KPI worth fighting for: how to tell signal from dashboard decoration
A KPI only matters if it can change a decision. Everything else is decoration on a dashboard.
At some point in every strategy or operations meeting, a slide with metrics appears. Ten, twenty, thirty indicators. Traffic-light colours. Up and down arrows. Everyone looks, nods, and moves to the next agenda item.
I have learned to watch for one simple sign: if nobody said "so we need to do X" after that slide - the slide did nothing. It created a feeling of control without producing a decision.
This is not a visualisation problem. It is a problem of metric selection.
The only criterion a useful KPI must pass
A useful KPI is a number whose change changes behaviour. Not "good to know", not "interesting to watch" - specifically: if the value drops, some specific person does something specific differently.
If there is no such person and no such action, it is not a KPI. It is a number in a report.
The test is simple. Ask yourself: "If this metric drops 20% next month - who finds out within 24 hours, and what exactly does that person do differently?" If the answer is vague or nonexistent, the metric is decorative.
How decorative monitoring accumulates
Companies end up with bloated dashboards through a few familiar paths.
The first is copying someone else's system. A list of recommended metrics from an article or a book gets added to the report. The metrics belong to a different context, and the connection to decisions was never established.
The second is accumulation. Every quarter someone asks to add one more indicator. Old ones are never removed. A year later the dashboard is an archive of requests, not a current picture.
The third is fear of missing something. The assumption is that seeing more means controlling more. In practice it is the opposite: the more indicators there are, the higher the chance that the important signal drowns in background noise.
Signal metrics versus accounting metrics
I divide metrics into two classes.
Accounting metrics are for reporting, history, and audit. Revenue for the period, order count, shipment volume. Important numbers, but they describe what has already happened. You cannot manage them in real time - you can only analyse them after the fact.
Signal metrics are leading indicators. They tell you what is happening right now and what will happen if nothing changes - conversion at a key funnel step, response time on an incoming request, defect rate at an intermediate quality check. The same distinction shows up in security: virus counts make a bad KPI for executives for exactly the same reason - they are accounting metrics, not signal metrics.
Signal metrics belong on the operational dashboard. Accounting metrics belong in periodic reports.
The problem with most monitoring systems is that they are packed with accounting metrics and contain almost no signal metrics. Visually rich, historically deep, operationally useless. The same problem appears in self-service reporting: giving everyone freedom to build their own metrics and shared reports without a shared definition of which numbers drive decisions produces the same dashboard proliferation.
How to cut down to a working set
I go through the metric list with four questions:
- Who specifically looks at this indicator regularly - not "everyone", but a name and a role?
- What decision does that person make based on this indicator?
- How often does the indicator actually drive something - or is it just recorded?
- What happens if this metric disappears from the report for three months?
If the first three questions have no concrete answers, the fourth becomes the test. If nobody notices the metric is gone, it was decorative.
A working operational dashboard typically holds five to ten indicators. Not twenty-five. Not forty. Five to ten - the ones that passed the test.
A simple test for the next meeting
At the next metrics discussion, try an experiment: pick three indicators from the current set and ask the people in the room to name the last decision made on the basis of each one.
If the answers are specific - those are live metrics. If the answers are about "we just monitor the situation" - that is a signal that the monitoring system is working as decoration, not as a management tool.
Fixing this is not difficult. But it starts with the question, not with adding another indicator.