When the goal becomes a target, the metric dies

A dental plan operator, 260 people, a 40-seat contact center. Leadership decides to attack average handling time: cut it 30% in one quarter, with a bonus attached. Within two months, average time falls 40%. The target is celebrated at the all-hands.

The following quarter, formal complaints double and repeat calls — customers calling back about the same issue within seven days — rise 60%. The team didn't cheat: they did exactly what was asked. Ending the call sooner is the most direct way to cut average time, and the metric didn't tell a solved problem apart from a closed call.

01The idea: the measure stops measuring #

Charles Goodhart, then an economist at the Bank of England, made an observation about monetary policy in 1975: any observed statistical regularity tends to collapse once pressure is placed upon it for control purposes. The short version that became famous — 'when a measure becomes a target, it ceases to be a good measure' — is credited to anthropologist Marilyn Strathern, who put it that way in 1997 while discussing the audit of British universities.

The distinction matters so you don't misquote: Goodhart wrote about monetary aggregates; the managerial version of the line, now repeated in every metrics review, is Strathern's paraphrase. Both describe the same mechanism — behavior adjusts to whatever is measured, and the indicator loses its link to what it used to represent.

Jerry Muller, in The Tyranny of Metrics (2018), documents the pattern across health care, education, policing and the military. He names what he calls metric fixation: the belief that replacing judgment with numbers produces objectivity. The recurring result is short-term manipulation — cherry-picking easy cases, reclassifying records, pushing the problem to another team — and the quiet abandonment of whatever isn't measured.

02The four kinds of gaming you'll find #

Gaming is rarely dishonesty. Most of the time it's a rational response to a badly designed system, made by people who want to keep their jobs:

1. Selection. Prioritize the cases that close fast and push the hard ones to the back of the queue, where they age without showing up in the metric.
2. Reclassification. Log the case under a different category so it doesn't land in the count being tracked.
3. Transfer. Close your own ticket by opening a request with another team; your metric improves, the customer's experience doesn't.
4. Sacrificing what isn't measured. Cut time by dropping the confirmation, the record and the explanation — exactly what prevents the repeat call.

One aggravating factor specific to mid-size companies: with small teams the metric belongs to the team, not the individual, and the pressure of a shared bonus tends to silence whoever flags the problem. Reporting that a target is being hit badly means taking money from a colleague. That's why target design is also a culture question — the subject of the leader who only asks for status isn't leading.

03The math behind the pretty number #

Stated assumptions, at the scale of that 40-seat center, in a mid-size Brazilian company (figures in Brazilian reais). Redo it with your numbers:

Want to see how this looks inside a real operation? Explore the platform.

The third line: 40 seats, 5,800 calls a month, repeat calls going from 8% to 12.8% of cases, which is about 280 extra repeat calls a month; at roughly R$ 92 per completed contact, that's around R$ 310 thousand a year. The math ignores the effect on cancellations, which the company couldn't isolate — and which is probably larger than everything else here.

04How to design measurement that holds #

Andrew Grove, in High Output Management (1983), proposed the cheapest defense against this effect: paired indicators. Every volume or speed metric comes with a quality metric that degrades when the first one is pushed. Average handling time travels with repeat calls; units produced travels with scrap rate; deals closed travels with 90-day delinquency.

The practical rule that follows is simple: no metric goes into a bonus alone. It also helps to cap the number of paid targets, to separate tracking metrics from contractual ones, and to review periodically whether the measure still describes what you wanted to observe. It's the same discipline of telling signal from noise found in variation isn't error, and the same care that separates objective from indicator in from MBO to OKR — where the advice not to tie OKRs directly to pay comes from exactly this problem.

05What data-driven management answers #

Muller doesn't conclude that you should stop measuring — nor would that be possible in a company that has to decide. His conclusion is about proportion: metrics are an input to judgment, not a substitute for it, and the cost of measuring (time, distortion, induced behavior) belongs in the calculation before one more dashboard gets built.

What makes this workable is having the raw data close to the decision. When every contact, incident and reopening is recorded inside the flow of work, the quality pair doesn't require a new project — it's already in the same place. And the manager can look at the individual case behind the number before concluding anything from the average.

At Relevanti, the operations, processes and intelligence modules keep the indicator and the case side by side — see the platform, the per-team cut in solutions, or bring your target list to a conversation.

No indicator survives the weight of a bonus intact. A manager's job isn't finding the perfect metric: it's designing the pair that keeps the metric from lying.