Explore one measure by splitting it across dimensions. Use it to investigate where a result is concentrated, such as which regions, stores or product categories contribute to sales. A split shows a breakdown, not a causal explanation.

Configure the analysis

1. Add Components → Charts → Flow & breakdown → Decomposition tree (a new tree is 600 × 320 px) and choose an Analysis model in Data.
2. Put Net Sales in Analyze.
3. Add Region, Store and Product Category to Explain by. These are the dimensions users can choose when expanding nodes.
4. Set Filters to the population and period to investigate. Add Color, Tooltips, or Icon only when useful for node context.
5. Check the root value against a table with the same measure and filters.

Field group	Holds
Analyze	One measure. The root node shows its total.
Explain by	The dimensions offered when a node is split, in the order listed. Each level queries one of them.
Color, Tooltips, Icon	Optional extra fields for the nodes.
Time axis	Optional date field for a page Date filter that filters by time axis.
Filters	Component filters.

Start with a small, useful set of dimensions. Very high-cardinality fields can create more branches than users can inspect, and Smart split skips a field with too many members (the tree then shows Skipped {field}: more than {limit} members.).

Follow a path

1. Click + on the root node and choose Region.
2. Click a region to select it, then click + on that node and choose Store.
3. Add another available dimension to investigate the selected subset.
4. To choose a different path, click × in a level's header (Remove level). It removes that level and every level after it.

If a dimension is missing from the + menu, check Explain by, the dimensions already used in the path and the current data result.

Every level inherits the selections above it. A Store value under East China is not a store total for all regions. Recheck the path after changing a report filter.

The path you build in the editor is saved with the report and is what readers see first. Readers can build their own path in view mode; it is not saved, and reopening the report shows the saved path again. A tree can have up to 50 levels and 5,000 data points.

Selection and cross-filtering

Action	Result
Click a node	Selects the path to that node and cross-filters the linked components.
Click the selected node again	Deselects it and removes the filter it applied.
Click the root node (tooltip Clear selection)	Clears the selection and every filter the tree applied to other components.

Which components are filtered is set in Actions; see Cross-filtering.

Use Smart split deliberately

With Style → Analysis settings → Smart split on (default), the + menu also offers High value and Low value: the tree evaluates the unused Explain by dimensions, splits by the one whose member scores highest (or lowest) and selects that member. The recommendation is deterministic; no LLM is used. Comparison mode sets how members are scored and appears only while Smart split is on.

Comparison mode	How it compares candidates
Absolute (default)	Compares member values directly. Useful for finding the largest or smallest result.
Relative	Compares a member value with the absolute average of its candidate dimension. Useful for finding a value that stands out within that dimension.

On a tie the field listed first in Explain by wins. In Relative mode a field whose average is 0 is skipped.

A relative recommendation need not have the largest absolute sales. A high/low recommendation also does not mean good/bad performance. Use a manual split when the business question specifies a dimension or a candidate is unavailable under the current filters. When you add a manual level below High value or Low value levels, those levels are fixed to the dimension they picked.

Set a meaningful bar scale

In Style → Data bars → Scale to:

• Level maximum compares peers within a level.
• Parent value compares children with their parent.
• Visible tree maximum uses a common reference across the visible tree.

These choices change bar lengths, not the measure values. For ratios, averages and distinct counts, child values may not add up to their parent; do not read every bar as an additive contribution. Data bars also has Positive bar, Negative bar and Conditional color, which overrides both; see Conditional colors.

Use Tree layout, Nodes, Connectors, Category labels, Values, and Level headers to keep the path readable. Leave enough width for category names; a level with more nodes than fit scrolls inside the component.