Marketing
Market segmentation
Split a market into groups that want different things, size them, and decide which to target.
Segmentation, targeting and positioning start here: split a market by what people want, size each group, and decide which to serve. Then, among the customers you already have, see who is warm and who is cooling.
Segment by needs
Groups that want different things need different offers. Here k-means groups survey respondents who answered alike about what matters to them when they buy; the profile then shows what sets each group apart, and how big each one is in people and in spend.
Rating questions work best. Each is standardised first, so a 1–7 scale and a 1–10 scale count the same.
Silhouette width: near 1, every respondent sits firmly in its group; near 0, groups overlap. A rough guide (Kaufman and Rousseeuw): above 0.5 reasonable, 0.25 to 0.5 weak, below 0.25 no real structure.
100 = average. Blue: above the average respondent; terracotta: below. The small number is the segment’s mean answer. Names come from each segment’s needs indexing 110 or more.
- k-means finds the groups the numbers support, not the ones that exist: change the questions or the number of segments and the groups change. Keep a split only if each group is big enough to serve and different enough to need its own offer.
- Segments are lettered by size, largest first. The names describe the answers and are not personas.
Choose the target
Segmentation pays only when you pick. Score each segment above on its size, how much each person spends, and how well its needs match what you are good at, with weights you set.
- Size and spend are scored against the best segment (100 = the biggest, or the highest spenders). Fit is 50 for a segment whose needs are average, rising when it cares most about your strengths and least about your weaknesses.
- The scorecard ranks the segments in this survey. Check that the winner is reachable through channels you can use, and big enough in the real market, before committing.
Customer value (RFM)
RFM scores every customer from 1 to 5 on recency (days since the last order), frequency (number of orders) and monetary value (total spend), then groups them so each group can get the message that fits.
Defaults to the day after the latest order.
Placed like a sunflower’s seeds, 137.5° apart: the best customers sit at the dark heart, the faintest at the rim. Point at a seed, or pick a segment in the table, to find them.
Rows are the recency score; columns average the frequency and monetary scores. Each cell shows its segment and number of customers.
- Scores are quintiles within this customer base, so a “5” means top fifth here, not an absolute standard. Customers with identical values share a score.
- RFM looks backwards. Pair it with margins and acquisition cost before deciding where to spend.
Sources
- All example data on this page is fictional, generated for the demo: the shopper survey (600 answers from five planted kinds of shopper, with noise) and the online-shop orders.
- k-means with k-means++ starts: Arthur and Vassilvitskii, “k-means++: The advantages of careful seeding”, Proc. SODA (2007). Silhouette width: Rousseeuw, Journal of Computational and Applied Mathematics 20 (1987); the rough guide to reading it: Kaufman and Rousseeuw, Finding Groups in Data (Wiley, 1990).
- Needs-based segmentation and targeting: Kotler and Keller, Marketing Management, chapter on identifying market segments and targets.