# Email segmentation ideas: 10 examples with data and measurement

Page: https://alidemirbas.com.tr/en/blog/email-segmentation-ideas-examples

2026-11-14 · CRM & Lifecycle · Ali Demirbaş

Ten email segmentation ideas, each with the data it needs, a concrete example and a way to measure its effect with a control group from the same segment.

Email segmentation splits your list by data you already hold and decides who gets which message and how often, or who gets nothing at all. The ten email segmentation ideas are signup source, preferences, non-buyers, purchased category, order recency, discounts, click engagement, browsing, delivery city and open tickets.

Whether an idea works shows up only when you compare people inside the same segment. Hold back a random part of the segment and send it the usual email. Comparing the segment with the rest of your list mistakes the segment's own buying habits for the effect of the email.

RFM (recency, frequency, monetary) scoring has its own post, and triggered flows live on the journey pages. Here each idea is a campaign segment, and the email segmentation examples use placeholders where your own data goes.

## Ten segmentation ideas, the data and the measurement

Each of the ten ideas rests on one data field or event type, and each one says up front what the control group gets. The second column usually decides whether you can use an idea. Without the field, or without a link to an email address, there's no segment to build.

| Idea | Data needed | Control group gets | Main metric |
| --- | --- | --- | --- |
| Signup source | Name of the signup form | Content not split by source | People ordering within 30 days |
| Topic and frequency preferences | Preference center fields | No preference question | People unsubscribing within 90 days |
| Subscribers with no order yet | Subscriber matched to orders | Standard newsletter | First orders within 30 days |
| Purchased category | Category on the order line | Standard newsletter | People ordering within 7 days |
| RFM | Order dates, count and value | Standard newsletter | People ordering |
| Discount behavior | Coupon or discount on orders | Coupon campaign | Gross profit per person |
| Click engagement | Last click date | Current send frequency | Orders, unsubscribes, complaints |
| Browsed category | Product view events | Standard newsletter | People ordering within 7 days |
| Delivery city | City in the address | Copy not split by city | People ordering |
| Open support ticket | Ticket in the support system | Current promotion flow | Unsubscribes, complaints, next order |

What the control group gets defines what you measure. If the control receives the standard newsletter, the result tells you what the tailored email added over that newsletter. It doesn't tell you what sending email at all is worth.

## Segments built from signup and preference data

Fields written at signup are the segment data you can use before someone places a first order. Two ideas rely on them.

### Signup source

A signup source segment groups subscribers by the form or channel they joined through. You need a source field written to the profile at signup, such as the form name. You can't infer it later, so if you never stored it, this idea starts with new signups only.

For example, people who opted in at checkout and people who came through a giveaway form get different content in their first month. The first newsletter to the giveaway group says only what you know: "The winner is announced on {date}. This newsletter brings new products and offers every {day}."

Comparing source groups with each other won't show the effect, because people from different channels may behave differently anyway. Hold back a random control within each source group. The control gets the first-month content that isn't split by source, and you compare people ordering within 30 days.

### Topic and frequency preferences

Topic and frequency choices come from the subscriber, not from behavior you infer. This takes the preference center fields and the date each field was last updated.

Someone who picked "offers only" and "once a month" doesn't get the weekly new-arrivals newsletter. Testing a stated preference with a control group would mean ignoring what the people in the control chose. What you can test is asking the question at all.

Show half of your new subscribers a topic and frequency choice in their first email, and don't show it to the other half. Compare the two groups by people who unsubscribe, file spam complaints and click within 90 days.

## Segments built from order data

Order records feed four segments once each email address matches a customer ID. Without that match, order data never reaches your email tool and none of these ideas can be built.

### Subscribers who haven't ordered yet

People on your list with no order at all make up this segment. You need the subscriber record matched to the customer ID in your order table. Guest orders match on email address, and an order that doesn't match keeps a buyer in this segment by mistake.

A campaign to this segment answers questions people ask before a first order: "Returns are accepted within {return_days} days. Shipping is {shipping_terms}." A flow that moves on intent signals, such as a quote request or a new lead, belongs to the [lead nurture journey](https://alidemirbas.com.tr/en/lab/journeys/lead-nurture).

The control gets the standard newsletter, and the main metric is people placing a first order within 30 days. The segment drops buyers on its own, so it shrinks during the test. Assign people from the members on the send date and count results over everyone assigned that day.

### Purchased category

A purchased category segment groups customers by the product categories in their past orders. You need a product or category field on the order line. An order total alone won't do.

For example, customers who bought cat food in the last 365 days get an announcement of new products in the cat category. The subject line says only what you know: "{product_count} new products in the cat category this week."

The main metric is all orders, not only orders in that category. If the category email pulls people away from another category they would have bought from, category orders can rise while total orders stay flat. The hypothetical example section works through this idea with numbers.

### Last order, order count and spend

RFM scores customers on time since their last order, number of orders and spend, then groups them. The segment rests on the date and value of each order, plus a decision on how to handle returns. The analysis window, scoring scale and segment names are covered in the [RFM analysis](https://alidemirbas.com.tr/en/blog/rfm-analysis-customer-segmentation) post.

Customers who used to order often but haven't ordered in a long time might get a campaign about new products, with no discount. The control gets the standard newsletter. Freeze the RFM score as it stood on the day the test starts, so a recalculated score doesn't move people between groups mid-test.

### Discount buyers and full-price buyers

A discount behavior segment groups customers by how many of their orders used a coupon or discount. The field it relies on is the discount amount or coupon code on the order line.

With this segment, customers who ordered without any discount in the last 365 days are left out of the general coupon campaign. They get that week's new-arrivals newsletter instead.

Order counts can mislead here, because the no-coupon group may order less but leave more profit per order. The control gets the coupon campaign, and the main metric is gross profit per person. You can work out product margins with the [gross margin calculator](https://alidemirbas.com.tr/en/calculators/gross-margin).

## Segments built from behavior data

Behavioral segmentation relies on events in your emails and on your site, and here it feeds two ideas: email clicks and site browsing. Both need the event tied to an email address.

### Click-based engagement level

An engagement segment groups subscribers by the time since their last click in your emails. You need click events and the date of the last click. Opens alone won't do.

Apple's [Mail guide for Mac](https://support.apple.com/guide/mail/protect-email-privacy-mlhlp1205/mac) says that with Protect Mail Activity on, remote content downloads in the background when a message arrives. It doesn't wait until the message is viewed. Klaviyo's [engaged segment guide](https://help.klaviyo.com/hc/en-us/articles/115000200072), as updated on October 27, 2026, says Mail Privacy Protection (MPP) inflates open rates. The same guide suggests setting the Apple Privacy Open filter to False to leave out automatic MPP opens.

For example, people who clicked in the last 90 days keep getting the weekly newsletter. People who haven't clicked for 90–180 days get one email a month. Cutting back and then stopping email to long-silent subscribers is a separate flow, covered in the [unengaged subscriber sunset journey](https://alidemirbas.com.tr/en/lab/journeys/unengaged-sunset).

To measure it, a random control inside the 90–180 day group stays on the current frequency. The main metric is people ordering within 60 days, with unsubscribes and spam complaints as guardrails. Fewer emails may bring fewer orders. The question is whether any loss in orders is worth the change in unsubscribes and complaints.

### Browsed but not purchased category

A browse segment collects people who viewed products in a category but didn't order from it. You need product view events tied to a profile with a known email address. That means a logged-in visitor or one who arrived through an email link. Anonymous visits stay out of this segment until the visitor is identified.

One way to use it: take people who viewed at least three shoe products in the last 14 days but bought none. Their weekly newsletter features the shoe category. The triggered flow that sends one reminder after repeat views of the same product is the [browse abandonment journey](https://alidemirbas.com.tr/en/lab/journeys/unresolved-interest-recovery).

The control gets the standard newsletter. Fix the view threshold before the test, because changing it changes who is in the segment.

## Segments built from location and support records

Location and support records do two different jobs. Location tailors a condition in the message to the person's city. A support record takes a group out of promotions for a while.

### Delivery city and time zone

A location segment groups people by the city or time zone in their delivery address. The source is the city on an order or profile address. Apple's Mail guide also says the same option hides your IP address from senders, so a location inferred from an email open may be wrong.

An in-store event announcement, for example, goes only to people with a delivery address in that city. The text says only what you know: "{event} at the {city} store on {date}."

The effect of a store event shows up in store visits. If you can't tie a visit to an email address, clicks are the only thing you can measure on the email side. Write that limit into the test plan.

For content that changes by city, such as delivery times, the control gets copy that isn't split by city. Compare people ordering in the two groups.

### Excluding people with an open support ticket

An exclusion segment removes people in a given state from a send. You need open complaints or return requests passed from your support system to your email tool.

A customer with an open complaint gets no promotional email until the ticket closes, while order and shipping notices continue. Owning and closing the complaint itself is a separate flow, described in the [complaint ownership and closure journey](https://alidemirbas.com.tr/en/lab/journeys/issue-ownership-and-closure).

A random control among people with open tickets keeps getting promotions as it does today. The main metrics are unsubscribes, spam complaints and people ordering within 60 days after the ticket closes. This group may be small, so if there are few events, report counts rather than rates.

## Measure a segment idea against a control group from the same segment

The effect of a segment idea comes from comparing two random groups from the same segment. One gets the tailored message, and the other gets whatever it would have received otherwise. The general logic and sizing of control groups are in the [control group](https://alidemirbas.com.tr/en/blog/control-group-ab-testing) post. Email adds five decisions:

1. Freeze the segment on the send date. A dynamic segment changes as people meet or stop meeting its conditions, so assign from that day's members.
2. Assign people at random and write down what the control gets: the standard newsletter, the current frequency or nothing.
3. Count results by assigned group. Counting only the people who opened or clicked breaks the comparison that random assignment set up.
4. Write the metric, the measurement window and the decision rule before you send. Say which difference counts as a win and which guardrail stops the send at what limit.
5. In a small segment, keep the same people in the same group across several sends, because one send may stay unclear.

An unclear result doesn't mean no effect. It means the measurement couldn't tell a difference of that size from chance. Splitting the result afterward by device or city can produce winners by chance, a risk the [A/B test segmentation](https://alidemirbas.com.tr/en/blog/ab-test-segmentation) post covers in detail.

## Hypothetical example: a category segment with a 20% control group

In this hypothetical example, a store segments 20,000 people who bought from the cat category in the last 365 days. All of them gave marketing consent. The segment is frozen on the send date and people are assigned at random.

The category email goes to 80% of the segment and the usual weekly newsletter to 20%. The 80,000 people on the list outside the segment also get the weekly newsletter.

The decision rule was written before the send. A win means the 95% confidence interval for the difference in people ordering within 7 days sits entirely above zero. If the unsubscribe rate is more than 0.2 points above control, the email doesn't roll out, whatever the order difference.

*Hypothetical example: counts are unique people with at least one order in the 7 days after the send, by assigned group. Rates are rounded to two decimals.*

| Group | People assigned | Ordered within 7 days | Rate |
| --- | --- | --- | --- |
| Category email (treatment) | 16,000 | 480 | 3.00% |
| Weekly newsletter (control) | 4,000 | 100 | 2.50% |
| Rest of the list | 80,000 | 800 | 1.00% |

Comparing the treatment group with the rest of the list shows a gap of 2.00 points. The control group shows that the same segment orders at 2.50% even without the category email.

So 1.50 points of the visible gap is the segment's own habit, and the measured difference from the category email is 0.50 points.

The 95% confidence interval for that difference runs from −0.05 to 1.05 points. The lower bound is below zero, so under the decision rule the result is unclear: the effect could be zero or about one point.

On the guardrail, 48 people in the treatment group and 8 in the control unsubscribed.

The unsubscribe rates are 0.30% and 0.20%, so the gap stays under the 0.2-point limit. But eight events in the control are too few to read that rate with any precision.

With the rates unchanged, raising the control share to 50% pushes the interval entirely above zero. When the segment shrinks to 8,000 people, even a 50% control isn't enough.

*Hypothetical example: order rates are fixed at 3.00% for treatment and 2.50% for control, and only segment size and control share change.*

| Segment and control share | Treatment / control people | Difference and interval | Decision |
| --- | --- | --- | --- |
| 20,000 people, 10% control | 18,000 / 2,000 | 0.50 points (−0.23 to 1.23) | Unclear |
| 20,000 people, 20% control | 16,000 / 4,000 | 0.50 points (−0.05 to 1.05) | Unclear |
| 20,000 people, 50% control | 10,000 / 10,000 | 0.50 points (0.05 to 0.95) | Win |
| 8,000 people, 50% control | 4,000 / 4,000 | 0.50 points (−0.22 to 1.22) | Unclear |

The interval is a 95% confidence interval for the difference of two rates, using the normal approximation. It holds when each group has dozens of events and breaks down below about 10 expected events in a group.

A 50% control means half the segment misses the category email on this send. Even in the winning row, the effect comes to an estimated 50 extra orders in a treatment group of 10,000.

The interval for that estimate runs from 5 to 95 orders. If you want to keep the control share small, hold the same assignment over several sends and read the results together. You can get campaign-level click and unsubscribe rates from the [email performance calculator](https://alidemirbas.com.tr/en/calculators/email-performance), but the decision comes from the difference measured against the control group.

## What to check when you build the segment in your platform

In Klaviyo and Mailchimp, a segment is a rule made of conditions applied to people's data. Their help pages give a few details that matter for a test.

Klaviyo's [getting started with segments](https://help.klaviyo.com/hc/en-us/articles/115005237908) article, as updated on October 28, 2026, calls lists static and segments dynamic. A segment grows as people meet its conditions and shrinks when they no longer do. The same article describes a segment snapshot, which copies the current members into a static list. That can be one way to freeze a segment on the send date.

Mailchimp's [getting started with segments](https://mailchimp.com/help/getting-started-with-segments/) page gives a plan limit, as of November 2026. On the Free or Essentials plan, a segment can hold up to 10 conditions with one and/or operator. The advanced segment builder on Standard and Premium plans has no limit on conditions. Check your plan's limit before you build an idea that nests several conditions.

## Consent is the first condition of every marketing segment

Behavior, category and preference conditions don't stand in for consent, and both platforms treat consent as a separate condition. The Klaviyo article above says segments don't necessarily carry opt-in criteria. To limit an email segment to subscribers, it suggests adding the email marketing consent condition with AND (every condition must be true).

Mailchimp's page says that when you send an email, anyone who isn't subscribed is removed. This post doesn't cover which consent the law requires where your recipients live. Confirm that for your own markets before you build segments on top of it.

## Pick your first two segments from your own data

To pick your first segment, check the "data needed" column against your own systems. Is the field filled in, is it current and is it tied to an email address? Among the ideas whose data is ready, base one on order data, such as category, discount or RFM. Base the other on send frequency, through click engagement or preferences.

Before you send, write four things for each one on a single page. Note what the control gets, the 7- or 30-day measurement window, the difference that counts as a win and the guardrail limit. If the segment is too small to detect that difference under the rule, raise the control share or keep the same assignment across several sends.