segmentation.

Where segmentation fits on a long engagement.

The test of a segmentation is whether anyone does anything differently as a result. Clustering will always produce groups, and they will always be describable, but a segmentation that does not change a decision is a presentation rather than an analysis.

Simple rule-based segments frequently outperform clustering in practice, because they are stable, explainable and directly actionable. A cluster whose membership shifts when the algorithm is rerun cannot be the basis of an ongoing campaign, and explaining it to a stakeholder is harder than the value usually justifies.

What an assigned team does with segmentation.

Segments drift. The behaviour that defined a group two years ago describes a different set of people now, and a segmentation that is not revisited becomes a description of history that people continue to act on.

Reviewing it periodically is ongoing analytical work rather than a project deliverable, held within an agreed committed monthly capacity.

What we use segmentation for.

• Segments that change a decision Groups defined so that different treatment actually follows, rather than a description.
• Rules preferred where they suffice Stable, explainable definitions instead of clusters that move on every rerun.
• Definitions revisited as behaviour shifts Segments reviewed, so they describe the current population rather than a past one.

How segmentation capacity is assigned.

Segmentation work is assigned under managed data services, with actionability treated as the acceptance criterion rather than statistical separation.