The symptom
Performance drift is what happens when the number keeps looking fine after the thing the number was meant to represent has stopped being fine. It is a specific and recognisable symptom, distinct from a straightforward decline: the measure does not fall. It holds steady, or improves, while the underlying reality it was built to track moves in the opposite direction. The gap between the two is invisible from inside the reporting system, because the reporting system only shows the measure.
The most common first reading of this symptom is that it is not really happening — that the measure is simply reflecting genuine improvement and any concern is a case of looking for problems where none exist. This reading is understandable, because drift by its nature produces evidence in its own favour: the number went up, and a number going up is, on the face of it, good news. It is also the reading that keeps drift invisible for as long as possible, because it treats the measure as equivalent to the reality rather than as a proxy for it — a distinction that only becomes obvious once the two have separated far enough to be undeniable.
Why it happens
The research explanation for this symptom is old and specific. In 1956, V. F. Ridgway published one of the earliest systematic examinations of what happens when organisations rely on quantitative performance measures, published in Administrative Science Quarterly through Cornell University’s Johnson Graduate School of Management. Ridgway documented cases including public employment interviewers who, once evaluated on the number of interviews conducted, began conducting fast interviews and placing fewer applicants — and law-enforcement investigators, given a monthly case quota, who selected easier cases as the month-end deadline approached. His conclusion, stated plainly, was that “the cure is sometimes worse than the disease”: measurement, applied indiscriminately and trusted too completely, can produce exactly the kind of distortion it was meant to prevent.
What leaders typically conclude when they first notice a discrepancy is that the process around the measure has a flaw — a data entry error, a definitional ambiguity, an edge case that needs tidying up. This is a reasonable first hypothesis, and it is sometimes correct. What the evidence suggests, though, is that it is more often something else: not a flaw in the process, but a rational response to the process by the people working inside it. A measure creates an incentive the moment it is introduced, whether or not that was the intention, and people are generally quick to discover the paths of least resistance a new incentive creates. Nothing about finding that path typically involves breaking a rule. It usually involves following the rules exactly as written, in the one direction the rules happen to permit.
There is a competing explanation worth taking seriously before settling on distortion: sometimes the measure has simply gotten easier to satisfy for a legitimate reason — genuine process improvement, better tooling, a real change in the underlying work. Distinguishing a genuine gain from a gamed one is not always straightforward from the measure alone; it usually requires looking at what happened to the categories and exceptions around the measure, not just the headline figure.
What it looks like in practice
Consider an organisation where a new measure is introduced specifically to improve responsiveness — how quickly incoming items are addressed. Within two months, the responsiveness figure has improved markedly, which is celebrated as an early win for the new system. In the same period, the volume of items marked “pending information” — a legitimate category for cases genuinely waiting on something outside the team’s control — has tripled. The first explanation offered is a data entry issue: perhaps the new categorisation options have confused staff, or the definition of “pending” needs clarifying.
A closer look finds something else. Staff have discovered, correctly, that items marked pending do not count against the responsiveness clock. Nothing about this is against any written rule. The category exists for genuine cases, and its use has simply expanded to cover cases that are ambiguous rather than genuinely stalled. The measure has not been broken. It has created a route around itself, and the route is entirely legitimate — which is what makes it hard to challenge and easy to defend, on both sides, when it is eventually noticed.
This shape of drift recurs with recognisable variations. Sometimes the escape route is a category, as above. Sometimes it is timing — work reclassified or logged just before or after a measurement cutoff. Sometimes it is scope — a definition of “resolved” or “complete” that technically applies before the harder part of the work is actually done. What these variants share is that none require dishonesty in any individual instance. They require only that the measure and the reality it represents have enough daylight between them for a legitimate path to open up — and that path, once discovered by anyone, tends to be discovered by everyone, because it works.
A useful way to think about the speed of this spread is that it rarely depends on any formal communication of the workaround. Nobody needs to announce that pending items stop the clock; the pattern is discoverable by anyone paying reasonably close attention to their own workload, which is precisely why it tends to appear in the data as a fairly sudden, organisation-wide shift rather than a slow trickle — the trickle happens in discovery, but adoption, once discovered, spreads quickly through ordinary imitation of what visibly works for a colleague under the same pressure.
For how this pattern presents in other kinds of organisation, see other sectors.
Questions worth considering
- What behaviour does this measure actually reward, as distinct from what it was designed to reward?
- If a category or exception used within this measure grew sharply, would anyone be looking in a place likely to notice?
- Has this measure been checked against a second, independent view of the same activity recently — or only against itself, over time?
More on this topic: Performance.
Your team is facing similar challenges?
Subscribe to The Riddle Ledger
Perspectives are sent as they are published, with no promotional email, and you can unsubscribe at any time.