Segmentation
On this page
Definition
Segmentation is the division of users into groups that are similar in behaviour or value within a group and different between groups. Segments are rule-based (RFM, metric thresholds) or discovered algorithmically (clustering). The goal is the same: different groups need a different product and a different message.
How to compute
First choose the axes — behaviour, value, channel, lifecycle. Rule-based segments are set by metric thresholds. For clustering, normalise features and feed them to k-means, hierarchical clustering, or a mixture model; pick the number of clusters by internal metrics and by interpretability.
Pitfalls
A segment with no action delivers nothing — start from a hypothesis and an intervention, then split. Clusters overfit noise and fall apart on a rerun. Segment composition drifts over time, so recompute regularly. Mixing behavioural and value axes in one split yields uninterpretable groups.