Build The Dashboard Around The Decision

The short version

Most dashboards are built from the metrics that happened to be available, then reviewed by people who cannot say what decision the page supports. Start from the decision instead, choose the insight family that answers it, and split by one dimension at a time.

Choose the family before the chart

Picking the right kind of insight matters more than picking the right chart type. The families answer different questions:

Family	Answers
KPI number, trend, table	What is the figure, how is it moving, how do several compare
Trend and breakdown charts	How a metric moves, and how it splits across a dimension
Funnels	Where drop-off happens, in delivery or in events
Cohort views	Retention, LTV, and ROAS by cohort over time
Channel views	Scorecards and install trends across platforms
Comparison views	Period over period, quadrants, rankings
Diagnostics	Anomaly scans and campaign drilldowns

A line chart of a number that should have been a funnel is not a styling problem. It is answering a question nobody asked.

One breakdown at a time

Most insights can be split by platform, campaign, ad group, ad, placement, device, demographic, targeting, or traffic source.

The temptation is to split by several at once, on the theory that more detail is more insight. What actually happens is that each cell gets smaller until the numbers stop being reliable, and you reach a confident conclusion from a sample of eleven.

Split by one dimension. If it does not explain the movement, try a different one rather than adding it to the first.

The disagreement is the finding

When an MMP is connected, cohort retention, LTV and ROAS curves, and SKAN summaries become available. These use the MMP as the source rather than the ad platform.

They will disagree with platform-reported numbers. That is expected, it is not a bug, and it is frequently the most useful thing on the page. Platforms report on their own attribution; an MMP reports across them with different windows and rules. The gap tells you how much of your reported performance is claimed rather than incremental.

Teams that treat this disagreement as something to reconcile away lose the signal. Teams that track the size of the gap over time learn something about their measurement that neither source shows alone.

Four habits that make dashboards get read

Start from the decision. If you cannot name what the page is for, you are building a report nobody will act on.
Headline first, diagnostics below. The top number should answer "are we fine." Everything under it should answer "why not."
One dashboard per audience. A weekly review page and a diagnosis page are different documents. Merging them produces one page that serves neither.
Fewer insights, read carefully. A dense page nobody reads is worth less than four charts someone actually looks at.