In e-commerce, segmentation means dividing your customer list into groups by shared traits, above all by shopping behavior, and choosing a separate offer, message or sending time for each group. Last order date, order count, amount spent and the categories a customer has shopped in are a reliable basis for this grouping, because you can pull every one of them from your store’s order and product records.
When you finish reading, you will be able to decide which types of segment make sense for your store and use a spreadsheet program to pull your first four or five segments from an order export. You will also know which segment belongs in an automated flow and which belongs on a campaign list.
What does it mean to segment?
To segment is to divide an audience into subgroups whose members share similar traits. In marketing, the criteria for the split usually fall under four headings: demographic (age, gender, income), geographic (country, province, region), psychographic (lifestyle, values) and behavioral (purchasing, usage, loyalty). That definition correctly answers the question of what customer segmentation is, but for an online store it leaves one thing out: what data do you actually have?
Market segmentation divides the entire market you target into groups, including people who are not your customers yet; it is what you use when deciding which age group or which city to advertise to. Customer segmentation, by contrast, works with the people you already have: it groups customers who have placed an order, become members or given you permission to contact them. In that second case, the most reliable information is what the customer has done. If your sign-up form never asked for a date of birth or gender, you have no demographic data. Your system does, however, record every customer’s last order date, how many times they have shopped and which products they bought.
Segments are also often confused with personas. A persona is a representative profile, such as “a working mother in her thirties who does her shopping on weekends,” and it guides you when you write content. A segment, on the other hand, is the list of real people that a rule produces on a given day. The rule might be “customers who shopped in the baby category in the last 90 days,” and the list it returns that day might hold 412 phone numbers. In Pushouse, Website AI Reporting also builds the rule from a sentence like this: when you type “customers who bought coffee in the last 30 days,” the sentence becomes an audience rule that is applied to purchase history.
Five segment types you can build in an online store
You do not need to set up all of the types below at once. Depending on how often customers buy your products again and how many customers you have, starting with two or three of them is often enough.
Recency, frequency and spend: RFM
RFM stands for recency (the time since the last purchase), frequency (the number of orders) and monetary value (total spend). You calculate all three from your order records and give each customer a score on each measure. Scoring can use five levels, but for a store with a few thousand customers, three levels are often enough and make the result easier to read. Suppose that, after looking at your own data, you chose these thresholds:
- A customer whose last order was 60 days ago or less scores 3, one whose last order was 61 to 180 days ago scores 2, and one whose last order was longer ago than that scores 1.
- Once cancellations and returns are taken out, 4 or more orders score 3, 2–3 orders score 2 and a single order scores 1.
- When you rank customers by total spend, the top third scores 3, the middle third 2 and the bottom third 1.
Take four customers. Ayşe ordered 20 days ago, has five orders in total and is in the top spending third; she scores 3-3-3 and belongs to the store’s loyal core. Mehmet placed his last order 240 days ago and has four orders and high spend; he scores 1-3-3. He is the most valuable type of customer you are about to lose, and he deserves a personal message rather than a general discount announcement. Zeynep placed her first order 15 days ago and her spend falls in the middle third; at 3-1-2 she is a new customer, and what you are hoping for from her is a second order. The fourth customer, Can, made a single low-value purchase 400 days ago and scores 1-1-1. For most stores, it is enough to include him in the general list for big seasonal campaigns only, rather than putting separate effort into him.
Three measures at three levels give 27 possible score combinations. Do not try to manage them one by one; merge them into four or five meaningful groups. You could group them like this:
- A customer with a recency score of 3 and a frequency score of 2 or 3 belongs to the loyal core.
- A customer with a recency score of 3 and a frequency score of 1 is a new customer.
- A customer with a recency score of 1 or 2 and a frequency or spend score of 3 is a cooling high-value customer.
- A customer who scores 1 on all three is a one-time buyer.
- A customer who fits none of these rules stays on the general list.
One-order customers and repeat buyers
Even if you never calculate RFM scores, the first split to make is between customers with a single order and those with two or more. Check what share of the customers in your own data stopped at one order. If that share is high, a reasonable starting point for most stores is to focus on bringing existing customers to a second order before looking for new ones, because you already know what these customers bought and how to reach them. Members who have never ordered fall outside this split; with no order records, they cannot get an RFM score either. Reach them through a separate flow with first-purchase incentive messages. In a campaign send, do not put these members in the same group as customers who have ordered, and do not add a member who is currently receiving messages from the incentive flow to a campaign list at the same time.
Category interest
The categories a customer buys from tell you what to say to them. In a store that sells cosmetics, a customer who buys skincare and one who buys makeup may react differently to the same announcement. You can build a category segment in two forms. The first is the “bought” form: customers who bought a hair care product in the last six months. The second is the “bought, but not the complementary product” form: customers who bought a shampoo but not the hair mask from the same range. The second kind is the natural target for a cross-selling message.
For products that run out regularly, add a time element to the category. A customer who buys a monthly pack of coffee may need a new pack about a month later. Messages timed to how long a product lasts are called WhatsApp reorder reminders; they work like a category segment tied to a calendar.
Cart and stock behavior
Some segments are short-lived or wait for an event. A customer who added products to the cart but did not complete payment has to be reached before it is too late, while a visitor who asked to be notified when a sold-out product is back expects to hear from you the moment it is restocked. A list you pull once a month cannot keep up with these people; you need automations that run at the moment of the behavior or event. A WhatsApp cart recovery flow responds to an abandoned cart, and a WhatsApp back-in-stock alert flow responds to the awaited product coming back into stock.
You can also build longer-term groups from the same behavior. A customer who has abandoned a cart several times in the last two months without ever ordering may still have a question about the price, the shipping fee or the delivery time. In your next campaign, it is worth sending them a message that spells out the delivery time and the return terms. Cart activity does not appear in an order export; you can build this group from the records of the tool you use for cart reminders. A customer who asks for a stock alert on a product they have never bought has also shown interest in that category; you do not need to wait for them to order before adding them to the category segment.
Location
The province in the delivery address matters for seasonal products and when shipping times vary by region. A winter coat campaign is relevant to a customer in Erzurum weeks before it is relevant to one in İzmir. Rather than using a location segment on its own, combine it with a category, as in “customers who bought outerwear in the last year and whose delivery address is in Eastern Anatolia.” For date-bound campaigns such as Mother’s Day, send the reminder about the last order day a day or two early to customers in provinces where deliveries take longer.
Build your first segments from an order export
You do not need new software to get started. For a first version, an order export from your store’s admin panel, a product export if needed, and a spreadsheet program are enough. Work through these steps in order:
- Export the orders from the last 24 months. The file should include columns for order number, order date and status, the customer’s phone number and email address, the amount paid, product code and category. On some e-commerce platforms the order export has no category column; in that case, export the product list separately and match categories by product code (SKU). In Excel, you do this with the VLOOKUP function.
- Remove canceled orders, orders whose payment was never completed and returned orders. If they stay in the file, customers’ frequency and spend will look higher than they really are. If marketplace orders also come into your admin panel, keep them separate; otherwise they get mixed into the frequency and spend figures for your own site’s customers.
- Merge records that belong to the same person. A customer who checks out as a guest can look like a different person on every order. Putting phone numbers into a single format (for example, ten digits with no leading 0 or +90) and merging by number removes most of the duplicates.
- For each customer, calculate the last order date, the number of valid orders and the total net amount; a pivot table can do this for you. If the file has a separate row for each product, count orders by the number of distinct order numbers, not by the number of rows.
- Choose thresholds based on your own data and write the segment name next to each customer.
- Add a column for contact permission; the order export may not include this information, in which case take it from your member records. Count the people with permission in each segment separately. A customer without permission can appear in a segment, but they do not go on the message list.
A useful measure when choosing thresholds is how long your customers wait between two orders. For the customers who have placed a second order, calculate the number of days between their first two orders and take the median. Use the median rather than the average, because a few customers who come back after two years pull the average up more than they should. In a store where the median comes out at 40 days, a reasonable starting point is to treat twice the median, 80 days, as the point at which a customer is cooling off, and four times the median, 160 days, as the point at which they have lapsed. You can also base the recency thresholds in your RFM scoring on these two points. In Excel, the MEDIAN function finds the median, and the PERCENTILE function finds the cut-off points that divide spend into thirds.
For products bought rarely, such as furniture or major appliances, these intervals stretch into years and frequency stops being a meaningful measure. For stores like these, category, spend and location segments are more useful. In stores that sell products that run out regularly, such as food, cosmetics or pet supplies, frequency is often the measure that tells you the most.
Look at how many people each segment contains as well. Writing a separate message for a group that makes up only a small slice of the whole list may not repay the time you spend on it. If most of the list piles up in one group, review your thresholds. If that large group is the general list, the thresholds for the other groups are too tight; if it is a group that is naturally large, such as one-order customers, split it in two by last order date.
If you would rather not redo all of this by hand before every send, you can type the group you are looking for into Website AI Reporting as an ordinary sentence. You can read the rule that appears on screen and correct it, see the group’s net revenue and how many of its members can be messaged, and then download the list as an Excel file or pass it to a bulk message send.
Adapt the WhatsApp message to each segment
On WhatsApp, a campaign message sent to someone who has no interest in it costs you money directly. According to Meta’s WhatsApp pricing documentation, pricing has been per message since July 1, 2025: a template message is charged when it is delivered, and all marketing templates are charged.
We covered how the offer, the opening line and the link change from group to group within a campaign in our guide to which customer groups need a separate campaign message. The three examples below show how the message differs for three of the groups you built above: the cooling high-value customer, the new customer and the category customer who has not bought the complementary product.
For the cooling high-value customer, Mehmet in the example above, the first step can be a reminder instead of a discount:
Hi {{customer name}}, new colors have arrived this month in the {{product name}} range you’ve bought from before. You can browse the products we picked for you here: {{category link}}
If you’d rather not receive our promotional messages, you can tap “Stop promotions” below.
If no order comes in after the first message, the second step can be a personal discount code. Instead of keeping track of the two steps by hand, you can build them as a single flow in Pushouse’s WhatsApp We Miss You module: a flow can consist of several steps, and if the customer places an order in the meantime, the next step does not go out.
For the new customer, Zeynep, a message that offers help after her order has been delivered is a better fit than a sales offer:
Hi {{customer name}}, your {{product name}} has arrived. We’ve put together a few short tips to help you use it for longer: {{link}}. If you have any questions, you can reply to this message.
Send the offer for a second order close to the date when the customer will need the product again. In a category segment, for example customers who bought a shampoo but not the hair mask, the message highlights the complementary product:
Hi {{customer name}}, the hair mask from the same range as the {{product name}} you use is {{discount rate}}% off this week. Have a look here: {{product link}}
To stop promotional messages, just tap “Stop promotions” below.
The opt-out line in the examples points to the “Stop promotions” button added at the bottom of the message. Do not ask customers to type a single word such as “STOP”; if your sending tool does not act on that word, the customer stays on the list. Once the segment is ready, you use it as the recipient list in a WhatsApp bulk messaging send.
Contact permission is the first filter on every segment
However accurate a segment is, a promotional message goes only to people who have given permission. Türkiye’s Law No. 6563 on the Regulation of Electronic Commerce makes commercial electronic messages conditional on the recipient’s prior consent (Article 6). Recipients may refuse these messages whenever they wish, with no need to state a reason, and once a refusal reaches the service provider, sending must stop within three business days at the latest (Article 8). WhatsApp’s Business Messaging Policy also sets two conditions: the person must have opted in to messages from you beforehand, and you must honor their request when they want to opt out.
For consent to be valid, it must also be registered in the Message Management System (İleti Yönetim Sistemi, İYS), Türkiye’s national system for recording consent to commercial electronic messages. Article 7 of the Regulation on Commercial Communication and Commercial Electronic Messages treats consent that is not registered in İYS as invalid. So when you fill in the permission column in your export, check it against the records in İYS.
Article 6 of the Law also provides an exception: if recipients gave their contact details so that you could get in touch with them, messages that explain how to use or maintain the goods or services they received from you, or that tell them about changes to those goods or services, need no separate consent. The exception only removes the consent requirement in the Law; the permission requirement in WhatsApp’s messaging policy still applies, so even the usage tips you share after delivery may go only to customers who agreed to hear from you on WhatsApp. Once you add a discount or another product to those tips, the message may become promotional; send a message like that only if the customer has given permission for commercial messages.
In practice, think of permission as a filter that sits above every segment. When a customer says they no longer want your messages, removing them from that one campaign’s list is not enough; take them off all your sending lists and out of your automated flows as well.
Mistakes that make segments pointless
The first mistake is building a segment once and then holding on to the list. What defines a segment is its rule; the list is only that day’s result, and a customer who was in the “cooling” group in March may have placed an order in April. Regenerate the list from the rule before every send, and update RFM scores at least once a month. Sending to an old list means writing “we miss you” to a customer who shopped with you yesterday.
The second mistake is a segment that makes no difference. If you send two groups the same text with the same offer, separating them gains you nothing. Limit the number of segments to the number of genuinely different messages you can write. In a store whose customers number in the hundreds, managing three groups is easier than managing twelve.
The third is demographics based on guesswork. Guessing gender from first names to build a “female customers” segment produces mismatches, and a message that reaches the wrong person can put the customer off. Turkish names such as Deniz, Umut and Evren are used as both women’s and men’s names. If you have no self-declared information from the customer, rely on behavior: a rule such as “customers who shop in the women’s clothing category” is both more accurate and easier to defend than a guess based on a name.
The fourth is reading the result from the totals. The total number of orders received in a campaign week does not tell you which of them came because of the message. Before sending, use Excel’s RAND function to add a column of random numbers to each segment’s list, sort the list by that column, and set aside the first ten percent as a control group that receives no message. Two weeks later, compare the order rate of those who received the message with that of those who did not.
In a segment of a few hundred people, the group you set aside is a few dozen; in a group that small, both a difference of a few orders and no difference at all may be down to chance. Do not base your decision on a single send; read the results of several sends with the same offer together. If there is still no difference, the orders from the people who got the message would most likely have come without it; the message to that group is not bringing in new sales, so change the offer or the timing. If a difference shows up in the first two weeks and disappears after a month, the message may only be pulling forward orders that would have come anyway. You can see how cart, stock, order and campaign messages are set up together on our WhatsApp marketing page; if you would like to write your first segment rule with us using your own order data, get in touch.
