Customer segments that describe behaviour, not demographics

You have thousands of customers and treat them all identically, because working out who is who means exporting orders and building a spreadsheet nobody maintains.

The most useful thing you know about a customer is not who they are. It is what they have done: how much they have spent, how often, how recently, what they bought and what they sent back.

That data is already in your orders. The obstacle is that answering “who has spent over £500 and not bought since March” is a query, and a query is a thing most merchants cannot run.

Segments as saved questions

A segment is a definition rather than a list. Define it once and membership updates itself as people’s behaviour changes — someone becomes a lapsed customer by lapsing, not by anybody moving them.

That distinction is what stops it decaying. An exported list is accurate on the day it was made, and a segment is accurate whenever you look.

The four that earn their keep

Most catalogues get nearly all the value from a handful:

Best customers — high spend, recent, frequent. Worth telling first about anything.
Lapsed — used to buy, has not lately. Already converted once, so cheapest to win back.
One-and-done — bought once, never again. The largest group in almost every shop, and the biggest single opportunity.
High returners — buy plenty, send most of it back. Look profitable on revenue and frequently are not.

That last one is uncomfortable and worth looking at anyway. A customer with a fifty per cent return rate on clothing is costing you postage, handling and stock that spends its life in transit.

Recency is the signal

If you use one dimension, use recency. Somebody who bought last week is far more likely to buy again than somebody who spent twice as much two years ago, and lifetime-value tables that ignore recency will confidently point you at people who have already gone.

What segments are for

A segment that only ever gets looked at is a report. The value is in acting: a win-back to the lapsed, early access for the best, a genuine reason to return for the one-and-done, and an honest conversation with yourself about the returners.

Segments feed the win-back enrolment directly, which is the point — the query and the action are the same system rather than an export and an upload.

Be careful about what you infer

Behavioural data is factual: they bought this, on that date, for that amount. Inferences drawn from it are guesses, and some guesses are intrusive. Merchants have learned this expensively — targeting people on conclusions they never volunteered reads as surveillance, not service.

Segmenting on what someone did with you is fair. Building a picture of their life from it is a different activity, and customers can tell the difference.