Segment customers and present the findings

Averages hide differences. A market where half the customers want the cheapest option and half want premium service has an “average customer” who does not exist. Segmentation groups customers with similar needs or behaviour, so that each group can be targeted with the right product, message and price.

Approaches to segmentation

• A priori segmentation uses known variables such as age, region or channel. Simple, but people in the same demographic group often want different things.
• Cluster analysis (k-means, hierarchical) groups respondents by distance on a set of variables. The cluster analysis tool on this site does this.
• Latent class and latent profile analysis (LCA, LPA) assume that the population is a mixture of unobserved segments and estimate the segments with a statistical model. They give formal criteria for the number of segments and each respondent’s probability of belonging to each segment. LCA is used for categorical indicators, LPA for continuous ones.

How many segments?

The tool estimates models with one to five classes and compares them:

The BIC (Bayesian information criterion) balances fit and complexity; lower is better, and it is lowest for three classes. Entropy above .80 means respondents are clearly assigned to classes (.981 here). The four- and five-class models add tiny classes (4.4% and 0.6%) that are hard to target and probably not real. The best log-likelihood of the three-class model was replicated in 15 of 20 random starts, so the solution is stable. Finally, a segmentation must make substantive sense: can you describe and act on each segment?

Naming the segments

Class 1 hunts for deals in physical stores, with little brand loyalty but low price sensitivity: perhaps in-store promotion browsers, reached by in-store displays and promotions. Class 2 is price-sensitive yet loyal and not deal-driven: loyal value seekers, kept with consistent everyday prices and loyalty rewards. Class 3 buys mostly online, is very loyal and loves deals: digital loyal deal hunters, reached by app-based personalised offers. Good segment names are short, memorable and describe what makes the segment different. Then profile the segments on other variables (demographics, NPS, spend) to make them recognisable, and download the class membership file for targeting.

Bringing the course together

A complete marketing research project combines several of the methods in this course. For the app redesign from Lesson 1, the findings would be: NPS is lowest for app customers (−5.0 versus 20.4 in store); price value and customer service drive satisfaction but perform below average; order tracking and easy returns are must-have features, while live chat and AR try-on would delight; strong security and low fees top the MaxDiff ranking; and three segments need different approaches.

Presenting to managers

• Lead with the answer. Start with the recommendation and the two or three findings that support it, then the detail.
• One message per slide or chart, written as a full-sentence headline: “Reliability is the largest service gap”, not “SERVQUAL results”.
• Show uncertainty honestly. Margins of error, confidence intervals and whether differences are significant.
• Translate statistics into business terms. Willingness to pay in currency, reach in customers, importance in percentage of impact.
• End with actions: what to do, for whom, by when, and how to measure success.
• Describe the method briefly in an appendix: sample, fieldwork dates, questionnaire, analyses.

Congratulations on completing the course. The final assessment covers all nine lessons; pass it to get your certificate.