Dashboards That Answer a Question, and Dashboards That Do Not

The typical marketing dashboard displays thirty numbers, is checked briefly each morning, and changes nobody's behaviour. It is not badly built. It was built to display data rather than to answer a question, and those produce different artefacts. For time-based reporting, online timesheets are a simple example of data that only becomes useful when the dashboard answers a defined question.

The test

What decision does this dashboard support, and who makes it?

If there is no answer, it is a data display. Data displays have a use — reassurance, awareness, an audit trail — and they should be labelled as such rather than treated as decision support.

Most dashboards fail this test, and the failure is visible in how they are used: opened, glanced at, closed, with no action in between.

The central failure: numbers without context

A number alone cannot be interpreted.

Conversion rate: 2.4%. Is that good? Compared with what? Is it inside normal variation or outside it? Has it been drifting?

The single highest-return change to any dashboard is putting each metric next to its own recent history, with the normal range visible.

A sparkline over twelve weeks answers, at a glance, the only question that matters: is this unusual? Almost every "the number moved, what happened" investigation would not have started if the range had been visible, because most movements are inside it.

Comparison against the previous period alone is not enough. Week on week is noisy, and it invites reading noise as signal. Against the same week last year introduces a different confound. The distribution of the last twelve weeks is more informative than either.

What a decision dashboard contains

Few metrics. Five to seven. A dashboard with thirty is a place where the important number hides.

Each metric with its history and normal range.

Absolute numbers next to rates, so a denominator change is visible. See metrics that mislead.

A stated definition, or a link to one, next to anything ambiguous.

A marker for known events. Campaign launches, price changes, site deployments, consent banner changes. This one addition answers more questions than any additional metric, because most unexplained movements have an explanation on that list.

An indication of data quality. If the reconciliation ratio is drifting, that belongs on the dashboard, because it changes how much weight the other numbers deserve. See data quality.

What to leave off

Metrics nobody has ever acted on. Review what has actually been used in the last quarter and remove the rest. This is uncomfortable and it typically removes half.

Vanity metrics. Impressions, reach, sessions. Test: if this doubled and nothing else changed, would anything be better?

Everything the platform offers by default. Default dashboards are built to display the platform's capabilities, not to answer your question.

Modelled numbers presented as observed, without a label.

Precision nobody needs. Two decimal places on an estimate with a wide interval is precision theatre.

Different audiences want different things

Trying to serve everyone with one view is why dashboards accumulate metrics.

Practitioners need diagnostic detail — campaign level, daily, enough granularity to find what changed.

Managers need the small number of metrics that indicate whether anything requires attention, with enough history to tell signal from noise.

Executives usually need three numbers and a sentence, and are best served by a written summary rather than a dashboard. The most common mistake is building an executive dashboard nobody opens when what was wanted was a paragraph once a month.

Build separate views. They share a data layer and they are different products.

The maintenance problem

Dashboards decay, and the decay is invisible.

A metric definition changes and the historical series becomes inconsistent, with no marker on the chart.

A source breaks and a chart shows zero, which looks like a business event rather than a broken pipeline.

Someone adds a metric for one meeting and it stays forever.

The underlying data changes shape — a new channel, a renamed campaign type — and a filter silently stops matching.

Practices that help: an owner for every dashboard; a review each quarter that removes rather than adds; alerting on data freshness so a stale chart announces itself; and a changelog on the dashboard, because "why does this look different from last month" otherwise has no answer.

Alerts do the job dashboards are asked to do

A dashboard requires someone to look. An alert arrives when something happened.

For anything genuinely important, build an alert instead, on a threshold defined in advance.

Alert on movements outside the normal range, not on fixed thresholds — a fixed threshold either fires constantly or never.

Every alert needs an action. One that produces no action trains people to ignore alerts, including the ones that matter.

The most valuable alerts in marketing data: conversion volume dropping per channel, a source going to zero, the reconciliation ratio moving, and data freshness.

Rebuilding one

If an existing dashboard is not working:

Ask who uses it and what they do differently because of it. Frequently nobody and nothing, which is the answer.

List the decisions it should support.

Keep only the metrics those decisions need.

Add history and normal range to each.

Add the event markers.

Delete the rest, and see whether anyone asks. They usually do not.

The summary

A number without its history cannot be interpreted, and adding the range is the highest-return change available.

Five to seven metrics, each supporting a decision someone actually makes.

Mark the events — campaigns, deployments, banner changes — because that answers most of the questions the dashboard generates.

Build alerts for the things that matter, because a dashboard only works when someone looks and an alert works when they do not. For a decision-focused public-sector approach, see GOV.UK guidance on using data to improve services.