Metrics Before Dashboards

Why deciding what to measure — before building anything visual — is the highest-leverage work in analytics.

Every dashboard problem is secretly a metrics problem.

It’s the most common pattern in analytics work: someone asks for a dashboard. The request sounds concrete and easy to act on — pick a tool, wire up the data, ship some charts. Three sprints later, the dashboard exists, nobody opens it, and the conversation shifts to why adoption is low.

The problem is almost never the charts. It’s that nobody decided what the numbers mean or what decision they’re attached to.

Start with the decision

Before any chart gets built, there’s one question that matters: what will someone do differently depending on what this number says?

• If the answer is “nothing,” you don’t have a metrics problem — you have a decoration problem. Skip it.
• If the answer is “we’d investigate,” the metric is a tripwire, and it should be designed like one: cheap, loud, and hard to misread.
• If the answer is “we’d change the roadmap,” the metric is a compass, and it deserves the full treatment — definition, owner, target, review cadence.

Most “failed” dashboards fail because they were built before anyone answered this question. The charts were fine. The thinking wasn’t finished.

Name the metric precisely

Vague metrics produce vague arguments. “Engagement” can mean almost anything, which means it means nothing. A well-named metric answers four things in one sentence:

1. Definition — the exact formula, in plain language, with the edge cases called out.
2. Grain — is this per user, per day, per session, per account?
3. Window — trailing 7 days? Calendar month? Point-in-time?
4. Owner — the one person who will be asked when the number looks wrong.

That last one is underrated. A metric with an owner survives; a metric with only a SQL file slowly rots.

One number beats ten charts

There’s a strong temptation, once the data is flowing, to show everything. Resist it. Every chart on a dashboard is a claim on the reader’s attention, and attention is the scarce resource.

The dashboards I’ve seen actually used share a shape:

• One hero metric per audience, the number they’d want on a T-shirt.
• Two or three supporting metrics that explain why the hero moved.
• Nothing decorative. If a chart can’t change a decision, it goes.

A checklist that saves sprints

Before building the next dashboard, run through this:

• What decision does this support, and who makes it?
• What would we do if the number went up? Down? Flat?
• Is the metric written down — definition, grain, window, owner?
• What’s the smallest set of charts that answers the question?

If you can get through that list, the dashboard is almost an afterthought. That’s the point: the dashboard is the output of clear measurement design, not a substitute for it.

Decide what to measure first. Then the charts draw themselves.