DASHBOARDS & GRIDS

Choosing the right chart for business data

Pick the chart by the question, not by taste. Change over time, comparison, share, distribution, relationship - each has one or two right answers.

Business EXPLAINER 5 min read
What is the question?
  1. Change over timeLine
  2. Comparison between itemsBar
  3. Part of a wholeStacked bar
  4. Distribution of valuesBuckets
  5. Intensity across two dimensionsHeatmap
  • KPI one number and its change
  • Table the records
  • Leaderboard a ranking
  • Pivot two dimensions, totals
The question picks the chart. Taste picks the colour.

Start from the question

Every chart answers one kind of question well and the others badly. A line shows how something changed; a bar shows how things compare; a stacked bar shows how parts make a whole; buckets show how values are spread; a heatmap shows where two dimensions meet. Decide which question the chart is for, and the chart type follows. Decide the chart first and the question gets bent to fit it, which is where most bad charts come from.

QuestionChartExample
Change over timeLineOrders per week across the quarter
Comparison between itemsBar, sortedOrders by region: North 395 · West 247 · East 168 · South 142
Part of a wholeStacked bar; pie only for two to four partsEach region's share of this month's orders
Distribution of valuesBuckets (histogram)Products by price band: 0–25, 25–50, 50–100
Intensity across two dimensionsHeatmapOrders by store and hour of day
One number and its changeKPI tileOrders this period against last
The records behind a numberTableThe 14 urgent orders, one row each
RankingLeaderboardStores by revenue, best first
Two dimensions with totalsPivotRegion by month, with a totals column

The pie earns its place only when there are two to four parts and the point is that one of them dominates. Past that, slices become indistinguishable, and a sorted bar says the same thing legibly.

The non-charts

Four of the most useful shapes on a dashboard are not charts at all.

ORDERS952vs last periodURGENT14vs last periodREJECTED36vs last period
KPI tile
ORDERREGIONSTATUS #48213NorthUrgent #48214WestRejected #48215NorthShipped
Table
01020304 NorthWestEastSouth 395247168142
Leaderboard
REGIONJANFEBMARTOTAL North18412289395 West5082115247 East415176168
Pivot
The non-charts. A number, the records, a ranking, two dimensions with totals.

A KPI tile is one number and its change: orders this period, against the last. It answers "how are we doing?" in a glance and belongs at the top. A table is the records behind the number - the 14 urgent orders themselves, one row each - for when someone has to act on them. A leaderboard is a ranking with a bar beside each item, so the order and the gap are both visible: North 395 at the top, South 142 at the bottom. A pivot is two dimensions with totals: regions down the side, months across the top, so North's 395 is visible as 184, 122 and 89, West's 247 as 50, 82 and 115, East's 168 as 41, 51 and 76, and the totals column makes the comparison. The KPI tiles beside them read 952 orders in all, 14 urgent, 36 rejected. Each is a chart for a question that charts handle badly.

Period comparison

The most common question after "how much" is "compared with what", and the answer rarely needs a second chart. On a KPI tile, the change against the previous period sits under the number as a percentage with its direction. On a bar chart, the previous period is a thinner, lighter bar behind each current one. On a line, it is a second, muted line for the prior period on the same axis. The comparison lives inside the chart, where the eye already is, instead of in a second chart the eye has to travel to and match up.

Common mistakes

Wrong · a pie with twelve slices
Right · a bar, sorted
Wrong · two axes, one chart
Right · two charts, one scale each
Twelve slices or four bars. Two axes or two charts.

A pie with twelve slices. Nobody can compare two slices that differ by a few degrees. Sort the values and draw bars.

Dual axes. Two measures on two scales in one chart invite the reader to see a relationship that the scales invented. Two charts, each with its own axis, stacked, show the same two measures honestly.

Truncated bar axes. A bar chart that starts at 300 instead of zero makes 395 look three times 302. Bars encode length; the axis starts at zero.

Six series on one line chart. Past three or four lines, the chart is a tangle. Split by series, or show the one that matters and grey the rest.

Colour with no meaning. A different colour per bar, when the bars are all the same measure, tells the reader to look for a difference that is not there.

Colour

One accent for the thing that matters: the current period, the region in question, the series the sentence is about. Everything else in a neutral grey or a lighter tint of the same accent. The same meaning gets the same colour on every chart in the dashboard, so that once a reader has learned that the accent means this period, they never have to learn it again. A dashboard with three accents is a dashboard with three things that matter at once, which is usually a sign that it has not decided what it is for.

How an assistant chooses

An AI assistant answering a question about business data applies the same rules, automatically, every time. A question about change over time comes back as a line. A question that compares regions comes back as a bar or a leaderboard. A question for one number comes back as a KPI tile with its change. A question for two dimensions comes back as a pivot with totals, and a question for "show me the orders" comes back as a table. The rule is the question, which is why the answers arrive as KPI, pivot, leaderboard, chart or table and not as whichever chart was configured last. How can AI answer questions about business data? follows an answer through that choice.

See it on real data.

The demo instance runs dashboards, data grids and the AI Assistant on real business data. No sign-up.