What makes a good segment? The five tests

A good segment passes five tests: you can measure it, reach it, it behaves distinctly, you can act on it, and it's worth it. Fail one and the segment is a label. Fail two and it's a liability.

The tests exist because segments fail quietly. A weak segment doesn't produce an error message; it produces a campaign that underperforms for reasons nobody can pin down. Running the tests before you build saves you finding out after you've spent.

Test 1: measurable. Can your data actually identify who's in it?

Every word of the definition must map to a field you hold. Matt's "greenhouse people" passes: bought from the greenhouse category, or browsed it twice, both are fields his platform records. "Premium customers" fails as written: no field records premiumness. It's a hunch, and hunches are fine, but they need a proxy before they're buildable. Rewritten as "average order valueAverage order valueThe average amount spent per order: total revenue divided by number of orders. AOV thresholds are a common way to make value-based segments measurable, such as "customers in the top 20% by average order value".View in glossary in the top 20%", it passes. (Matt's top-20% threshold turns out to be £85; now it's a rule a database can run.)

Test 2: reachable. Can you contact them through a channel you're allowed to use?

A perfectly defined segment with no valid email consent and no ad-platform match is an audience of ghosts. Reachability includes permission, which UK marketers cannot treat as a footnote: a segment you cannot lawfully contact fails this test as surely as one with no contact details at all. When Matt checks, 600 of his 3,800 lapsing customers have no marketing consent on record. They stay in the analysis and out of the campaigns. Path 5 covers the rules; for now, make consent part of every segment definition rather than an afterthought at send time.

Test 3: distinct. Does the group behave differently from everyone else?

If a group converts, spends and returns exactly like everyone else, the segment is a fact, not a difference. Matt checks "weekend browsers" and finds their repeat rate (31%) and average order (£58) are all but identical to the weekday base (30%, £61). The segment is real and useless. Distinctness is what justifies different treatment, and you check it with one comparison: pick a behavioural number and compare the group against the rest. No meaningful gap, no meaningful segment.

Test 4: actionable. Does it imply what you'd do differently?

The definition should carry its own campaign inside it. "Bought a lawnmower 10 to 12 months ago" practically writes the servicing reminder itself. Matt's "local customers" (within 30 minutes of the shop) passes too: click-and-collect offers, event invitations, café vouchers. But notice that "customers aged 25 to 34" would fail: age alone rarely tells you what to say. A segment that's measurable and distinct but implies no action is an insight, and it belongs in a report rather than a campaign tool.

Test 5: worth it. Is the group big enough, and valuable enough, to repay the effort?

Every segment costs something to keep: creative, logic, monitoring, attention. The way to know whether a group repays that upkeep is to run the sum: members × expected conversion × average order. Small can still pass, if the value is there. By contrast, a segment of 40 low-value customers with a niche trait fails not because it's small but because nothing you could earn from it covers its upkeep.