Analytics Dashboard Design Services

Which metrics go on the main view and which in a drill-down

The most common analytics dashboard question we get from founders is where a metric belongs. The main view is for the handful of numbers a user acts on every time they open the product; drill-downs are for the numbers that explain them. Deciding this before any chart is drawn is most of analytics dashboard design.

A working rule: a metric earns the main view when a user would change something this week based on it alone. If their first reaction is to ask why it moved, it is a diagnostic metric and belongs one click away, with the date range and filters carried into the drill-down so the context is never lost.

• Main view: 3 to 5 outcome metrics (revenue, active users, conversion, retention, cost), each with its trend against the previous period and a target where one exists.
• Drill-down: the breakdowns that explain a change, by segment, channel, cohort, plan or account, with the same time range and filters pre-applied.
• Detail: individual events, rows and exports, reachable from any drill-down and searchable, never on the first screen.
• Everything else: if a metric is not acted on and does not explain an outcome, leave it out of the product and keep it in the warehouse.

Analytics dashboard design patterns that hold up

Most analytics dashboards that work share the same handful of patterns. None of them are visual; they are about state, context and timing.

• Overview strip: the main-view metrics in one row, each a link into its own drill-down, so the first screen is a table of contents as much as a report.
• Compare to a period, always: a number without 'vs last 30 days' is a trivia fact. Show the delta, its direction and whether that direction is good.
• Filters that travel: date range, segment and plan filters persist from the overview into every drill-down and back, and are visible as removable chips so users can see what they are looking at.
• Drill-down with a way back: breadcrumbs or a persistent overview rail, so users never lose the question they came with.
• Response times by the clock: Nielsen Norman Group's limits are 0.1 s for a change to feel instant, 1 s before flow of thought breaks and 10 s before attention is lost. Filter changes should land under 1 s; anything slower needs a skeleton state and a progress indicator, not a spinner.
• Honest empty and partial states: 'No data for this range yet', 'Tracking started 12 March' and 'Last synced 4 minutes ago' prevent the most expensive dashboard failure, which is a user trusting a wrong number.

An example from our work: Echo AI's usage metrics

When we redesigned Echo AI, an AI chat and IDE developer tool, the first deliverable was the design system, and the first components were the ones developers actually touch: the editor chrome, the AI usage metrics, the model selection and the conversational surfaces beside them. Each became a named component with defined states, so the metrics read the same way on every screen and later screens agree with earlier ones. The whole product was redesigned on Echo AI's existing codebase in 5 to 6 weeks. The full story is on the Echo AI case study page under Work, and our shortlist of SaaS dashboard design agencies compares studios that do this kind of work.