Choosing a chart

Start from the question you're answering and pick the chart that answers it, with live examples for each kind of question.

Start with the sentence you want a reader to say after looking: "Revenue is up on last year", "West is our largest region". That sentence points to the chart.

When you choose Chart these results in the explorer, Halo suggests a first type from your fields: a scorecard for a measure alone, a line for a date, a column for a category. Treat it as a starting point. Chart types has every option.

A single number

Use a scorecard when one number is the answer. Add a comparison, since a number means little alone: 2.4M is good or bad depending on last year.

A gauge suits a number with a known scale and target, such as a rate that should stay near 95%.

Use gauges sparingly. Several scorecards read faster than several gauges.

A trend

A line shows change over time better than anything else. Pick the coarsest time grain that still shows what matters: months for a two-year story, days for last month.

Up to about five lines stay readable. Past that, filter to the ones that matter, or use one small chart for each.

Use columns when each period stands alone, like a monthly total you'd compare one by one.

A comparison

Bars compare sizes accurately, because eyes judge length well. Sort them largest first unless the groups have an order of their own. Sideways bars leave room for long names.

For two measures over the same groups, a combo shows both. Prefer one shared axis, or two charts, since a second axis can make any two lines seem to move together.

Composition

For shares of one whole, a donut works with a handful of slices. Past six or so, or when slices are close in size, use a bar instead.

For how the mix changes over time, stack areas.

For what moved a total from one figure to another, a waterfall shows each step up and down.

For stages people pass through, a funnel shows the drop-off between each.

Distribution

Averages hide a lot. A histogram shows how a number's values are spread: whether most orders are small with a few large ones, or two groups of customers behave differently. A box plot compares that spread across groups.

A heatmap shows patterns across two groupings at once.

A relationship

A scatter shows whether two measures move together, and picks out the points that don't. Here most segments keep to a similar order size, and one stands apart.

Add a third measure as size with a bubble chart, though small differences in size are hard to read.

Exact values

When people will look up or copy figures, give them a table. A pivot compares across two groupings.

Good habits

• One message per chart. Two ideas usually want two charts.

• Say what it shows. A title like "Revenue by region, largest first" does half the reading.

• Show fewer labels. A few value labels read better than all of them.

• Keep color for meaning. Series colors are set for you, and color rules call out what needs attention.

• Offer the table. Every chart has one, for exact values and screen readers. See Exploring a chart.