What makes a good analytics dashboard — and what quietly ruins one

Show a dashboard to the person who uses it and ask what they would do differently after reading it. If they cannot answer in a sentence, it is not a dashboard. It is a report with rounded corners.

Most are built backwards: start from the data that happens to exist, fit it all on one screen, wonder why nobody opens it twice. The better order is a person and a question — who reads this, and what will they do next. Everything on screen either serves that or leaves.

What it is actually for

Making a decision faster, not reporting everything measurable. A support lead wants to know whether response times are slipping before customers notice. A finance team wants to see whether spend tracks to plan. Those are different questions with different readers, and each deserves its own view — an executive and an analyst need different dashboards entirely. One screen serving everyone serves nobody.

Six habits that make one work

• Start from the decision and the reader, then build backwards.
• Lead with the number that matters — largest, first, before any supporting detail.
• Match the chart to the question: a trend is a line, a comparison is a bar, a status is often just a number.
• Practise restraint. If a metric changes no decision, it does not belong on the screen.
• Give every number context — a target, a trend, a comparison.
• Keep units, date formats and colour meanings identical throughout, at a size people can actually read.

Context is the one most often skipped, and it does the most damage. £120,000 in revenue means nothing alone. Against a £100,000 target it is a strong month. Against last month's £180,000 it is a problem. Same number, opposite meanings, decided entirely by what sits next to it.

How they quietly go wrong

Dashboards rarely fail loudly. They fill up gradually, one reasonable request at a time, until nobody can find the number they came for.

• Every metric on one screen, so nothing is emphasised and everything must be hunted.
• Vanity metrics — total page views, cumulative sign-ups — crowding out figures that change decisions.
• Decoration over legibility: gradients and 3D effects making charts harder to read.
• No hierarchy, every element the same weight, the eye with nowhere to start.
• Unstated or mixed time ranges, so nobody knows what is being compared.
• Charts that mislead: truncated axes, pie charts with nine slices, dual axes implying a relationship that is not there.

That last one deserves real care, because a misleading chart is worse than no chart. An axis starting at eighty instead of zero turns a two percent move into a cliff. The data is accurate and the picture lies, and someone makes a confident decision on it.

None of this needs exotic tooling. It needs editing — cutting metrics, ordering what survives, labelling it honestly. That editing is the work, and it is what our dashboards and data design actually does.