A logistics client emailed us a screenshot of their KPI dashboard last spring. Eleven categories on a stacked bar chart, every color hand-picked from their brand guide. Two of the bars were indistinguishable in print. Three more collapsed into the same muddy teal when their warehouse manager — who is red-green colorblind — looked at it on his monitor. The designer who built it had ten years of experience and a beautiful portfolio of landing pages. None of that experience had prepared him for the problem on screen.
That's the gap we keep running into. Color training in our industry is overwhelmingly marketing-shaped. Pick a hero color, build a tasteful palette around it, ship the page. Those instincts produce charts that lie.
The rules change when color carries meaning
On a landing page, color is mood. On a dashboard, color is data. A user reading a bar chart is making a quantitative judgment with their eyes — "this category is larger than that one" — and if your palette steps through hues in ways that aren't perceptually equal, you've put your thumb on the scale. The viewer thinks they're reading the chart. They're reading your unconscious aesthetic preferences.
This is the case for perceptually uniform colormaps. Viridis, Cividis, Magma, Inferno — they were designed so that equal steps in data map to equal steps in perceived brightness. They look less pretty than a brand-flavored gradient. They tell the truth.
What we use as defaults now
For sequential data — anything ordered, like time or revenue — we reach for Viridis or Cividis first. Viridis covers most cases. Cividis is the one we use when accessibility for deuteranopia and protanopia is a hard requirement, which on dashboards is just "always." Both render to grayscale without losing rank, which matters more than people expect: stakeholders print dashboards, drop them into slide decks, screenshot them into Slack.
For diverging data — things with a meaningful midpoint, like profit and loss against zero — ColorBrewer's RdBu and BrBG still hold up. They were designed by Cynthia Brewer in the early 2000s and the cartography community has been testing them on real readers for two decades. We trust them more than anything we'd invent.
For categorical data, things get harder.
Eight categories is the real threshold
Up to about seven distinguishable hues, you can pick by hand and survive. Beyond that, neighboring categories on the legend start to alias — two greens that look different on your monitor merge on a client's laptop, two oranges fail under the conference-room projector. We've watched it happen too many times to keep guessing.
Our current rule: if a categorical chart needs more than seven series, we either pre-aggregate ("Other" buckets work harder than designers admit) or we shift the encoding. Sometimes the answer isn't a different palette. It's a small-multiples grid instead of one chart with twelve overlapping lines.
When we do need eight to twelve categories, we use Tol's qualitative palettes or Okabe-Ito. Both were built for colorblind safety and tested across deuteranopia, protanopia, and tritanopia. Okabe-Ito tops out at eight categories, which is honest of it — beyond eight, you should not be relying on color alone, and Okabe-Ito quietly refuses to pretend otherwise.
Brand color almost never survives contact with data
This is the conversation we have with clients most often, and it goes poorly maybe one time in five. Their brand palette is teal and coral. Their dashboard wants seven categorical series. We can't get there without inventing colors that aren't in the brand guide, and if we do invent them, the brand designer will not love what comes back.
Our move now: brand colors anchor the chrome of the dashboard — headers, accents, button states. The chart palette is its own thing, chosen for the data. We tell clients this on day one. The ones who push back usually come around when we show them a side-by-side: their brand palette stretched across nine bars, and Tol's bright palette doing the same job legibly.
Test against the worst monitor in the building
Last practical note. Designers tend to work on calibrated, high-gamut displays. Their clients do not. Before any dashboard ships, we screenshot every chart and view it on the cheapest monitor we own — a 2018 Dell from the back office — and on a phone in direct sunlight. If categories blur in either, the palette fails. This catches more problems than any colorblind simulator does, because it captures something the simulators miss: the real-world degradation between "your designer's eyes" and "the actual room where someone makes a decision."
The dashboard our logistics client emailed got rebuilt with Tol's high-contrast bright palette, a small-multiples layout for two of the breakdowns, and an aggressive "Other" bucket. Their warehouse manager could read it for the first time in two years.