In Short: The Dashboard Is the Easy Part
KPI dashboards fail long before anyone opens Power BI: the metrics were chosen in a workshop that listed everything measurable, half have no target, none have an owner, and the Monday meeting glances at the wall of numbers and moves on. The craft is one-third design and two-thirds choosing and governing the numbers - which is good news, because those parts are entirely in your control.
Here is the practice as we run it: choose, present, govern.
Choosing: Few, Decision-Linked, Owned
A KPI earns its place by passing four tests.
- A decision hangs on it: someone changes behaviour when it moves; if nothing changes, it is background, not key
- It has a target or expected range: variance is the signal; a number without context is trivia
- Its cadence matches its review: daily dashboards for monthly-moving metrics manufacture noise
- It has an owner: one name accountable for the number's accuracy and for the response when it turns red
Hold the line on count: five to nine on the primary view. Every addition dilutes the attention every existing number receives, and the pressure to add is exactly how last year's dashboard became this year's wallpaper. Secondary metrics live behind drill-downs, where the interested can find them.
Definitions are chosen here too: what counts as revenue, when a customer is active. Contested definitions get settled in the semantic model, once, or the dashboard inherits every argument.
Presenting: Variance First, Story Second
The reader's question is never "what is the number"; it is "is it where it should be, and is it getting better?" Presentation follows.
- Lead each KPI with value, variance and direction together: actual, versus target, trend arrow or sparkline
- Use semantic colour only for meaning: red is a state, not a brand accent - the design system rules apply doubly here
- Prefer cards with context and compact trend visuals over gauges and gimmicks; the fastest-reading page wins
- Structure as headline then drill: the nine numbers on top, the why pages behind - decomposition and driver visuals do their best work there
- Design for the meeting: if the Monday review runs off this page, the page's order should match the agenda
Governing: Why Dashboards Stay Believed
Every long-lived KPI dashboard we audit survived on governance, not aesthetics. One model feeds it (no parallel extracts drifting), refreshes are monitored so stale data cannot masquerade as fact, definitions are documented where readers can find them, access respects who may see what, and each metric's owner reviews it quarterly: still decision-relevant, still correctly defined, still targeted sensibly. Retire without sentiment - a pruned dashboard is a trusted one.
The 2026 addendum: those same governed definitions are what Copilot quotes when an executive asks Teams how the quarter looks. The dashboard surface may fade; the discipline behind it is now the substrate for every AI answer about your business, which makes it more valuable, not less.
Sources and Further Reading
Frequently asked
Five to nine on the primary view. A KPI is a key indicator; a wall of forty numbers is a report pretending to be a dashboard, and it trains executives to stop looking. Secondary metrics belong on drill-down pages behind the headline.
Four properties: it connects to a decision someone actually makes; it has a target or expected range (a number without context is trivia); it moves at the cadence you review it; and someone owns it - both the number's accuracy and the response when it goes red.
Yes, or at least an expected range or trend direction. The entire point of the presentation is variance: is this number where it should be? Green/amber/red against target, or versus prior period, is what turns a value into a signal.
Almost always governance, not design: definitions drift between reports, a pipeline breaks silently and the numbers go stale, or two dashboards disagree. One governed semantic model, tested refreshes and a named owner per metric prevent all three.
The surface is shifting - executives increasingly meet numbers as Copilot summaries in Teams - but the discipline transfers unchanged: chosen metrics, agreed definitions, targets and owners are exactly what makes AI summaries trustworthy rather than fluent noise.
