Colour begins working before a reader has absorbed the title, checked the axis or read a single value. It directs the eye, groups related marks and suggests what deserves attention.

That makes colour part of the reporting language, not a finishing touch. Used well, it helps someone locate the signal quickly and interpret it correctly. Used carelessly, it can make an ordinary result look urgent, imply an order that does not exist or conceal the one exception that matters.

The two charts contain the same values. The first is colourful, but every colour competes for attention and none has an assigned meaning. The second is quieter. Most marks provide context; one exception carries the visual emphasis.

The difference is not taste. It is the amount of work the reader has to do.

Colour is an instruction

Most business reports are scanned before they are read. A finance director looks for the variance that needs explaining. An operations manager checks where service has fallen below target. A board member wants to know whether the pattern is improving, deteriorating or simply noisy.

Colour can help answer those questions in seconds, but only when its role is clear. In analytical work it usually performs one of four jobs:

  • Attention: this result is the one to examine.
  • Grouping: these marks belong to the same series or category.
  • Order: values progress from less to more.
  • Direction: results sit on different sides of a meaningful reference point.

If colour is not doing one of those jobs, it may be decoration. Decoration is not inherently bad, but it should not compete with the data.

Consistency matters too. When the same accent identifies the selected item throughout a report, readers learn the convention and stop decoding it on every page. When colours change meaning between charts, each view becomes a new puzzle.

This does not mean that green must always mean good or red must always mean bad. Those associations vary by context and they are not equally visible to everyone. It means that a report should establish its visual rules, apply them consistently and support important meanings with labels, symbols or position.

Fast can still be wrong

Rapid interpretation is only useful when the interpretation is sound. Colour does not merely make a chart easier to see; it can change what the chart appears to say.

A rainbow scale applied to ordered values creates bright boundaries where the data may change only slightly. A light-to-dark ramp applied to departments suggests that one department ranks above another. A diverging palette without a meaningful midpoint turns an arbitrary value into a visual event. A saturated corporate palette can give every series the prominence normally reserved for an exception.

These are not cosmetic shortcomings. They can alter which result is noticed first, which differences appear substantial and where a reader believes the centre of the story lies.

Match the palette to the question

There is no universally correct data palette. The right choice depends on the relationship encoded by the data.

Sequential: how much?

A sequential palette represents ordered values from low to high. Lightness should progress steadily enough that a larger value reliably looks stronger than a smaller one. This makes sequential colour useful for rates, concentrations, age bands and other magnitudes, particularly on maps and heatmaps.

The common failure is an uneven ramp: several pale steps collapse into one another while the final dark step appears to contain most of the change. That distorts the apparent distribution even when the numeric intervals are equal.

Diverging: on which side of the reference?

A diverging palette has two ordered arms around a centre. It is appropriate when that centre has a defensible meaning: zero profit, a target, the national average or no change.

The centre is not simply the middle of the data range. If revenue variance runs from -2% to +18%, placing zero at the visual centre may require unequal numeric ranges on the two sides. The business meaning should determine the scale, not the convenience of a default setting.

Categorical: which group?

A categorical palette separates things that are different but not inherently ordered: regions, product lines, channels or scenarios. Its colours should be distinct at the size they will actually appear, not just as generous swatches on a design page.

More categories do not automatically justify more colours. Direct labels, small multiples or a highlighted selection with neutral context often work better than asking readers to remember twelve legend entries.

Highlight: what needs attention?

Highlighting is the simplest palette and often the most useful: one accent against neutral context. It works for a selected business unit, the latest period, an outlier or a result below threshold.

The restraint is the point. If five things are highlighted, the reader still has to decide which one matters. A strong accent earns its power from everything around it being quieter.

Rising intensity: where should urgency build?

Some monitoring views have a clear direction and a deliberately uneven distribution of attention. Passing is expected; failure requires action. Giving every step in the palette the same visual intensity can make the normal state as prominent as the exception.

The CDR Ecosystem Health treemap is a useful example. Cell area shows how prevalent each schema node is in product responses, while colour shows how reliably that node complies with the standard. Most nodes are compliant. If the large compliant regions use the strongest green available, the chart spends much of its visual energy confirming that nothing is wrong.

A rising-intensity palette changes that balance. Green still means good and red still means bad, but the compliant steps are pale and close to the background. Yellow begins to interrupt the field, orange carries greater weight and red becomes difficult to overlook. The palette preserves the ordered compliance bands while directing attention towards the smaller number of nodes that may need investigation.

This is an intentionally asymmetric treatment, not a neutral default. It is appropriate when the direction of the measure and the action thresholds are explicit: compliance, risk, service failure or overdue work. It would be misleading for a measure such as election margin, where both sides of the midpoint deserve equal visual weight. Labels and stated thresholds should still reinforce the meaning; colour supplies urgency, not the definition of the problem.

Specialist work sometimes calls for cyclical scales, multi-hue sequential scales or bivariate palettes. The same principle still applies: choose the palette for the structure of the measure, and make that structure clear to the reader.

Why equal numbers do not look equally different

Screens produce colour with red, green and blue light, so RGB is a practical way to store and display it. It is less useful as a way to design analytical scales. Equal numerical changes in RGB do not usually produce equal-looking changes to a human observer.

HSL and HSV make colour easier to describe by separating hue from lightness or value and saturation. They are intuitive controls, but their geometry still does not match perception particularly well. A yellow and a blue with the same HSL lightness can appear markedly different in brightness.

CIELAB was designed around human vision and made perceptual comparison far more useful. OKLab and its cylindrical form, OKLCH, refine that idea for modern display colour. They are not a perfect model of every viewing condition, but they provide a much stronger basis for building even ramps and measuring differences between steps.

OKLCH gives a palette designer three useful controls:

  • Lightness describes how light or dark the colour appears. In many quantitative scales, this does most of the work of communicating order.
  • Chroma describes colour intensity. Raising it can increase emphasis; lowering it can let context recede.
  • Hue describes the colour family. It can separate categories or distinguish the two sides of a diverging scale.

Separating those dimensions makes the design intent explicit. A sequential palette can vary mainly in lightness without wandering unpredictably through hue. A categorical palette can distribute hues while keeping their apparent weight more consistent. A warning colour can become more prominent through controlled chroma rather than simply being made darker.

Build a colour system, not a collection of swatches

A mature reporting environment needs more than a list of brand colours. Brand palettes identify the organisation; analytical palettes explain the data. The two should feel related, but they have different responsibilities.

A practical reporting colour system usually defines:

  • neutral ink, rules and backgrounds;
  • one or two accents for selection and emphasis;
  • sequential scales for ordered measures;
  • diverging scales for approved reference points;
  • categorical slots with a recommended maximum number of series;
  • semantic colours for warnings, failures and success states;
  • rules for labels, contrast and colour-vision differences.

The governance is as important as the hexadecimal values. Teams need to know which palette belongs to which analytical question, when a semantic colour may be used and what must accompany colour to make the meaning unambiguous.

This also protects brand quality. A consistent reporting system looks recognisably like the organisation without forcing every chart to use the most saturated colours in the corporate identity.

Accessibility supports correct interpretation

Accessibility is sometimes treated as a compliance pass after a design is finished. In reporting it is part of whether the report can be understood at all.

Colour-vision accessibility is not one alternative palette. Deutan, protan and tritan perception change different relationships between colours, so a palette that remains clear in one simulation may still collapse in another.

These examples use the same six-category task and canvas. The Standard palette prioritises broad hue separation. The Deutan palette reduces its dependence on red-green differences. The Protan palette also treats dark reds cautiously because reduced red sensitivity can change both hue and apparent brightness. The Tritan palette avoids making blue-yellow contrast do all the work. In every case, lightness differences and labels remain part of the encoding.

Colour should not be the only way important information is conveyed. A failed status can also use a label or icon. A selected line can be thicker as well as differently coloured. A map legend can show explicit ranges. Direct labels can remove the need to distinguish similar series from colour alone.

Contrast also has to be tested in context. A colour that looks clear as a large square may disappear as a thin line, a small point or text on a pale cell. The relevant question is not whether a swatch passes in isolation, but whether the finished mark remains visible and its meaning remains recoverable.

The W3C guidance on use of colour makes the underlying requirement straightforward: colour cannot be the only visual means of communicating information or prompting a response.

Test colour where it will be used

Palette strips are useful, but they are forgiving. They give every colour a large, equal area and place each step in a predictable sequence. Reports rarely do that.

A palette should be checked across the marks it is intended to serve: thin lines, small points, adjacent bars, dense heatmap cells, labels and large filled regions. It should also be checked on the actual report background, at realistic screen sizes and under colour-vision simulation.

That is why we built Spectra as a lab for analytical colour rather than a conventional colour picker. It generates palettes in OKLCH, measures their perceptual behaviour and applies them to the Datashow Chart Index so that problems appear before the palette reaches a live report.

The scorecard checks measures such as lightness progression, separation between steps, colour-vision robustness, canvas contrast and whether colours remain inside the display gamut. Those measures do not replace judgement. They make the trade-offs visible, then the chart previews show whether the palette works in practice.

A short review before publication

Before a report goes to its audience, ask:

  1. What job is colour performing in each chart?
  2. Does the palette match the structure of the data: categorical, sequential, diverging or highlight?
  3. Is a diverging midpoint genuinely meaningful?
  4. Does the most prominent colour correspond to the most important information?
  5. Can the chart still be understood when colour is reduced or perceived differently?
  6. Are text, icons, line weight or position reinforcing critical meanings?
  7. Has the palette been tested in the finished chart, on the finished background, at the finished size?

Colour cannot rescue a weak measure or an unsuitable chart. It can, however, make a sound report much faster to read and much harder to misread. That is enough to treat it as part of the analytical method.

Explore the palettes in this article, change their lightness, chroma and hue, and test them across the full chart library in Spectra.

Further reading