The first step is to measure whatever can be easily measured

At the review, response time, conversion, volume and satisfaction each have a decimal. Three recent examples of the recurring customer problem are read aloud because the system has nowhere to record them.

The review is now constrained to the recorded categories.

Easy numbers deserve attention. They are fast, comparable and often less vulnerable to whoever tells the best story. The rot begins in the next move: anything difficult to count receives a made-up score, then a small weight, then no budget. Customer anxiety, degraded trust, future maintenance and work transferred elsewhere receive no equivalent entry.

The fashionable answer is more metrics. Sometimes that helps. Sometimes it creates six proxies and a composite index that nobody can explain without notes. Extra proxies leave the original weakness intact.

Start with the decision. What must be understood to make it responsibly? Use direct measures where possible. Where judgement remains necessary, keep it visible as judgement: examples, independent ratings, written reasons and ranges rather than a decorative 7.3.

Compare each measure with evidence beyond its definition: faster replies alongside repeat contacts, sales alongside returns, output alongside defects. The contradiction is often more useful than the score.

At the next review, the customer problem is still difficult to count. Three recent examples are read aloud while the owner decides how it will be recorded. One becomes a recurring category; the other two remain individual cases attached to the decision.