Email list segmentation is splitting one subscriber list into smaller groups, so each group gets a different email, a different timing, or no email at all.
Building segments is the easy part. The harder question is which ones are worth the upkeep.
The familiar figures show better results for segmented campaigns. The stronger evidence says the obvious segments, your riskiest customers and your best ones, are not always the groups that respond.
Email list segmentation, in short. A segment is a group defined by a rule, not a fixed set of names. It earns its place when three things hold: the group behaves differently in a way you can see, you have a genuinely different treatment for it, and a randomized comparison shows that treatment beats what the same people would otherwise get. The best-known figures come from a 2017 comparison that was not an experiment. The field experiments point toward testing segments, not assuming them.
What is email list segmentation?
It is dividing subscribers into groups by a rule, such as what they bought, what they told you, when they joined or how they have engaged, and sending each group something different.
The point is relevance. Someone who has never bought and a customer on their fifth order do not need the same email. A subscriber who stopped reading months ago may need none.
What is the difference between a list and a segment?
A list is a fixed collection of people, added by signup or by hand. A segment is a rule that keeps selecting people as they meet or stop meeting its conditions.
Klaviyo’s help center draws the line this way: lists are static, while segments are “defined by a set of conditions” and “grow and shrink based on customer behavior”. People can join a list through a form or be added to it by hand. Nobody can be added to a segment except by meeting its rule.
That is why the same documentation recommends one main list per channel, holding everyone who has consented to marketing, and doing the dividing with segments.
What can email segments be built from?
Four useful signals to start with are what people bought, what they told you, when they joined, and how they have engaged with your email.
- Purchase behaviour. Never bought, bought once, bought repeatedly. Bought one product but not a related one. Spent over a threshold.
- Stated interest. Preferences a subscriber picked on a signup form or preference page.
- Signup date. Recent joiners versus people who joined long ago.
- Engagement. Who has clicked or bought recently, and who has gone quiet.
Recency, frequency and monetary value combine three purchase signals into one score. RFM segmentation covers how that scoring works and where it breaks.
Does segmentation actually improve results?
The best-known figures show better results for segmented campaigns, but not that segmenting caused them. They come from a 2017 comparison of senders’ own campaigns, not a controlled test, so they show an association rather than an uplift to expect.
Mailchimp’s study sampled about 2,000 of its users who sent about 11,000 segmented campaigns to almost 9 million recipients. It compared those campaigns with the same users’ unsegmented ones. Segmented campaigns had opens 14.31% higher, clicks 100.95% higher and unsubscribes 9.37% lower. The page is dated February 2017.
Why is that comparison weaker than it looks?
The page describes no random assignment. It compares each sender’s segmented campaigns with that sender’s unsegmented ones, so the gap mixes segmentation with differences in what was sent and to whom.
The study says as much. Its figures are “a little tough to interpret”, because “the hows and whys of each user’s implementation of the various segments has a tendency to muddy the waters.”
Which segment types had worse complaint rates?
In Mailchimp’s 2017 comparison, segments built by signup date and by subscriber activity had higher abuse-report and unsubscribe rates than the same senders’ unsegmented campaigns.
Signup-date segments had opens 29.56% higher, but unsubscribes were 33.76% higher, abuse reports 29.55% higher and bounces 55.18% higher. Activity segments had abuse reports 10.34% higher and unsubscribes 5.49% higher.
The other two types went the other way. Segments built from interest groups, meaning what subscribers ticked on a signup form, had clicks 74.53% higher, unsubscribes 25.65% lower and abuse reports 17.78% lower. Segments built from stored subscriber fields, such as customer type or location, also had lower unsubscribe and abuse rates.
So the segments built on who people are or what they told you went with lower complaint rates, and the two built on timing and past activity went with higher ones. The study’s own explanation points at how segments were used, not only at what they were built from: an activity segment can be used to chase the least engaged subscribers, and old signups carry inactive addresses that bounce.
Should you target the customers most likely to leave?
Not automatically. In two field experiments published in the Journal of Marketing Research, Eva Ascarza found that customers at the highest risk of churning “are not necessarily the best targets” for retention programs.
In the first study, a wireless provider in the Middle East tested free credit on prepaid top-ups. Targeting the 40% of customers at highest risk would have cut churn by 1.9 percentage points. Targeting the 40% most responsive to the offer would have cut it by 6.0.
In the second, a North American membership organization tested a thank-you gift with its renewal letter. There, targeting the highest-risk 40% would have increased churn by 4.4 points, while targeting the most responsive 40% would have reduced it by 4.3.
The two groups barely overlapped. Of the top 10% by risk, only 16% were also in the top 10% by response in the first study, and 6% in the second.
Neither offer was an email, so those figures belong to those two campaigns. The lesson travels: a segment chosen because it looks important is not the same as a segment that responds.
Do your best customers respond best?
Not necessarily. A study in the Journal of the Academy of Marketing Science (May 2024 issue) modelled a beauty retailer in six countries and a US apparel retailer. Response to email and direct mail varied by customer value segment, and the authors found high-value customers less responsive to expensive direct mail, not more. In their words, high-value status “does not mean greater responsiveness to marketing actions”.
The same paper carries a sharper warning. Its model of past sales found email significantly effective for medium- and high-value segments. In the randomized field experiment that followed, “email is not effective for any of the segments”. Using the experiment’s estimates, the authors calculated that reallocating the same marketing budget by segment would lift revenue 6.5% against business as usual. That reallocation was a calculation, not a tested group.
A segment-level finding from a peer-reviewed model did not survive the experiment. That is the case for testing a segment rather than trusting it.
Which segments are worth building?
A segment is worth building when the group behaves differently in something you can see, you have a genuinely different treatment for it, and a randomized comparison can show that treatment beats what those people would otherwise get.
- Does the group behave differently in a way you can observe? Purchases, clicks and stated preferences are observable. A guess about who someone is, with nothing behind it, is not.
- Do you have a genuinely different treatment? A different email, offer or timing counts, and so does leaving the group out. If every segment would get the same email, the segment is a filter, not a strategy.
- Can you check it? Randomly split the segment between its own treatment and whatever it would otherwise get, then compare margin per person.
A practical set to test first:
- Purchase stage. Never bought, bought once, bought repeatedly. A first-time buyer needs a different email from a repeat customer.
- Stated interest. One of the two segment types in the 2017 figures that went with lower complaint rates, not higher ones.
- Value. Spend over a threshold, or the fuller RFM version. Keep the warning above in view: high value does not mean high response.
- An engagement exclusion. People who have stopped engaging, taken out of the main send. The example Klaviyo gives is anyone who received but did not open your last 20 emails. Opens are now partly hidden (see below), so check clicks and purchases before removing anyone.
How do you check whether a segment is working?
Randomly split the segment into two groups. One gets the segment’s own treatment. The other gets whatever it would otherwise receive: usually the standard campaign, or no email if leaving people out is the proposal. Compare margin per person, after product, discount and shipping costs, between the two groups over the same period, not opens.
That per-person difference, multiplied by the number of people who get the segment’s treatment, is what the segment added. Subtract what it costs to write, build and maintain its emails, and you know whether it pays for itself. If the groups are too small for the difference to stand out from ordinary variation, the answer is “not yet”, not “no”.
Ascarza’s approach is a fuller version of the same idea. Run the campaign on a randomly selected pilot group first, “instead of targeting specific customers on the basis of some prespecified rule”. Compare treated and untreated customers to find which kinds of customer the campaign actually moved, then target people who look like them, not simply the pilot’s responders. Her summary of the two studies was that “half of the retention money is wasted”, and that estimating how response varies shows which half.
Can you still build engagement segments on opens?
Only partly. Apple’s Mail Privacy Protection “prevents senders from seeing if you’ve opened the email message they sent you.” For readers who turn it on, an “opened in the last 30 days” segment is not measuring what its name says.
Build engagement on things a reader has to do, such as clicking, buying or replying. Treat opens as a weaker supporting signal, if you use them at all. This matters most where a segment decides who counts as lapsed, as it does for a win-back email.
Can a segment include people who never opted in?
Yes, in some tools, so matching a segment is not the same as agreeing to marketing. In Klaviyo, segments “pull from everyone in your account, including people who have not necessarily opted in to receive marketing emails from you” unless the segment is confined to a list. Build every campaign segment inside the consented list.
Is segmentation still an advantage?
Not by itself. The DMA’s Marketer Email Tracker 2026, based on 250 marketers, calls segmentation “firmly established” alongside lifecycle tracking and lifetime-value measurement. It puts the opportunity in “how consistently these capabilities translate into performance.”
Having segments is the baseline. The difference is whether they change what people do.
Where does this advice stop applying?
It stops where a segment is too small for a randomized comparison to give a useful answer, where the research offers differ from your emails, and where the 2017 figures are read as a forecast.
- Small lists. A segment can be too small for a holdout to show a clear difference. With a small list, use fewer and broader segments, and treat an unclear result as inconclusive.
- The research figures. Ascarza’s studies tested prepaid credit and a renewal gift, and the retail experiment ran at one beauty retailer. They show that segment responses can surprise you. They do not tell you which of your segments will respond.
- The 2017 comparison. Mailchimp’s figures come from one observational comparison, including the cases that went wrong. They are associations, not an uplift to expect.