What Measurement Cannot Fix

A facility that installs monitoring and changes nothing has bought a dashboard. Knowing what measurement can and cannot address keeps expectations honest.

What it identifies and cannot fix

A layout that forces travel. The data shows it clearly; fixing it is a racking project.

Insufficient equipment, which shows as waiting and requires capital.

Demand variation driven by customers, which shows as peaks and requires either flexible staffing or a commercial conversation.

Product mix and packaging that make handling slow, which is a supplier and design question.

Building constraints: dock capacity, aisle width, ceiling height.

Understaffing, which measurement makes undeniable and does not resolve.

In each case the data converts an argument into an arithmetic, which is genuinely valuable and is not the same as a solution.

What it will not identify

Why people do things. A workaround shows as an anomaly; the reason for it comes from asking.

Whether the standard is achievable. Rate data shows what is happening, not whether it is sustainable, and confusing the two is the mechanism behind quota-driven injury.

Near-misses, unless specifically instrumented.

Fatigue, which affects everything and is not directly observable in transaction data.

Quality problems that pass inspection.

Anything about the work nobody records, which in most facilities is a substantial fraction.

The measurement paradox

Measuring changes behaviour, which is sometimes the point and frequently a distortion.

People optimise what is measured at the expense of what is not. A facility measuring picking rate and not accuracy gets a higher rate and more errors.

Balanced measures reduce this and do not remove it.

The distortion is proportional to the consequence attached. Aggregate measurement used for improvement produces little gaming; individual measurement tied to pay or discipline produces a great deal.

The Hawthorne problem

Performance changes when measurement begins, temporarily and for reasons unrelated to any intervention.

Which makes before-and-after comparison unreliable in the first weeks.

Establish a baseline over a long enough period — several weeks at minimum, covering the normal variation — and expect the initial period to be unrepresentative.

Re-measure after the novelty passes, which is the number that matters.

What to do with a constraint you cannot remove

Name it. A constraint stated is manageable; one that is denied produces repeated failed initiatives aimed at the wrong thing.

Quantify its cost, which is what makes the capital case.

Work within it deliberately rather than pretending it is not there.

Stop measuring against a target the constraint makes unreachable, which is demoralising and produces gaming rather than improvement.

The honest positioning

Monitoring tells you where the losses are, with numbers.

It does not remove them, and it does not tell you which are worth removing, which is a judgement about cost, disruption and capital.

A programme that promises improvement rather than visibility will be judged against the improvement, and the improvement usually requires decisions and money that the monitoring project does not control.

Naming a constraint you cannot remove

Some constraints are fixed for years, and how that is stated matters.

Name it specifically: dock capacity, aisle width, building height, a supplier's packaging.

Quantify its cost, which is what makes the eventual capital case.

Stop measuring against targets it makes unreachable, which is demoralising and produces gaming rather than improvement.

Work within it deliberately — schedule around it, buffer for it, sequence to it.

Revisit annually, because the economics change and a constraint accepted in one year may be worth removing in the next.

Establishing the baseline before the novelty

Measurement itself changes behaviour, temporarily, which contaminates early comparisons.

Expect an improvement in the first weeks unrelated to any intervention.

Allow a settling period before the baseline window opens.

Take the baseline over several weeks, covering the normal variation.

Re-measure after the novelty has passed, and report that figure as the result.

Say this in advance, so that the early improvement is not claimed and the later regression is not read as a failure.