Segmentation that reflects how people actually buy
"Engaged in the last 30 days" is not a segment, it is an activity filter. Useful segmentation describes buying behaviour, and it should change what you send.
Most stores segment on engagement because engagement data is the easiest thing to reach. The result is segments that do not differ in any way that should change your message. Real segmentation comes from purchase behaviour — what someone buys, how often, at what price sensitivity, and how they entered — and it is only worth building if the messages actually diverge afterwards.
Attributes that change the message
A segment is only useful if you would write something different for it. Analyst builds segments from behavioural attributes that pass that test, and drops the ones that do not.
Category affinity — what someone buys, not just that they bought
Replenishment interval, which determines when to speak at all
Discount sensitivity, separating full-price buyers from promotion-only buyers
First-order value and acquisition source, which predict long-term value
Return and support history, which should suppress some sends entirely
Segments have to earn their existence
Ten overlapping segments produce conflicting sends and a maintenance burden. The agent consolidates segments that do not behave differently, flags overlaps that cause double-sending, and retires segments that no longer match how the catalogue sells.
From segment to different message
The point of the exercise is divergent content. Retention writes materially different sequences per segment — different objections, different proof, different offer logic — rather than swapping a first name into one template.
How it works
Build from purchase data
Analyst derives behavioural segments from order history rather than engagement windows.
Prune and consolidate
Segments that do not behave differently are merged; overlaps that cause double-sending are resolved.
Write differently for each
Retention produces genuinely divergent sequences per segment and measures them separately.
What you get
Segments defined by buying behaviour rather than by activity windows
Overlap resolved, so subscribers stop receiving conflicting sends
Genuinely different messaging per segment, not a merge tag
Segment performance measured separately so weak ones can be retired
Handled by these agents
Works with
Frequently asked questions
How many segments should I have?
As many as you can write genuinely different content for, and no more. For most stores that is a handful. Segment count is a cost, not an achievement.
Is RFM segmentation still useful?
As a starting frame, yes. As a finished model, it is coarse — it ignores category affinity and replenishment interval, which usually matter more for what you should actually send.
How often do segments need rebuilding?
Whenever the catalogue or the customer mix shifts materially. The agent re-derives them on a schedule and flags when a segment has stopped describing a coherent group.
Can segmentation reduce how much I send?
Often, and that is frequently the gain. Sending less to the right people usually beats sending more to everyone, especially once deliverability is accounted for.
Keep reading
Win-back campaigns
Re-engage lapsed customers using your real churn window instead of a round number.
Email flow automation
Complete lifecycle coverage, built once and then actually maintained.
Design a loyalty programme
A loyalty programme designed from repeat-purchase economics — not points for their own sake.
Post-purchase upsell
Raise repeat rate and order value in the window after purchase, without souring the experience.
Put the team to work on your store
Connect your storefront, analytics, and email stack, set a goal, and let the agents run the work end to end. Start on the free plan with your own model key.