Explain k-means user segments to a marketing team

What's really being asked

The interviewer wants to know if you can take an unsupervised result that has no inherent meaning and give it business meaning that a non-technical team can act on. The algorithm is irrelevant to the audience; the personas and actions are everything.

The full answer

Start by naming each segment with a memorable, descriptive label such as Weekend Browsers, Loyal Power Users, or Price-Sensitive Newcomers, derived from the features that most distinguish each group. For each persona, state the two or three defining behaviors in plain terms, for example high purchase frequency but low basket size. Quantify each segment's size and its share of revenue so marketing can prioritize. Then give a concrete recommended action per segment, such as a win-back email for a churning group or a loyalty perk for high-value users. The visuals should be a labeled scatter or bubble chart with segments color-coded and bubble size showing population or revenue, plus a compact profile table with one row per segment and columns for size, average spend, and key traits.

The mistakes people make

Explaining how k-means iterates, how you chose k via the elbow method, or what inertia means. Presenting raw cluster IDs like cluster 0 through 4 with no human-readable names. Showing the clusters without tying each to a recommended action.

What usually comes next

How did you decide on five segments? How stable are these segments over time? How would marketing measure whether targeting a segment worked?

A concrete example

You present a single slide: five labeled bubbles on a spend-versus-frequency plot. The largest bubble, Casual Dabblers, is forty percent of users but only ten percent of revenue, suggesting a nurture campaign. The smallest, VIP Regulars, is five percent of users but thirty percent of revenue, justifying a concierge program. Marketing leaves with names, sizes, and a next step for each.