The Sense and Nonsense of Chasing KPIs

Today, we can measure almost everything. Revenue, margins, lead times, utilization rates, customer satisfaction, number of calls, tickets, reports, conversions, clicks. Modern software can turn almost any process into a dashboard.

But this creates a strange paradox: the more we can measure, the easier it becomes to forget what we are actually trying to achieve. And that is exactly where KPIs often go wrong.

Not every metric is a KPI

A metric tells you something about your organization. A KPI tells you something about your progress towards an outcome that matters.

The difference may seem small, but it is fundamental.

The number of completed cases can be a metric. If your strategic objective is to process cases faster, turnaround time might become a KPI. And if speed can come at the expense of quality, you may also need an indicator for errors or customer satisfaction.

A KPI therefore never really exists in isolation. Without an objective and context, it is just a number.

This idea was also at the heart of the Balanced Scorecard, introduced by Robert Kaplan and David Norton in 1992. Their criticism was that organizations focused too heavily on financial results. Those figures primarily tell you what has already happened. They proposed combining financial measures with indicators relating to customers, internal processes, and an organization's ability to learn and improve.

The original idea was not: find one number and manage everyone against it.

Quite the opposite: look at performance from multiple perspectives, because no single number tells the whole story.

The problem starts when the number becomes the goal

Imagine an employee is expected to produce a certain number of reports each year. At the end of the year, their KPI is red. They missed their target. On paper, the conclusion seems simple: they underperformed.

Then their team lead adds some context.

This employee is constantly called in when other teams or offices run into problems. They handle exceptions, help colleagues, and put out fires that are not tracked anywhere. The number itself is not wrong. They really did produce fewer reports.

The mistake is assuming that this means they created less value.

Worse still, if we judge them exclusively on that number next year, we can predict what will happen. They will spend less time putting out fires and more time producing reports.

The KPI turns green.

The organization performs worse.

This effect is surprisingly well researched

A well-known formulation of Goodhart's Law captures the problem neatly:

When a measure becomes a target, it ceases to be a good measure.

That does not mean targets are inherently bad. It means people adapt their behavior to the way their performance is measured.

Donald Campbell described a similar effect: the more heavily a quantitative indicator is used in decision-making, the greater the pressure to influence that indicator, and the greater the risk of distorting the process it was intended to measure.

Holmström and Milgrom added another important insight in 1991. People typically have multiple responsibilities, but some are much easier to measure than others. Strong incentives around measurable work can therefore draw attention away from work that is just as valuable, but much harder to capture on a dashboard.

Then there is surrogation. This occurs when the indicator gradually takes the place of the objective it was supposed to represent.

Suppose we want to provide better customer service. We choose average handling time as an indicator. Over time, the discussion is no longer: how can we help customers better? Instead, it becomes: how can we reduce our handling time?

Research shows that managers can indeed begin to treat performance measures as if they were the strategy itself. This effect can become even stronger when compensation is directly tied to a single measure.

So should we stop using KPIs?

Not at all. Measurement is not the problem. The problem begins when we confuse measuring with understanding.

A good dashboard therefore does not need as many KPIs as possible. It needs as few indicators as possible that, together, provide enough information to make good decisions.

A few simple principles can help.

Start with the objective, not the data.
First ask what you are trying to achieve. Only then decide which number can tell you something meaningful about your progress.

Combine outcome and leading indicators.
Revenue tells you what has happened. Pipeline, usage, or customer behavior can tell you something about what may happen next. Both are valuable, but they serve different purposes.

Measure the counterbalance too.
If you optimize for speed, monitor quality as well. If you optimize for volume, look at value too. This makes trade-offs visible instead of allowing them to remain hidden.

Ask what happens when someone deliberately tries to maximize the number.
If the easiest way to hit a KPI is bad for the customer, colleagues, or the organization, you have designed an incentive problem.

Leave room for context.
Not everything that creates value is easy to measure. A dashboard can inform a conversation. It cannot replace one.

Perhaps that is the most important question about KPIs

Software gives organizations access to more and more data. AI will only accelerate that trend. As a result, the technical challenge is increasingly less about: can we measure this?

In most cases, we can.

The strategic challenge is:

If this number changes, what does it actually tell us about what we are trying to achieve?

And when a KPI turns red, perhaps the first question should not be who is responsible.

The better question is:

What is this number not measuring?