Dashboards that get used. 6 design patterns from 5 years of practice

Why so many dashboards go unused

The pattern is always the same. A dashboard gets built, the marketing team applauds at kickoff, three weeks later nobody opens it. Industry surveys of BI adoption have shown the same number for years: roughly 60% of dashboards built are no longer opened regularly within the first month, and only about one-third of licensed BI users log in weekly. That's not a tool question, it's a design question.

Usually not the tool. Looker Studio, Power BI, Tableau, Metabase, all four can render great dashboards. What fails is the decision before the dashboard: who reads this? Which decision do they make? Which metric drives that decision?

In 2026 the problem doesn't get smaller, it gets bigger. Power BI Copilot, Tableau Pulse, and Looker's Gemini integration now auto-generate "insights" next to every chart, small text boxes that explain in natural language why a number moved. That doesn't solve the adoption problem; it amplifies it. When a platform pushes ten AI-generated findings per view per day to every stakeholder, cognitive load goes up, not down. That's exactly when visual layout decides which information actually gets perceived.

When the answers are fuzzy you build a dashboard that shows everything, and decides nothing. Six patterns I've found reliable in practice:

1. One question per dashboard

"Marketing performance" is not a dashboard topic. That's a domain. A dashboard should answer one question:

• "Are we hitting Q2 lead targets?"
• "Which campaigns are below ROAS threshold?"
• "How does pipeline value distribute across channels?"

If I can't describe the dashboard in one sentence, it's too broad. Better three focused dashboards than one mega-dashboard with fifteen tabs.

2. KPI hierarchy is visual hierarchy

The most important metric is visually the largest. Sounds obvious. I see it implemented wrong every day.

An executive summary dashboard has a headline number large at the top (revenue, leads, pipeline), below that two or three secondary KPIs (CAC, conversion rate, ROAS) at medium size, then optional drilldowns. If a C-level glances at the dashboard for two seconds, they should see the headline. Not the custom charts at position three.

3. Comparison period belongs NEXT to the number, always

A number without context is an assertion. A number with context is a basis for decisions:

• ❌ Revenue: 1.2M
• ✓ Revenue: 1.2M (+15% vs last year · +3% vs last month)

Comparison period consistent across the whole dashboard: same reference, same position, same colour for up/down. Without this context every stakeholder asks at the meeting, and that's exactly the data-maintenance loop the dashboard was supposed to eliminate.

4. Colour is information, not decoration

Rainbow dashboards look "professional" but read as visual noise. My standard:

• Grey for most values (neutral data)
• Green for good, red for bad, amber for warning
• Brand accent (e.g. indigo) for the headline number, sparingly

If colour means nothing, it shouldn't be there. Bar charts don't need a rainbow palette, one brand colour for all bars is clearer and faster to read.

5. Table when number, chart when pattern

Most common wrong call: forcing a table into a chart because charts "look better". But tables and charts have different jobs:

• Table when the exact number is the information (financial reports, campaign audits, top-N lists)
• Chart when the pattern is the information (trends, distributions, comparisons)

Nobody wants to read off a pie chart whether campaign X did €124,500 or €142,500 in revenue. A sorted table is the right tool for that. Charts answer questions like "are we growing?", tables answer questions like "which campaign exactly?".

6. Annotations explain why

Conversion rate drops 30% in week 12. Three days later half the marketing team has a theory. A week later we're still debating in the meeting whether it was the weather, a bug, or the ad budget.

Answer: annotations. Right on the chart, on the data point: "24 Mar, tracking bug in checkout, fixed 26 Mar". Or: "01 Apr, campaign X paused, budget reallocated". Looker Studio supports it natively, Power BI via bookmarks, Tableau via annotation layers.

What changed in 2026: LLM-based insight layers (Power BI Copilot, Tableau Pulse, Looker AI) now generate first-draft annotations automatically. They correlate outliers with calendar events, campaign starts, deployed tracking releases, and suggest a text. That saves the analytics person about 70% of the typing, but it doesn't replace them: every AI suggestion has to be reviewed before it ships, otherwise wrong causalities end up pinned to a data point forever. The visual pattern stays identical: a small marker icon at the data point. Hover or tap reveals a one- or two-sentence context. What shifts is only the who-writes-it question. From "the analyst types it" to "the analyst reviews an AI draft".

Numbers without context = arguments. Numbers with annotations = decisions. That's the difference between a dashboard that gets debated and one that gets used.

What I don't test, and you shouldn't either

Three things I exclude from dashboard design because they rarely survive use:

• 3D charts. Never. Distort values, harder to read than 2D.
• Three-axis charts. If I need three data series in one chart, it's a hint there should be two charts.
• Custom fonts in dashboards. Power BI and Looker render system fonts faster, more legibly, more consistently. "Brand font" on charts is vanity, not value.
• Unreviewed AI insights in the main layout. An LLM "insight" can appear as a suggestion, but it shouldn't land in the executive block without review. One hallucinated causality in a headline number costs more trust than all the time savings from AI will ever return.

If your dashboard gets opened regularly, great. If not, chances are high one of these six patterns is missing. Audit checklist: one question, KPI hierarchy, comparison period, sparse colour, table/chart split, annotations.

Questions about specific dashboard setups: anna [at] datascale.de or via Juri through the Audit Sprint.