Variant grouping should follow the products, control needs and available evidence. A count of forty variants is an operating hypothesis, not a universal Google rule.
The earlier “internal lottery” explanation and 25–35 sweet spot were not supported by platform evidence. Do not split every group just because it crosses a count threshold.
Consider the original hammock example: two groups of one hundred variants versus eight groups of twenty-five. That comparison is a possible account design, but without the exposure, dates and results it cannot establish which structure performed better.
Before changing it, record:
- Which variants represent the same product family and require valid grouping attributes.
- Whether margin, price, availability or demand differs enough to justify separate controls.
- Whether a smaller group would retain enough useful data for the intended decision.
- Which settings and other account changes would confound a before-and-after comparison.
Keep valid product IDs and relationships. Do not create duplicate offers simply to manufacture test cells. Consult the current product specification and maintained segmentation guide for the actual feed setup.
A good result is an interpretable product structure with a recoverable baseline. A lower CPC alone is not enough if conversion quality or contribution deteriorates.
Take the idea further
Review the segmentation guide and its limits