Grouping and simplification
Real data has long lists of small segments. Grouping keeps the analysis readable and its explanations robust.
Clustering
Section titled “Clustering”The Clustering panel groups segments before the analysis.
Discrete dimensions (country, channel, product): small segments are grouped into an Other segment.
| Setting | Effect |
|---|---|
| KPI | The metric the shares are measured on. |
| Method | Smallest segments: every segment below the threshold goes to Other, labelled Other (< X% of KPI). Long tail: the smallest segments go to Other until together they reach the threshold, labelled Other (last X% tail of KPI). |
| Threshold | The share used by the method, in %. |
Continuous dimensions (numbers): Buckets is the number of ranges a
numeric dimension is cut into. Its values are ranked and cut into contiguous
ranges such as [10;100[, each carrying about the same share of the KPI.
Dates and boolean dimensions are not grouped.
In the web app, the clustering settings can be pinned to the story, with the Clustering panel or with the dimension hierarchy.
Out of scope segments
Section titled “Out of scope segments”Some segments exist on one side of the comparison only: a country opened, a store closed, a product launched or discontinued. Left in the analysis, their whole weight lands in volume, then in mix and performance effects that mean nothing.
In Compare, declare the Eligible dimensions of this detection. Every combination of their values present on one side only is taken out of the steps and shown as one Out of scope block right after Start, split into new and discontinued segments. The steps are then read on the comparable scope, and the waterfall still adds up:
In-scope gap + Out of scope = Total gap
Leave the eligible dimensions empty to turn the detection off (default). You can rename the block, for example Perimeter effect; leave the label empty to keep Out of scope.