Customer segmentation: cutting the list in a way that changes what you do

Customer segmentation divides your customers into groups that behave differently enough to justify treating them differently. The last clause is the whole test. A segmentation that produces interesting groups nobody acts on has cost analysis time and changed nothing.

Small businesses usually over-engineer this. Four or five segments, defined by something observable, is enough to change targeting, pricing and messaging. Twenty micro-segments defined by a clustering algorithm produce a chart that gets shown once.

Ways to cut the list, roughly by usefulness

• By behaviour: how often they buy, what they buy, whether they have stopped. This is usually the most actionable and requires no data you do not already hold.
• By value: what they are worth, using the margin-based approach in customer lifetime value. The top group often deserves treatment nobody has thought to give them.
• By lifecycle stage: new, established, lapsing, lapsed. Simple, and it maps directly onto what you should send each group.
• By need or use case, which is frequently more predictive than firmographics and harder to collect.
• By firmographics — size, sector, geography — which is easy to collect and the least predictive of the five. It is where most segmentation stops, which is why most segmentation does nothing.

Making it change something

1. Start from a decision you would make differently. Who gets a call rather than an email, which group receives the price rise, where the retention effort goes.
2. Cut the list by something you can actually observe for every customer. A segment you cannot assign new customers to is a historical curiosity.
3. Keep the number of segments small enough that each gets genuinely different treatment. If two segments receive the same thing, they are one segment.
4. Write down what each segment gets. Not the description — the treatment. This is the step that separates segmentation from a chart.
5. Measure per segment afterwards. Otherwise you cannot tell whether the differentiated treatment did anything.
6. Review the boundaries yearly; customers move between segments, which is itself a signal worth watching.

A simple recency-frequency-value split does most of what elaborate segmentation promises. Sort customers by how recently they bought, how often, and how much they are worth; the groups fall out immediately and they map onto obvious actions — win back the recently lapsed, protect the frequent and valuable, and stop spending equally on everybody. It takes an afternoon in a spreadsheet.

Where it usually goes wrong

• Segmenting by data you collected because it was easy rather than because it predicts anything.
• Building segments that marketing understands and sales ignores, or the reverse. If both do not use them, they do not exist.
• Treating segments as permanent. Customers move, and the movement is the interesting part.
• Segmenting a list too small to matter. Below a few hundred customers, knowing them individually beats grouping them.

Where it lives

Ettex CRM holds the customer records and purchase history the cuts are made from, Ettex Sheets is where the arithmetic belongs, and Ettex Mail is where differentiated treatment usually shows up first — with the permission question covered separately in marketing database, because a segment is not a licence to contact anybody.

To be clear: there is no clustering engine, no predictive scoring and no automatic segment assignment. For a few hundred to a few thousand customers, a recency-frequency-value sort in a spreadsheet outperforms a model built on too little data, and you can explain it to the person who has to act on it.