Better Numbers Do Not Always Mean Better Outcomes

At the end of the month, every arrow points upward. More replies have been sent; average response time has fallen. You close the report and find a familiar message: “My original problem still isn't solved.”

This invented scene does not describe a particular company. It exposes a question a polished chart can hide: are the actions being counted still moving toward the intended result?

A popular Reddit question challenges the historical basis of the widely repeated “cobra effect” anecdote. This research did not establish that history, so the bounty-and-breeding story is not retold here as fact. Memorability helps a story spread. Historical verification requires separate evidence.

Remove the legend and the question remains testable

Suppose a small support team wants users' problems solved but rewards only the number of replies each day. A difficult issue takes thirty minutes to investigate; a template response takes thirty seconds. Under that rule, the person investigating complex problems can appear less productive.

This does not mean staff will inevitably provide poor service, nor is it a measured causal finding. It reveals a direction of incentive that can be investigated: people may optimize the rewarded action when it is narrower than the actual goal.

Look beyond attitudes. Do the same problems return? Do replies generate avoidable follow-up questions? Are difficult cases left untouched? Changes in incoming volume and complexity also need attention; not every change can be attributed to the reward.

A replacement metric can inherit the problem

“Tickets closed” seems closer to resolution than “replies sent.” But without a meaningful check, premature closure can improve that number too. Rewarding only satisfaction may discourage people from taking difficult cases likely to disappoint.

The remedy is not one magical measure that can never be misused. Let several kinds of information check one another: processing time, unresolved follow-ups, sampled case reviews, and feedback users choose to provide. Numbers suggest where to look; records help explain what happened there.

Explain how samples are chosen. Selecting only the best replies places the desired answer inside the invitation list before the review begins.

Give every target a sentence in ordinary language

Before choosing a percentage, write the goal plainly: “The reader can complete the task after following the instructions.” Then ask how a proposed number relates to that sentence and when the relationship could break.

For a blog, clicks, reading time, and saves each reveal something. A long visit may mean depth or confusion. A quick departure may mean disappointment or a clear answer. One number without context cannot reveal the reader's mind.

A target worth keeping allows a report to look worse temporarily and allows someone to say the outcome has not improved. We collect numbers to notice those messages earlier, not to remove their place in the monthly report.