A metric without a decision is decoration

A dashboard can be technically flawless and still change nothing. The data refreshes, the filters work, and the meeting opens with twenty minutes of numbers. Then everyone returns to work exactly as before.

The problem is not visual. The organization designed a screen before defining the decision that screen was meant to improve.

Research on data-driven management does find a relationship between evidence-based decision practices and stronger organizational performance. But that evidence does not say that accumulating indicators creates value on its own. Capability emerges when data enters a concrete process: someone sees a signal, interprets what it means, and takes an action with verifiable consequences.

The question that comes before any KPI

Before choosing a metric, complete one sentence: “If this value changes, we will decide to…”.

If the sentence has no clear verb —prioritize, stop, reallocate, investigate, contact, approve— there is no analytics use case yet. There is an information need without an operating design.

This also forces us to name the owner. “The business” does not make decisions; people and teams do, with different timelines, authority, and constraints. An alert that helps Operations may be noise for Finance. A monthly average may serve the board and arrive too late for the person handling an exception today.

The initial work is not adding more data. It is describing who decides, how often, which alternatives are available, and what a wrong decision costs.

A metric does not explain why it moved

The second mistake is treating movement as explanation. Two numbers changing together does not prove that one caused the other. Even in controlled experiments, Microsoft Research has documented recurring interpretation failures that can produce bad decisions: poorly defined metrics, unstable segments, ignored side effects, or conclusions drawn too early.

That is why a serious dashboard separates three layers:

1. Outcome: the final objective we want to protect or improve.
2. Operational signals: changes that allow intervention before the outcome becomes irreversible.
3. Guardrails: what we refuse to damage while optimizing the first two.

Increasing conversion while losing margin is not necessarily an improvement. Reducing handling time while increasing complaints is not either. An isolated metric invites people to optimize a number; a decision system makes the full trade-off visible.

Measurement is for learning, not confirmation

HM Treasury’s evaluation guidance frames evidence as input for deciding whether an intervention should continue, adapt, or stop. That logic is more useful than using a dashboard to defend a decision that has already been made.

A data-driven organization is not one that always obeys the number. It makes hypotheses explicit, observes relevant evidence, and changes course when results contradict expectations. It also records which decision was made and what happened afterward. Without that loop, every meeting starts from zero.

The right design is usually smaller than expected: one recurring decision, a few trusted metrics, agreed thresholds, and a later review. Once that circuit works, it can expand. Before then, adding visualizations only increases the surface nobody governs.

The review worth doing this week

Choose the most frequently viewed dashboard in your organization and take its main indicator. Ask three people what decision should change when that value rises or falls.

If they give different answers —or none— you do not need another tool yet. You need to define the decision process the tool is meant to serve. That work is less visible than launching another dashboard, but it is where Analytics starts creating value.