BuzbyTimeRead the requirements guide

Reading the numbers · 2.2

What happens when somebody is measured on it

What happens when somebody is measured on it. What decides it, what it costs, and what usually goes wrong. For a practical comparison point, see Monitask's Chromebook employee tracking.

The mechanism, not the moralityThe general result

Attach a consequence to a number and the number improves. Whether the thing it stood for improves is a separate question, and frequently the answer is no.

This is not a claim about dishonesty. Three of the four routes below require nobody to do anything they would describe as wrong. Teams comparing the wider operating context can also consult Zapier.

Four routes, three of them innocent

Effort moves to the measured thing. Attention shifts from what is not counted to what is.

Selection changes. Take the work that scores well, avoid the work that does not.

The definition is satisfied without the substance. An hour attributed to a client because that is where it technically belongs, when everybody knows it was internal.

Falsification. The rarest and the one everybody imagines first.

What this looks like in time data specifically

  • A billable percentage target: attribution drifts toward clients, and the figure improves while margin does not.
  • A utilisation target: internal work stops being recorded as internal.
  • An hours target: presence is recorded rather than work, and the record becomes a register of attendance.
  • A variance-to-estimate target: estimates inflate, and forecasting gets worse.

Each of these is common, each is predictable, and each is produced by a target that somebody set for a reasonable reason.

Surrogation

The subtler failure. People stop treating the number as a proxy and begin treating it as the goal. A manager who set out to improve margin and now wants the utilisation figure has not become cynical; the substitution happened without anybody noticing.

It is more common than deliberate gaming, because it requires nothing but repetition.

What reduces the damage

  • No individual targets on anything derived from a time record.
  • Several measures rather than one, chosen so that improving one at the expense of the work degrades another.
  • A qualitative channel, because the people entering the data know exactly how the number is being met.
  • Separating the numbers used for learning from the numbers attached to consequences.

The last is the strongest and the first to be abandoned when somebody wants accountability.

How to tell it is happening

A measure improves sharply and nothing the measure stood for improves with it. A distribution develops a spike just above a threshold. And the people closest to the work stop discussing the number in front of whoever set it.

All three are observable without accusing anybody, which is the only way to look for this without making it worse.

Targets that are almost safe

Some measures resist this better than others, and they share a property: the only way to move them is to do the thing. Cash collected. Whether the invoice went out. Whether the payroll run was correct.

Prefer those where a target is genuinely needed, and keep proxies out of anybody's objectives.

The team-level exception

A target on a team measure is less corrosive than one on an individual, because gaming it requires collusion and because colleagues object.

It is not safe, only safer, and the same signature applies: if the number improved and nothing else did, the number is what improved.

Undoing it

Removing a target is harder than setting one, because its removal is read as a judgement about performance. Announce the reason, in terms of the measurement rather than of anybody's work, and expect the behaviour to take a quarter to unwind.

The report that becomes a target by accident

Nobody has to set one. A figure shown at a monthly meeting, commented on twice, becomes a target without any decision being taken, and people manage it accordingly.

Which is why what appears in a recurring report matters as much as what appears in objectives. If you would not set it as a target, think before putting it on a slide every month.

Asking people how the number is being met

The cheapest diagnostic available and it requires that nothing happens to whoever answers. Asked properly, people describe exactly how a figure is produced, including the parts nobody intended.

A worked example

An agency sets a seventy per cent billable target. Within two quarters the figure is met. Margin is unchanged, internal project work has stopped being recorded, and the operations lead can no longer make the case for a hire because the hours supporting it have been attributed elsewhere.

Nobody did anything they would call wrong, the target was met, and the business lost the ability to answer a question it cared about more.

Where to start

List every number derived from time data that appears in somebody's objectives. For each, ask what the cheapest way to improve it would be, and whether that way involves doing better work. Most lists lose an entry or two.

Reviewing your own measures annually

For each recurring number, ask what it was introduced to answer and whether it still does. Measures outlive their questions routinely, and one nobody remembers the purpose of is being optimised for its own sake.

A short checklist

No individual targets on derived numbers. Several measures. A channel where people can say how a figure is being met. And an annual look at whether each measure still answers the question it was introduced for.

One sentence to carry

Any number attached to a consequence will be optimised, by reasonable people, in ways nobody intended. Design as though that is certain, because it is.

Also in reading the numbers