Lesson 4 of 5, 11 min with the exercise

Read the results properly

Look past the average to the failures that matter.

"92% accurate" tells you very little by itself. What were the other 8%? A wrong tag on a low-priority ticket and a wrong refund amount both count as one error, but they don't cost the same.

• Group failures by type, and count each group.
• Mark how serious each type is.
• Check whether failures cluster in one category, language or customer group.
• Compare with people. The current process has an error rate too, and you measured it earlier.

See it in practice

Invented example: 46 of 50 pass, which sounds fine. But all 4 failures are refunds over $500, sent to the wrong team. The fix may be a simple rule for high-value refunds, and the average would never have told you that.

Common mistake: Comparing against perfection. The fair question is whether it beats the current process at a cost worth paying, and whether its failures are acceptable.