QA score 90 sounds "very good" — but it could be a phantom measure: high scores because the criteria are easy, little scoring, or people click easy scores to get good scores — while real customer experience doesn't improve. QA scores are only meaningful when the criteria are spot on, the sample is representative, and the scores are used for training. Zenify recorded a 21% increase in conversion rate (approved) for businesses doing real QA.
TL;DR
- QA score 90 is only meaningful when the criteria are strict and the scoring is real.
- Virtual yardstick: nice score but customer experience remains the same.
- Imaginary causes: easy criteria, small/not representative samples, easy scoring.
- Contrast with CSAT/churn — High QA and customers leaving is a problem.
- QA score is for training, not to "show off your score".
What is QA score?
QA score is the aggregate score of conversation quality according to a criteria table — usually calculated as a percentage or on a scale (see QA in Customer Service). Common target: 90/100.
Problem: the number "90" does not indicate quality by itself — it depends entirely on criteria and scoring method.
When is a QA score of 90 a virtual measure?
1. Easy criteria
Check "will you say hello", "will you send a price" — things anyone can do.
Ignore real quality: whether the problem is solved, whether the customer is satisfied.
2. Few points, not representative
Only mark easy conversations, avoid difficult ones.
Score a few conversations/month → score does not reflect reality.
3. Dot easily
The examiner is afraid to give low scores → scores swell.
No calibration → each person scores one type (see Calibration).
4. Points do not go into training
Finish grading the report "90" and leave it there.
No one fixes anything → beautiful spots, unchanged quality.
Signs that QA score is fake
QA 90 but low CSAT, high churn — clear contradiction (see CSAT vs NPS vs CES).
Customers complain a lot but the QA score is still good.
All employees are 90 or older — easy doubt.
QA score does not change despite process changes — the index is not sensitive.
How to make a reliable QA score
1. Criteria for measuring what is truly important
Does measuring "whether the guest was addressed" — not just "whether they were greeted".
Attach criteria to business results (closing orders, returning customers).
2. The sample is representative and large enough
Mark all types of conversations evenly, including difficult ones.
At least 10 conversations/employee/month.
3. Standardize the scorers
Periodic calibration so that everyone can score at the same standard.
Check the difference between the scorers.
4. Check with other results
Compare QA score with CSAT, churn, complaint rate.
High QA but customers leave = QA is fake.
5. Include points in the training loop
Low score in any criterion → train that criterion (see Coaching from QA data).
Re-scoring after training to measure progress.