In 1976, sociologist Donald Campbell formulated a rule that sounds similar to Goodhart's law, but speaks of something different and, perhaps, more unpleasant: the more a social metric is used for decision-making, the more it distorts and corrupts the very processes it was intended to reflect.
The difference with Goodhart is fundamental. Goodhart is about a metric ceasing to be reliable—it detaches from reality and begins living a life of its own. Campbell goes further: the metric doesn't just lose touch with reality, it reshapes that reality to fit itself. The metric itself does not deteriorate—what deteriorates is the thing it was measuring.
Campbell wrote this while observing education and social programs. A classic example is standardized testing. As long as a test merely measures the level of knowledge, it works. As soon as school funding and teachers' careers start depending on test scores, education reorganizes itself around preparing for the test. Formally, scores go up. In reality, children learn how to pass the test, not what it was created to measure. The learning process itself has changed—not just the result of its measurement.
In HR, this mechanism works in exactly the same way, and it is harder to notice than Goodhart's distortion of numbers.
If time-to-hire becomes the key recruiting metric, it's not the reporting that changes—it's the hiring process itself. People start selecting candidates who are easier and faster to push through, abandoning complex searches, and closing positions with whoever is available. If team turnover becomes a manager's key performance indicator, managerial behavior changes: difficult conversations are delayed, and underperforming employees aren't fired, but quietly transferred to other departments. If engagement scores are tied to a manager's bonus, how the survey is conducted and the conversations surrounding it change.
In all these cases, the number looks better. And in all these cases, the process for which it was all started has gotten worse.
From this follows what Campbell essentially argues: you cannot simultaneously use a metric for control and rely on it as an honest measurement. The greater the decision pressure on a metric, the less suitable it is for diagnosis.
What helps:
separating diagnostic metrics from management metrics—what people are held accountable for and rewarded for should not be the same thing you use to understand how things really stand;
looking at a metric together with what it could have spoiled (time-to-hire together with quality of hire and probation retention);
limiting the weight of any single metric in evaluations, because distortion grows with the stakes;
and honestly checking what has changed in people's behavior after introducing a metric—if nothing has changed, it affects nothing, and if something has changed, you should make sure what that actually is.