Every metric can be improved without improving anything real. These are the six we see gamed most often — usually unintentionally, by teams optimising exactly what they were asked to.
1. Read rate
Why it misleads: WhatsApp read rates are high almost regardless of quality, so the number cannot distinguish a good campaign from a poor one. It also only counts where read receipts are on.
Watch instead: reply rate and conversion. Use read rate only to detect sudden drops.
2. List size
Why it misleads: it only goes up, and every contact added to it is a future cost. A list that doubled by importing unverified numbers is worse than one that did not grow at all.
Watch instead: engaged contacts — people who interacted in the last 90 days — as an absolute number and as a share of the list.
3. Containment rate
Why it misleads: it improves whenever you make reaching a human harder. A bot that traps people scores brilliantly.
Watch instead: containment alongside repeat contact rate and escalation-after-answer. Deflection and resolution look identical on containment alone.
The general pattern: any metric that improves when you make the customer's life harder needs a paired metric that gets worse when you do. Report them together or not at all.
4. Average response time
Why it misleads: the mean hides the tail, and the tail is what customers complain about. It also rewards fast, unhelpful replies — a "we are looking into it" stops the clock without helping anyone.
Watch instead: median and 90th percentile, plus subsequent response time, plus repeat contact rate.
5. Attributed revenue
Why it misleads: it credits the channel for purchases that would have happened anyway, and messaging targets exactly the people most likely to buy regardless.
Watch instead: incremental revenue measured against a holdout group.
6. Messages sent
Why it misleads: it measures activity and cost, not outcome. It rises when you spam people, which is the opposite of progress, and it is also the wrong billing unit since you pay per conversation.
Watch instead: conversations, and outcomes per conversation.
Building a report that resists gaming
- Pair every efficiency metric with a quality metric.
- Use medians and percentiles rather than means.
- Include at least one negative metric — blocks, opt-outs, complaints — prominently rather than in an appendix.
- Report trends across periods, not single-period snapshots.
- Segment before averaging, because a single blended average usually describes nobody.
And read the conversations
No dashboard will tell you that your tone is wrong, that an automated answer is technically correct and unhelpful, or that customers are confused by a particular instruction. Twenty transcripts a week will. Every team we know that does this finds something in the first month that no metric had surfaced.