy, response time, security, transfer rate, processing time, cost and retention rate. They show whether the AI is doing well or just "looking good on the board". Zenify recorded AI accuracy of 98.4%, response time of 1.2 seconds.
TL;DR
- 8 core indicators to evaluate customer care AI.
- Self-processing rate: How much can the AI carry.
- Accuracy: whether the answer is correct or not.
- CSAT: are customers satisfied?
- Measure weekly, not emotionally.
8 indicators to evaluate AI customer care
Index | Meaning | Reference criteria
1. Self-processing rate | How many % did AI solve | 60-80%
2. Accuracy | How many % correct answer | >90%
3. Response time | How fast | <5 seconds
4. CSAT | Satisfied customers | According to industry standards
5. People turnover rate | What falls on you | <30%
6. Processing time | Is it resolved quickly? Descending
7. Cost per treatment | How much cheaper than others | 30%+ off
8. Retention Rate | Will customers come back? Ascending
Explanation of each indicator
1. Containment rate
Percentage of AI that solves problems on its own without needing a human (see Containment rate).
The higher it is, the better the load, target 60-80%.
2. Accuracy
% of correct answers, checked by QA sample (see AI QA).
Zenify: 98.4%.
3. Response time
From customer sent to AI reply, Zenify: 1.2 seconds (see Reduce response time).
4. CSAT
Guest satisfaction after interaction, measured by both the AI part and the human part (see What is CSAT).
5. Tỷ lệ chuyển người
% of AI conversations have to switch people, too high means weak AI or tight boundaries.
6. Processing time (AHT)
Total resolution time, compared to before AI (see Reducing processing time).
7. Cost per treatment
Cost/voucher, Zenify: 34% cost reduction.
8. Retention rate
Customers return to buy/repeat, reflecting the overall experience.
Correct way to measure
Closing the goal: Do you need AI to reduce load, speed up or increase satisfaction?
Measure before/after: record baseline before implementation.
Weekly measurement: Do not measure emotions, do not change the scale continuously.
Qualitative combination: read a few AI conversations, the data doesn't tell the whole story.
Adjustment: If any index is weak, correct that.
Note
Don't just look at one indicator: High self-processing but low CSAT = AI answers quickly but wrongly (see What to do when AI is wrong).
Same benchmark comparison: compared to your baseline, not to advertising.
Zenify data: used as reference, measured for yourself.