BuzbyTimeRead the requirements guide

What monitoring does to people · 7.3

How people adapt to being measured

How people adapt to being measured. What decides it, what it costs, and what usually goes wrong. For another implementation reference, see Monitask's reference.

Adaptation, not cheatingThe framing that matters

People subject to measurement adjust to it. Almost all of that adjustment is rational, most of it is not deceptive, and none of it requires anybody to decide to behave badly.

Treating it as a discipline problem guarantees it continues, because the behaviour is a response to the system rather than to a choice. For another perspective on the surrounding workflow, see Aircall.

Six adaptations, in rough order of frequency

The normal-looking week. Entries adjusted so that the total resembles what is expected. The single most common, and it destroys the fragmentation data the record was collected for.

Batching. Recording on Friday from memory, producing the tidy fiction the first part of this blog describes.

Avoiding unmeasured work. If internal improvement is not counted, it stops happening.

Attribution drift. Hours landing wherever they cause the fewest questions.

Presenteeism. Being visibly present rather than usefully occupied.

Defeating the measurement. The rarest, the most discussed, and the only one anybody calls cheating.

Why the data becomes fiction

Each adaptation is small and the aggregate is a record that describes what people believed was expected of them rather than what happened.

The organisation then makes decisions on it, confidently, because it has a great deal of data. That is worse than having none, and it is the specific outcome this whole blog has been written to avoid.

How to see it

Look at the distributions rather than the totals.

  • Weeks clustering suspiciously close to a round number.
  • Entries in tidy blocks with no short items.
  • Almost nothing in the catch-all category.
  • Corrections at zero.
  • Entries appearing in batches on one day of the week.

Real work is untidy. A record that looks clean is usually a record that was cleaned, and the tidiness is the finding.

What causes it, in order

A target attached to the number. A manager who queried an entry. Categories that do not fit. Entry that costs too much. And an absence of any visible use, which invites people to guess at the purpose and optimise for the guess.

Four of the five are design decisions and the fifth is a communication one. None is about the character of the people entering data.

Undoing it

Slower than causing it. Remove the target, say publicly what changed and why, stop querying individual entries, and wait a quarter. The record improves gradually as people establish that the change is real.

Announcements alone do nothing. What convinces people is the first occasion on which an unusual entry produces no consequence.

The most useful question

Ask people, with nothing at stake, how they decide what to put in. The answers are specific, immediate and generally describe at least two of the adaptations above.

Nobody who has ever asked this has found nothing.

The manager who adapts too

Measurement changes the behaviour of the people reading it as much as the people generating it. A manager with a dashboard checks the dashboard, and the checking substitutes for the conversation that would have produced better information.

This adaptation is less discussed and it is at least as consequential.

Adaptation is information

Every one of the six behaviours says something specific about the system. Batching says entry is too expensive. Attribution drift says a category invites questions. The normal-looking week says somebody believes there is a right answer.

Read them as feedback rather than as failures and the system improves. Read them as failures and they go further underground.

Distinguishing adaptation from a problem

Almost all of it is adaptation. The exception is deliberate falsification for gain, which is rare, is a different matter entirely, and should not be used as the frame for the rest.

Organisations that design the whole system around the rare case produce the common ones.

What to do first

Plot the distribution of weekly totals for the last quarter. If it has a spike at a round number, you have your answer without asking anybody.

A short summary

Adaptation is rational and almost never deceptive. Six behaviours, the commonest being the normal-looking week. The resulting data describes expectations rather than events. Look at distributions rather than totals, because real work is untidy. Four of the five causes are design decisions. And undoing it takes a quarter of consistent behaviour rather than an announcement.

The tidy record as a warning sign

Auditors in other fields treat implausibly clean data as a prompt to look harder. The same instinct serves here: a workforce whose weeks all total the same figure is telling you something, and it is not that everybody works identically.

One line to carry

A record that looks clean is usually a record that was cleaned, and the tidiness is the finding.

Where to start

Plot last quarter's weekly totals. The shape of that distribution tells you whether you have a record or a performance.

The adaptation nobody objects to

Recording accurately because the record is useful to you. It is an adaptation too, and it is the one every design decision in this blog is trying to produce.

People adapt to whatever the system rewards. The question is only ever what it rewards.

A last observation

Every organisation gets the record it designed for, which is rarely the record it wanted. The gap between those two is entirely explained by the six adaptations above, and closing it is a design exercise rather than a management one.

Also in what monitoring does to people