Skip to content
Saiku Cloud

Charts and visualisations

Click Chart under the shelves and the same result renders as a picture. The numbers don’t change — only the way you read them. That choice matters more than it gets credit for: a treemap and a pie chart answer genuinely different questions about the same data.

The palette

Sixteen types, grouped by what they’re for.

Bars

Bar is the default and the most generally useful. One bar per category, height proportional to the measure. Reach for it to compare across categories — sales by region, headcount by department, error rate by service. When in doubt, this.

Stacked bar puts a second dimension inside each bar, so you see both the total and its composition. Good for “revenue by quarter, split by product line”. Less good when you actually want to compare the segments to each other — the ones not sitting on the baseline are hard to judge.

Waterfall starts each bar where the last one ended, so a sequence of gains and losses shows how a starting figure became an ending one. The classic gross-to-net revenue bridge.

Lines

Line is for trends. Put a date level on an axis, measures on the other, and the slope between points carries meaning.

That last part is the rule for using it well: a line chart claims the space between two points means something. Over time it does. Across regions it doesn’t — use bars.

Area fills in underneath, shifting the emphasis to accumulated volume. Stacked line and stacked area add series on top of one another for a running total.

Proportional

Pie shows parts of a whole. Honest for five slices or fewer; past that people can’t compare the wedges and a bar chart is kinder. Donut is a pie with a hole, which buys you room for a figure in the middle.

Sunburst is a pie that keeps going: the inner ring is the top level of a hierarchy and each outer ring the next one down. Use it when you want the overall shape and the breakdown in one picture — revenue by region → country → city.

Treemap packs rectangles instead, sized by measure and nested by hierarchy. It stays readable at hundreds of leaves, which is exactly where pie and sunburst fall apart. The right answer to “what dominates?” across a thousand customers or a whole SKU catalogue.

Matrix

Heatmap — two dimensions on the axes, one measure driving cell colour. Unbeatable for spotting patterns in two-dimensional data: sales by hour-of-day by day-of-week shows things no other chart makes obvious.

Radar plots several measures around a circle, one axis each. Useful for comparing a handful of entities across the same small set of metrics — a product against its competitors on five attributes.

Points

Scatter puts one point per row at the intersection of two measures. This is the correlation chart: does higher spend go with higher revenue?

Bubble adds a third measure as point size, so you can carry “how big is this customer” alongside the other two.

Geo

Map draws a world choropleth, colouring countries by measure. The place names come from your row hierarchy, and the first measure drives the colour.

One thing to know before you reach for it: it’s country-level only. Bind sub-national rows — states, counties, postcodes — and you get a blank map rather than an error.

Picking one

A rough decision tree:

  • Comparing categories? Bar. Stacked bar if you also want the composition.
  • Trend over time? Line, or area if the cumulative volume is the point.
  • Parts of a whole? Pie for five or fewer, sunburst if there’s a hierarchy, treemap if there are many leaves.
  • Two-dimensional pattern? Heatmap.
  • Relationship between two measures? Scatter, or bubble for three.
  • A sequence of additions and subtractions? Waterfall.
  • Geography, by country? Map.

Switching is one click, so trying three on the same data costs nothing. Start at bar and move only when you have a reason.

The chart editor

Defaults are reasonable, but the Chart editor is where a chart becomes something you’d put in front of a board:

Labels. A chart title and x/y axis labels. Legend on or off, and which side it sits on.

Y axes. Auto dual axis moves the smaller series to a right-hand axis when two measures’ scales differ by more than about 100× — so a percentage and a currency total stay readable together. You can also pin any series to left or right yourself.

Combo series. Give an individual series its own type — revenue as bars, growth rate as a line, on one chart.

Colours. Palettes (Theme default, Vibrant, Cool, Warm, Earth), per-series overrides, and colour-blind-safe mode, which deliberately overrides your picks for accessibility.

Trend lines. Linear, moving average, or weighted moving average, with a configurable period. They render on line, stacked line and area charts, off the first measure column.

Sorting and Top N. Order categories as queried, ascending or descending by the first measure — and cap the chart to the top N so a long tail doesn’t squash the part you care about.

Reference lines and bands. Draw a target, a threshold, or a shaded acceptable range across the plot. The difference between “here are the numbers” and “here’s how we’re doing against plan”.

Conditional formatting. Colour by value — greater than, less than, between — with a fallback colour for everything else.

Hierarchies. When ROWS has a parent level above a child (Year above Quarter), the chart plots only the leaf rows by default and prefixes them with the parent for context. Turn hide rollup rows off if you want every level drawn as its own bar.

Editor settings save with the workbook, so the look persists.

The compact views

Next to Grid and Chart, More offers three small renderings:

  • Stats — the numeric summary of the result.
  • Sparkline and Sparkbar — tiny inline charts, designed to sit beside numbers rather than stand alone. These are what you want in a dashboard tile where the trend matters but the axis doesn’t.

Exporting

Export → PDF saves the current view at print resolution — better than a screenshot for anything going into a report or a deck. XLS and CSV give you the underlying numbers instead.

When a chart makes things worse

  • Too many series. Ten lines on one axis is a plate of spaghetti. Use the editor’s Top N, or split the query.
  • Wildly different magnitudes. One series in millions, another in single digits. Turn on dual axis, or make it two charts.
  • Two data points. There’s nothing to see. The table is more honest, and quicker to read.

Where to go next