6.1 Dashboards

In Brief

Description

A dashboard is a single screen that brings a product’s or business’s most important metrics together so the whole team can read the current state at a glance. Instead of planning around forecasts, you watch what is actually happening now — active users, conversion, revenue, retention — and react when something moves.

A dashboard makes the relationships between parts of a business visible. A $10k investment in a channel may seem like a lot of money until you put it next to last year’s revenue of $250k on the same screen. Dashboards also create focus: choosing one or a handful of metrics for the whole organization tells everyone what matters, and a shared screen everyone checks keeps that focus from drifting.

A dashboard can serve any level — the company as a whole, a department, a role (such as the VP of Marketing’s dashboard), or an individual contributor. It can be anything from a manually updated spreadsheet to a custom system wired into your other business tools for a live view. Either way, the goal is the same: a reader should grasp the state of the business in seconds. The hard part is deciding what to leave off.

Choosing what goes on the dashboard is what most teams get wrong. “Vanity metrics” — total registered users, raw page views, cumulative downloads — climb steadily regardless of whether the product is improving, and they crowd out metrics the team could actually act on. Favor metrics that are actionable (a change tells you what to do), accessible (the whole team can understand them), and auditable (you can trust how they are calculated). Treat the dashboard as a prompt for investigation, not the answer.

How to

Prep AI Prompt

1. Decide what the dashboard is for. Write one sentence naming the decision the dashboard exists to support (e.g., “should we keep investing in this acquisition channel?”). If you can’t name the decision, the dashboard will collect metrics no one acts on.
2. Choose 5–7 candidate metrics that pass the actionability test. A metric earns its place only if a movement changes what the team will do next week. For each candidate, write the action you’d take if it moved 20% up or down; if you can’t name one, cut it.
3. Confirm the data source is ready. Check that the underlying events, metric definitions, and time-zone handling are stable enough to trust before you build on them.
4. Decide audience, cadence, and access policy. Whole-company, departmental, or role-specific? Reviewed in standup, weekly, or monthly? Public or restricted? Each combination changes what belongs on the screen.
5. Pilot one metric before scaling. Wire a single metric end-to-end (data → display → review ritual) and run it for a week. You’ll surface the data-quality and visualization gotchas before they’re embedded across seven panels.

Analysis AI Prompt

1. Read the layout before the numbers. Color, shape (a traffic-light red/amber/green), and status icons let you interpret a panel at a glance, and the size of each panel should match how important the metric is — the most important numbers get the most space.

2. Strip the chart down to the data. Remove decoration that carries no information: 3D effects, drop shadows, and redundant gridlines. Every mark on the chart should encode a number, not dress it up.

3. Check that the picture matches the data. The size of an effect on the chart should match its size in the underlying numbers. A truncated axis that makes a 2% change look like a doubling is misleading; use zero-based axes (start the scale at zero) for value comparisons.

4. Use a compact format for goal-tracking panels. A bullet graph — a single horizontal strip showing the current value, the target, and a qualitative range (poor / fair / good) — packs a metric and its goal into one line and is a good default for the panels you check most.

5. Investigate movement; don’t just report it. A metric that crosses its action threshold is the start of analysis, not the conclusion. Before acting, name the most plausible causes (including data-quality ones) and run one check that could disprove your leading explanation.