Segments that survive contact with a real catalogue

Most segment definitions are written once, against the catalogue as it looked that week, and are never checked again. Then the merchandising team renames a category, a supplier changes, the winter range goes out — and a segment that used to match 4,000 people quietly matches eleven. Nobody notices, because a campaign that sends to eleven people does not look like an error on a dashboard.

The fix is to build segments out of behaviour and time, which are stable, rather than out of catalogue structure, which is not.

The six that hold

1. Bought in the last 90 days. Time-bounded, catalogue-independent, and the single most useful audience you have. Everything else is a variation on it.
2. Bought twice or more, ever. The second purchase is the real conversion event in ecommerce; the gap between one-time and repeat buyers is larger than any demographic split you can construct.
3. Engaged in the last 90 days, by click or order. Not by open — opens have been unreliable since Apple's Mail Privacy Protection. This is your deliverability safety net.
4. Lapsed: bought once, nothing in 180 days. A win-back audience defined by absence, which no catalogue change can break.
5. High value: top 10 % by 12-month revenue. Relative, so it recalculates itself as the business grows. An absolute euro threshold goes stale within a year.
6. Never purchased, subscribed over 60 days ago. The audience your welcome series failed to convert. Worth a different message, and worth sunsetting.

Notice that none of these mention a product, a category or a brand. That is deliberate. Category segments are useful for a specific campaign and should be built for that campaign and thrown away afterwards.

RFM, without the spreadsheet mysticism

Recency, frequency and monetary value is a forty-year-old direct marketing technique that survives because it works and because it needs three fields you already have. Score each customer 1–5 on each dimension, and you get a grid that tells you who to protect and who to reactivate.

The practical shortcut: you do not need all 125 cells. Four groups cover most decisions — champions (recent, frequent, high value), at risk (previously frequent, not recent), new (recent, single purchase), and lapsed. If your platform cannot compute this, sort by last order date and total revenue and draw the lines yourself.

The engagement gate in front of every campaign

Whatever segment a campaign targets, intersect it with “engaged in the last 6–12 months” before sending. This is not a targeting refinement, it is deliverability hygiene: Google enforces a spam complaint threshold of 0.3 % and recommends operating below 0.1 %, and dormant addresses are where complaints and spam traps come from.

• Run the gate at send time, not at list build time.
• Make the window shorter before a large send, not longer.
• Exempt genuinely transactional messages, which are a different legal and technical category.

A sunset policy, written down

Sunsetting is removing people who have not engaged in a long time from marketing sends. It feels like deleting revenue. It is not: those addresses were not producing revenue, and they were actively taxing the ones that do.

1. At 6 months of no engagement, reduce frequency and try a different subject line style.
2. At 9 months, send a short “still want these?” message with a single obvious button.
3. At 12 months, stop marketing sends. Keep the record and the consent state; you are not deleting anyone.
4. Reinstate instantly on any click, order or form submission.

Two maintenance habits

• Alert on size change. If a saved segment's count moves by more than about 30 % between runs, something upstream changed. This catches renamed categories, broken imports and failed syncs.
• Re-read your definitions once a quarter. Ten minutes, all of them, out loud. You will delete two and fix one every single time.