Start with the question and the data you have, then choose the chart. Example field names are illustrative; use equivalent fields from your analysis model. To place a chart and fill its fields, see Adding components.

Choose by question and data

Question	Data you have	Use	Choose the alternative when
What is the current result?	One measure, optionally reference measures	Measure card	Readers also need the trend: a KPI trend card with a period field.
Are we meeting a target?	A measure and a target (measure or fixed number) per category	Bullet	One value on a scale with ticks: Gauge. One completion ratio: Ring progress.
Which category is largest?	A category dimension and a measure	Clustered column	Long names or many categories: Clustered bar.
What contributes to the total?	A category, and an additive measure split by a Legend field (or several part measures)	Stacked column and bar	One series must be compared across categories: a clustered column or bar.
How does the mix differ?	A category and additive, non-negative parts	100% stacked column and bar	A few slices of one whole: Pie. One whole per group: Grouped donuts.
How does it change over time?	An ordered date or period field and a measure	Line	Volume matters, or the series are parts of a total or shares of it: Area charts (area, stacked area, 100% stacked area).
How do measures with different units compare?	A category and two measures, such as sales and margin rate	Combo	Every series has the same unit and role: a line or clustered column.
Do two measures move together?	An observation dimension, such as Store, and two measures	Scatter	—
What happens on each day?	A day-level date field and a measure	Calendar chart	Precise change between dates matters: Line.
How are values distributed?	A row-level numeric field (a physical column)	Histogram	A sample dimension and a measure, compared across groups: Box plot. A low and a high measure per category: Range column.
Where are the patterns across two dimensions?	Two dimensions and a measure	Heat matrix	Exact values matter more than the pattern: a Pivot table.
How do profiles compare across criteria?	3–8 criteria on one comparable scale, and a measure	Radar	More criteria, many entities or mixed units: a clustered bar or a table.
How is a hierarchy divided?	Parent and child dimensions and an additive measure	Treemap	The sequence of levels matters: Sunburst. Readers choose the next split: Decomposition tree.
How many pass each stage?	Ordered stages (members or one measure per stage) and a count	Funnel	The categories are not process stages: a clustered bar.
Where does a quantity flow?	A source, a target and a non-negative weight	Sankey	The relationship is only parent and child: Sunburst.
How does a balance build up?	Ordered steps and signed changes	Waterfall	You have ending balances, not changes: Line.
Where does it happen, by region?	Region names or codes	Filled map	Markers sized by value: Marker map. Both use the built-in region maps and need no online map service.
Where does it happen, by address or coordinates?	Place names, or longitude and latitude	GIS marker map on an OpenStreetMap or Google basemap	Point density: GIS heat map. A Mapbox or Amap basemap: Mapbox/Amap. All three need an online map service, see GIS Map Settings.
Where is it on a floor plan or picture?	X/Y positions on your own image	Image map	—
Which exact values do readers need?	Any fields	Table	Cross-tabulated totals: Pivot table. A hierarchy: Tree table or Parent-child table.
What context should the reader see?	A dimension, words with weights, or image references	Dimension field	Category weight at a glance: Word cloud. A picture per item: Dynamic image.

Limits that often decide the choice:

• Many categories or close values: use a bar, not a pie. A pie reads well with a few slices.
• Criteria for a radar: about 3–8, all on one scale with the same desirable direction.
• One value: use a card, not a chart with a single column.
• An amount and a rate: a single value axis shows plain numbers and squeezes the rate near zero; use a Combo with two axes.
• Distribution charts need the observation grain: the Scatter Marker, the Box plot Sample and the Histogram's row-level Value decide what one point is.

In the palette, charts are grouped as Tables, Cards & KPI, Column & bar, Line & area, Proportion, Distribution & correlation, Flow & breakdown, Maps and Other; hover a tile to see what it is for.

Resolve common data problems

Check a new chart, and any chart that looks wrong, against a table at the same grain and with the same filters: save, open Preview, hover a few categories or points and compare the values. Only then configure Style.

Result	Check first
The value is different from a table	Same measure definition, aggregation grain, filters and included categories?
A total looks too large	Overlapping measures (a total stacked with its own parts), duplicated categories, or a measure aggregated at the wrong grain.
A rate looks wrong	Does the source store 0.42 or 42? Is the measure formatted as a percentage? Is it a ratio of totals or an average of row ratios?
A category is missing	Data filters, row limits, missing values and available display space.
A gap, blank cell or missing period	Absent data is not zero. Check the filters and whether the source returns 0 for that period.
A sharp change between periods	Whether categories appeared or disappeared under the current filters, and whether the last period is complete.
A chart does not respond to a filter	Exact model/field binding and filter subscriptions.
A tiny difference looks dramatic	Axis bounds, percentage normalization and color normalization.
Fields or series disappear after a chart-type change	Chart switch moves fields to the new chart's slots where they fit and can discard the rest. Recheck the data slots, axis units and style settings.
A chart is empty	Required slots, a valid model and whether any data matches the filters; the message tells you which, see Empty data and errors. Style cannot supply missing data.
Labels show K or M, or too few decimals	Display units and Decimal places, see Display units and decimal places.

A pattern between two measures shows association, not cause. See component filters for chart-specific scope. Keep units, comparison periods and the meaning of color visible in the report itself.