Daniel M. Jin Email

Craft 4 min read

The quiet dashboard

Beauty in BI is compression: fewer marks, three type roles, one accent, and a colour test the old dossier failed.

Most dashboards are loud. The good ones are quiet on purpose.

Beauty in business intelligence is not decoration. It is how little a reader must decode before the number lands.

Every mark earns its ink

Edward Tufte gave the idea a ratio: the ink that shows data, divided by all the ink in the graphic1. Push it toward one. Anything that does not show variation in the data costs attention.

In practice that means four cuts:

  • Lighten gridlines until they whisper, or drop them.
  • Label bars directly and delete the legend.
  • Replace bevels, shadows and 3D with flat fills.
  • Right-align numbers so the digits line up.

None of these cuts removes information. Each one removes a step between the eye and the answer.

Three fonts, three jobs

Type is the dashboard's grammar. One generic font for everything flattens the hierarchy.

RoleTypical faceJob
EditorialA serif such as IBM Plex Serif or Georgia, boldThe question and the one-line insight
StructuralA sans such as IBM Plex Sans or InterAxis labels, categories, tooltips, tables
QuantitativeA mono such as IBM Plex Mono or JetBrains MonoKPIs, timestamps and counts like n = 2,395,5292

Use tabular figures for anything that refreshes. Every digit then takes the same width, so a live KPI does not jitter.

Colour is a signal, not paint

Keep most of the surface neutral: greys, slates and white space. The old dossier set 85% neutral as its working rule3. Treat that as a habit, not a measured law.

Then give each saturated colour one job. Blue for focus, amber for watch, red for breach. A colour that means two things means nothing.

The test the old dossier failed

The old dossier paired emerald #047857 for gains with crimson #B91C1C for losses. It said the pair stays distinct in greyscale3.

I measured it. It does not.

WCAG scores the contrast between two colours from 1:1, equal luminance, to 21:1, black on white4. Emerald against crimson scores 1.18:1. Pure green against pure red scores 2.91:15. The "safe" pair is closer in lightness than the crude one.

Now simulate deuteranopia, one form of red-green colour blindness, in the table below6.

PairNormal visionSimulated deuteranopia
Emerald vs crimson#047857 vs #B91C1C, 1.18:1#696559 vs #776A12, 1.07:1
Pure green vs pure red#00FF00 vs #FF0000, 2.91:1#EFD63A vs #A39000, 2.18:1
Deep blue vs crimson#1E40AF vs #B91C1C, 1.35:1#0044AD vs #776A12, 1.58:1

The emerald and the crimson both turn a muddy olive-grey. The blue stays blue.

Each colour still passes on white, at 5.48:1 and 6.47:15, above the 4.5:1 bar for text7. So they work as labels. They fail as the only difference between good news and bad.

About 1 in 12 men and 1 in 200 women have some colour-vision deficiency8. On a large distribution list, assume someone cannot tell your red from your green.

WCAG's rule is plain: colour must not be the only visual means of conveying information9. The fix costs nothing:

  • Add a sign or an arrow: ▲ and ▼, or + and −.
  • Pick a pair that differs in lightness as well as hue.
  • Build status on blue, which stays blue under both protanopia and deuteranopia simulations5.

Blue survives on hue, not lightness: against crimson it scores only 1.35:1, and 1.23:1 under simulated protanopia5. Pair it with a sign or a word too.

What the Viz of the Day editors reward

Tableau Public features one viz on most weekdays. A Tableau team and a rotating group of community Ambassadors pick it from thousands of nominations a month10.

Their stated aims: spark data conversations, show what Tableau can do, and highlight community work10. Decoration is not on the list.

A board dashboard faces the same bar. Can a reader say the takeaway before reaching for the legend?

What this means for your BI team

  • Write a one-page style spec: three type roles, one accent, and a status pair tested under simulation.
  • Encode it once per platform, as a Power BI report theme and a Tableau custom palette, so every author starts from the same defaults.
  • Ban colour-only status: every red or green carries a sign, an arrow or a word.
  • Check contrast in review, the way you check spelling.
  • In a migration, treat the style spec as a deliverable: the two tools store styling in different places, so parity means rebuilding defaults, not pasting hex codes.

Beauty here is not taste. It is less work between a leader and a decision.

Notes

  1. Edward R. Tufte, The Visual Display of Quantitative Information, Graphics Press, 1983; second edition 2001, chapter 4; publisher's page, accessed 2026-10-17. ↩︎
  2. Kansas City 311 requests on 2026-09-24, both systems combined; audit claim 13 in tableau/real-data-audit.md. ↩︎
  3. The old dossier, creation-test/danieljin-site/claude-opus-5.5_max/site/research-beauty.html, section 03, published 2026-09-04. ↩︎
  4. W3C, Understanding SC 1.4.3: Contrast (Minimum), accessed 2026-10-17. ↩︎
  5. Computed on 2026-10-17 by creation-test/danieljin-bi/content/checks/palette_contrast.py; results in palette_contrast.json beside it. ↩︎
  6. Machado, Oliveira and Fernandes, "A Physiologically-based Model for Simulation of Color Vision Deficiency", IEEE TVCG 15(6), 2009, doi:10.1109/TVCG.2009.113. Full-severity matrices, applied in linear sRGB by the script in note 5. ↩︎
  7. SC 1.4.3 asks for 4.5:1 for normal text and 3:1 for large text; see note 4. ↩︎
  8. Colour Blind Awareness, Colour blindness, accessed 2026-10-17. ↩︎
  9. W3C, Understanding SC 1.4.1: Use of Color, accessed 2026-10-17. ↩︎
  10. Tableau, What is Tableau Public's Viz of the Day?, read on 2026-10-17 through a search index, because the page refused a direct fetch. ↩︎