Challenge · Customer segmentation
You already own your segmentation. It is in your orders table.
Three columns you already have, a spreadsheet, and an afternoon. The hard part is not the maths, it is deciding what each group gets.
20 August 2026 · 8 minute read
You can segment your customers usefully with three fields you already have: when they last bought, how often they buy, and how much they spend. Score each customer on each field, group the scores, and eight recognisable segments fall out, from your best customers to the ones who have gone. This takes an afternoon in a spreadsheet and requires no new tools. The value is not the chart. It is that each segment then gets a different message, and the two segments that most need different treatment are your highest-value lapsing customers and your recent first-time buyers, who almost always receive the same email today.
What three columns do you need?
A customer identifier, an order date, and an order value. That is it. That is all. It can come from Shopify or any commerce platform, from your billing counter, from a distributor sales file, or from a CSV your finance team already produces. If you sell offline and hold nothing but a phone number and a date against each sale, you still have two of the three, which is enough to start.
| Field | What it becomes |
|---|---|
| Last order date | Recency: how long since they bought |
| Count of orders | Frequency: how often they buy |
| Sum of order values | Value: what they are worth |
How do you turn three columns into eight groups?
- For each of the three measures, sort your customers and split them into five equal bands. The most recent fifth scores 5 on recency, the next fifth scores 4, and so on.
- Every customer now has three scores between 1 and 5.
- Group them by the patterns. High on everything is your best customers. High frequency and value but low recency is the group slipping away. High recency but low frequency is a new customer who has not come back yet.
- Name the groups in plain language. Not R5F5M5. "Your best customers", "going quiet", "just arrived". You will be showing this to people who did not build it.
The banding matters more than the exact thresholds. Splitting into fifths within your own base is what makes this work for any size of business, because it is relative to you rather than to an industry benchmark that does not fit.
What should each group actually get?
This is the part that gets skipped, and it is the only part that produces money.
| Group | What they need | What they usually get |
|---|---|---|
| Best customers | First access to new products, and no discount | The same discount as everyone |
| Regulars | A reason to increase basket size, not frequency | Reminders to buy again |
| Could become regulars | A specific nudge toward a second purchase | The newsletter |
| Just arrived | What else you make, within 30 days | A welcome email, then silence |
| Going quiet | A genuine reason to return, quickly | Nothing, until they are gone |
| Cannot afford to lose | A person, not an email | An automated win-back |
| Dormant | Cheap reactivation only | Expensive retargeting |
| Gone | Exclusion from paid targeting | Continued ad spend |
The line that pays for the exercise
Excluding the "gone" segment from paid targeting usually saves more money in the first month than the whole exercise costs in time. Most brands are still paying to reach people who last bought three years ago.
Why does the naming matter so much?
Because segmentation dies in the handoff between the person who built it and the person who has to write the email. A tab called "R5F5M5" gets ignored. A tab called "Cannot afford to lose, 890 people, average value high, last bought 7 months ago" gets acted on, because the person reading it can picture who is in it and what to say to them.
Write a short portrait for each group: who they are, what they want, what stops them, which channel reaches them. Two paragraphs each. That portrait is the brief for every piece of creative aimed at that segment, and it is the thing that makes the segmentation survive contact with the marketing calendar.
The segmentation is rarely the bottleneck. Almost every brand we work with could have done it a year ago. What is missing is somebody writing down what each group should hear, which is a marketing decision rather than a data one.
When is this not enough?
When your purchase cycle is long enough that recency is misleading, which is true for some considered purchases. When you have very few repeat customers, in which case the frequency axis carries no information and you should segment on first-purchase behaviour instead. And when your business is genuinely subscription, where the relevant questions are different.
For most consumer brands with repeat purchase, this works, and it works immediately.
If you are an agency
Segmentation is the highest-leverage first deliverable on a new commerce account, because it is quick, it uses data the client already has, and it produces an immediately actionable list of what to change. It also reframes the relationship: an agency that arrives with the client's own customer base grouped and named is doing something categorically different from one that arrives with a channel plan. Run it as a fixed-scope piece in week one.
How we do this ourselves
Audience Intelligence does this continuously rather than as a one-off. It reads your commerce and analytics data, builds the eight segments with demographics and behaviour attached, and writes a portrait of each one in your brand voice. Those portraits are then available to the creative and campaign work, so a campaign aimed at a segment is briefed from the portrait automatically. Every example we publish is aggregated and synthetic; real customer records never appear in our marketing.
How Audience Intelligence works