Email · Strategy
Subject line testing is the most-written-about and least-productive activity in email marketing. Who receives the message determines more than what it says — and in 2026, it determines whether it arrives at all.
There is an entire genre of email marketing content devoted to subject lines. Power words, emoji placement, character counts, curiosity gaps, the optimal number of words. I have read a great deal of it, written some of it, and come to think it is close to a distraction.
Here is the argument. A subject line operates on people who already received the email and already decided, at some level, whether your name in their inbox deserves attention. It is the last variable in a chain, and by the time it acts, most of the outcome is settled. Segmentation operates on the whole chain: it changes who gets the message, which changes engagement, which changes your sender reputation, which changes whether future messages reach the inbox at all.
That last link has become decisive. Since major mailbox providers began enforcing bulk sender requirements — spam complaint thresholds at 0.3%, with Google recommending below 0.1% — sending broadly to people who don’t want your mail is no longer merely inefficient. It actively degrades delivery to the people who do.
Segmentation isn’t an optimisation tactic. It’s infrastructure protection.
Start With the Only Segment That Matters Engagement, before anything clever
Before demographics, before purchase history, before anything sophisticated: split your list by recency of engagement.
Active — engaged in the last 30 days. Lapsing — 31 to 90 days. Dormant — 91 to 180. Dead — beyond 180.
This single split does more for a programme than every other segmentation strategy combined, for a reason that isn’t obvious: it lets you vary volume by segment rather than content.
Mail the active segment freely. Mail the lapsing segment less often, and only with your strongest material. Mail the dormant segment rarely. Don’t mail the dead segment at all except through one re-engagement sequence, after which you remove them.
The mechanism: mailbox providers evaluate reputation on aggregate engagement. A send to 20,000 people where 15,000 are dormant produces a weak signal and mediocre placement. The same content to the 5,000 active people produces a strong signal, better placement for the next send, and roughly the same absolute number of opens — because the dormant 15,000 weren’t opening anyway.
You send to fewer people and reach more of them. That result is counterintuitive enough that most people need two or three sends to believe it.
Segment Two: Lifecycle Stage Where someone is, not who they are
The second-most-valuable split is position in the relationship, because it determines what the message should even be about.
Never purchased. They need proof, objection handling and a reason to trust you. Sending them a loyalty offer is nonsense.
Purchased once. The most important and most neglected group in most lists. The gap between one purchase and two is where a business either compounds or doesn’t. These people need the second-purchase case made explicitly.
Repeat customers. They’ve decided. Sell to them less and give them more — early access, useful information, the sense of being on the inside. Discounting to people who already buy at full price is simply reducing your margin.
Lapsed customers. Bought once, then stopped. Different from never-purchased in a crucial way: you know what they wanted. That’s actionable information most lists throw away.
Lifecycle segmentation is usually available from your commerce platform without any additional data collection, which makes it the cheapest sophisticated segmentation there is.
Segment Three: RFM, For Anyone Selling Repeatedly Old, unglamorous, works
Recency, Frequency, Monetary value. A direct-mail technique from decades before email that has never been improved on for the job it does.
Score every customer on three axes — how recently they bought, how often, and how much — typically in quintiles. The combination produces groups that behave very differently and warrant genuinely different treatment.
Recent, frequent, high value: your best customers. Ask them for referrals and reviews. Never discount to them.
Recent, infrequent, low value: new and uncertain. Nurture, don’t push.
Not recent, frequently bought, high value: this is the group to focus on. They used to be excellent customers and have gone quiet. Something changed — a bad experience, a competitor, a life event. Winning back a former high-value customer is dramatically cheaper than acquiring an equivalent new one, and this cell is where the money is in almost every list I’ve looked at.
Not recent, infrequent, low value: let them go.
The reason RFM survives is that it uses only data every business already has, requires no tracking infrastructure, and produces groups that map onto obvious actions. You can build it in a spreadsheet in an afternoon.
“The customers who used to buy often and stopped are the most valuable segment in most lists, and almost nobody has a specific plan for them.”
Segment Four: Declared Preference Just ask
The simplest segmentation available and the most consistently skipped: ask people what they want and record the answer.
A single question at signup — which of these three things brought you here? — produces a segment that behaves better than most inferred ones, because it’s a statement of intent rather than a guess from behaviour.
Better still, put a preference centre behind your unsubscribe link. A meaningful proportion of people clicking unsubscribe don’t object to you; they object to frequency. Offering “monthly instead of weekly” or “product news only” converts a permanent loss into a retained subscriber. On most lists that recovers somewhere between a tenth and a quarter of intended unsubscribes, and unlike almost everything else in this article it takes an hour to implement.
The condition: if you ask preferences, honour them. Collecting a preference and ignoring it is worse than never asking, because you’ve demonstrated that your stated interest in their wishes was decorative.
Segment Five: Behavioural Powerful, fragile, build last
Based on what people do on your site or in your product rather than what they’ve bought. Viewed a category three times. Read a specific guide. Started a trial without completing setup. Abandoned a form.
Genuinely effective when it works, and the most fragile thing in this article. Behavioural segmentation depends on tracking, tracking breaks silently, and a segment that stopped populating in March looks identical to a segment nobody qualified for.
Build one or two, verify quarterly that they’re still filling, and don’t build a programme that depends on them.
What Segmentation Costs You The honest ledger
Almost every article on this subject presents segmentation as free. It isn’t.
Production time multiplies. Four segments receiving genuinely tailored content is four times the writing. Most people build the segments, discover they can’t fill them, and end up sending the same email to all four — which is worse than not segmenting, because they’ve added complexity for no differentiation.
Small segments produce unreliable data. A segment of 200 people gives you results you cannot distinguish from noise. Below roughly a thousand, treat performance differences with real scepticism.
Overlap creates duplicate sends. Someone in three segments receives three emails on the same morning unless you’ve configured exclusions. This is the most common practical failure, and it converts a segmentation project directly into an unsubscribe spike.
Maintenance accumulates. Every segment is a definition that can drift, a rule that can break, and a thing to check when something looks wrong.
So the realistic advice is: three to five segments, maximum, for most operations. Engagement recency, lifecycle stage, and one thing specific to your business. Anything beyond that needs either a team or a genuine reason.
Frequency Is a Segmentation Decision The variable nobody adjusts
Most programmes pick one send frequency and apply it to everybody. That’s a choice, and usually the wrong one, because the right frequency for your most engaged subscribers is far higher than the right frequency for your least.
The people who open everything you send would happily hear from you weekly, and treating them the same as everyone else means under-monetising your best audience to protect an audience that isn’t listening. Meanwhile the marginal subscriber who receives four emails a month is being pushed toward the complaint button that damages delivery for everybody.
A workable frequency ladder: weekly to active, fortnightly to lapsing, monthly at most to dormant, and nothing to dead outside the re-engagement sequence. Same content calendar, different distribution.
The effect is compounding. Your engaged segment gets more contact and generates more revenue. Your disengaged segment generates fewer complaints. Aggregate engagement rises, which improves placement, which raises engagement further. Frequency segmentation is the cheapest available improvement to a mature list and almost nobody does it, because the mental model is “the newsletter goes out on Tuesday” rather than “different people warrant different contact.”
Testing Segments Rather Than Subject Lines Where the experiments should go
If you’re going to run tests — and you should — point them at targeting rather than wording.
Test holding back a segment. Send to active only, then compare total opens, clicks and revenue against a previous send that went to everybody. Most people are astonished that the numbers barely move, or improve. That single test does more to change behaviour than any amount of argument.
Test frequency on a subset. Take a random half of your active segment and send twice as often for a month. Watch unsubscribes, complaints and revenue per subscriber. You’ll find your ceiling, and it’s usually higher than you assumed.
Test the same content to two lifecycle stages. Identical email to one-time buyers and repeat customers. The performance gap tells you how much lifecycle-specific writing would be worth before you commit to producing it.
Test the preference centre. Route half your unsubscribe clicks to a preference page and half to a straight unsubscribe. The recovery rate is the value of building it properly.
These tests take one send each and produce structural information. A subject line test, by contrast, produces a result that applies to one email and doesn’t generalise, which is why programmes can run subject line tests for two years and be no better at email than when they started.
Segment Hygiene: The Maintenance Nobody Mentions Quarterly, half an hour
Segments decay in specific, predictable ways, and a decayed segment is worse than none because you’re acting on a definition that no longer means what you think.
Check every segment is still populating. A rule referencing a tag you stopped applying, or a field you renamed, produces an empty or shrinking segment with no error. Look at the counts, not just the definitions.
Check for overlap. Run the numbers on how many contacts appear in more than one segment you actively mail. If the answer is high, someone is receiving multiple emails on the same day and you’re generating avoidable complaints.
Re-run the engagement classification. These segments are time-based and only mean anything if recalculated. Most platforms handle this dynamically; verify yours does rather than assuming.
Delete segments you haven’t sent to in six months. Every unused segment is clutter that makes the useful ones harder to find and increases the chance of sending to the wrong one at speed.
One More Thing Worth Trying The reply-driven segment
A segment almost nobody builds and which outperforms nearly everything above: people who have ever replied to you.
Most platforms won’t build this automatically, because replies land in your inbox rather than in the marketing tool. You tag them manually, which takes a few minutes a week, and the resulting group behaves unlike any other segment on your list. They open at dramatically higher rates, they buy more readily, and their engagement is a strong positive signal to mailbox providers.
The reason is straightforward: a reply is the highest-effort action a subscriber can take. Someone who has typed a sentence to you has crossed a threshold that a click doesn’t measure, and they tend to stay across it.
Two uses. First, treat them as your highest-value engagement segment for frequency purposes. Second, and more useful: when you’re about to send something to the whole list and you’re unsure about it, send it to the repliers first. Their response tells you whether the idea works before you commit your reputation to a full send. It’s the cheapest testing panel you will ever assemble, and it costs nothing but the discipline of tagging.
The Segments Worth Building
| Segment | Data needed | Effort | What it changes |
|---|---|---|---|
| Engagement recency | Opens and clicks — you already have it | An hour | Deliverability for the entire programme |
| Lifecycle stage | Purchase history from your store | An hour | Message relevance; second-purchase rate |
| RFM cells | Order dates, counts, values | An afternoon | Identifies lapsed high-value customers |
| Declared preference | One signup question; preference centre | An hour | Recovers a share of intended unsubscribes |
| Purchase category | Product-level order data | Half a day | Cross-sell relevance |
| Behavioural triggers | Site or product tracking | Days, plus upkeep | High relevance, high fragility |
| Demographic | Self-reported or inferred | Varies | Usually less than you’d expect |
That last row deserves a note. Age, gender and location feel like natural segments and are usually weaker predictors than behaviour. Two people of the same age in the same city, one of whom bought from you three times and one of whom never opened an email, have nothing meaningful in common from your perspective. Behaviour beats demography almost always. The exception is genuine geography — time zones, shipping, local events, seasonality in opposite hemispheres — which is operationally necessary rather than strategically clever.
A Four-Week Implementation
Week one: build engagement recency segments only. Send your usual content to active and lapsing, hold back dormant and dead. Record the results.
Week two: add lifecycle stage. Write one genuinely different email for one-time buyers making the case for a second purchase. Just that one.
Week three: run the RFM analysis in a spreadsheet. Find the lapsed high-value cell. Write them something specific — ideally something that asks what changed, from a real person, expecting a reply.
Week four: build the preference centre and put it behind your unsubscribe link. Run the re-engagement sequence on the dead segment and then remove whoever doesn’t respond.
Four weeks, no new tooling, and you’ll have a programme better segmented than most businesses ten times your size. Then stop and run it for a quarter before adding anything.
Spam complaint thresholds cited from Google and Yahoo published sender requirements. Segmentation frameworks reflect established practice; performance figures are illustrative of the author’s experience rather than published benchmarks. This article contains no affiliate links.