Segmentation improves email results reliably, and most programmes over-complicate it. Twelve carefully defined segments maintained by nobody perform worse than three that are actually used.
These are ordered by return on the effort required.
1. Engagement, before anything else
If you do one thing, do this. Split the list by recency of engagement — opened or clicked in the last 30 days, last 90, last 180, and beyond.
Send your main campaigns to the engaged segments. Put the inactive ones into a re-engagement sequence and then suppress them. This single change usually lifts open rates substantially, cuts complaints, and improves inbox placement for everyone on the list, for reasons covered in email deliverability.
It feels like shrinking your reach. In practice reaching 20,000 engaged people beats being filtered when sending to 50,000.
2. Lifecycle stage
Where the contact sits in their relationship with you: new subscriber, trial user, active customer, lapsed customer, former customer.
The content each needs is completely different, and sending the same campaign to all of them guarantees most of it is irrelevant. A new subscriber needs orientation; an active customer needs depth; a lapsed one needs a reason to return.
This segmentation is usually straightforward to build because the data already exists in your product or CRM, and it maps naturally onto automated sequences rather than requiring manual campaigns.
3. Behaviour in the product or on the site
What the contact has actually done: pages visited, features used, content downloaded, a cart abandoned, a plan limit approached.
Behavioural triggers consistently outperform scheduled campaigns because they arrive when the subject is on the person’s mind. A message triggered by someone reading three pages about a specific capability is relevant in a way no calendar-based send can match.
Start with two or three high-value triggers rather than building a comprehensive system. The cart abandonment sequence and the plan-limit notification are usually the two with the clearest return.
4. Firmographics, for B2B
Company size, industry, role. Useful where your product serves genuinely different use cases for different segments, and less useful than most people assume.
The common mistake is segmenting by industry and then sending near-identical content with one word changed. If the message is not meaningfully different, the segment adds maintenance cost and nothing else.
Role-based segmentation tends to be more valuable than industry: a technical evaluator and a budget holder need genuinely different messages about the same product.
5. Stated preferences
Letting subscribers choose topics and frequency. This reduces unsubscribes and complaints because the alternative to “too much email” becomes “less email” rather than “no email”.
A preference centre needs to be short — three or four topic choices and two or three frequency options. Long preference forms are abandoned, and the defaults then determine everything anyway.
How many segments
For most programmes, three to six active segments is the sustainable range. The test is whether each segment receives content genuinely different from the others. If two segments always get the same email, they are one segment.
Over-segmentation has real costs: more content to produce, more to maintain as data changes, more ways for someone to fall into no segment at all, and small segments where results are statistically meaningless.
Building segments that stay correct
Three practical points.
Prefer dynamic to static. A segment defined by rules updates itself; a list exported once is wrong within a month.
Decide what happens to contacts in no segment. Every programme has them, and they are usually the ones who quietly stop receiving anything.
Keep the data clean at the source. Segmentation on a field that is 40% empty or inconsistently formatted produces segments that do not mean what you think. Fix collection before building rules on top of it.
Personalisation versus segmentation
These get conflated. Segmentation changes who receives which message. Personalisation changes the content within one message.
Personalisation is useful where the variable content is genuinely relevant — a product the person viewed, a usage figure from their account, the name of their company. It is counterproductive when it is cosmetic: inserting a first name into a subject line stopped impressing anyone a decade ago, and a broken merge field actively damages credibility.
The ranking is clear in most tests: relevant segmentation beats cosmetic personalisation by a wide margin.
Measuring whether segmentation worked
Compare performance per segment against the unsegmented baseline, and measure revenue per recipient rather than open rate alone. A segment with a lower open rate and higher conversion is the better segment.
Watch total output as well. Segmentation that halves engagement-weighted volume because the team cannot produce enough variants has not improved anything. The right number of segments is the number you can feed.
A re-engagement sequence that works
Suppressing inactive subscribers is correct, and giving them one proper chance first is worth the effort.
A three-message sequence over two weeks performs best. The first asks directly whether they still want to hear from you, with an obvious way to say yes and an equally obvious way to leave. The second offers your single best piece of content, on the assumption that irrelevance rather than annoyance caused the silence. The third states plainly that this is the last message unless they act.
Expect a low response — frequently under 5%. That is not failure. The people who respond are genuinely interested, and the rest were costing you deliverability for everyone else.
Suppress non-responders rather than deleting them, so a future import does not quietly put them back.
Segmentation without good data
Many lists have an email address and little else. Three ways to build segmentation from nothing.
Infer from behaviour. Which content someone clicks reveals their interests more reliably than anything they would tell you on a form.
Ask in the welcome sequence. One question, one click, recorded as a tag. A single well-placed question early in the relationship produces better data than a long preference form later.
Use the signup source. What someone subscribed for is a strong signal about what they want next, and it requires no additional collection at all.
All three are more reliable than purchased enrichment data, which is frequently stale and sometimes simply wrong.
A reasonable starting configuration
For a programme with no segmentation at all, four segments are enough to begin and can be built in an afternoon.
Engaged, meaning opened or clicked in the last 90 days — this receives the main programme. Dormant, 90 to 180 days — this receives a lighter cadence and the re-engagement sequence. Inactive, beyond 180 days — suppressed after one final attempt. And new, in the first 30 days — receiving the welcome sequence rather than regular campaigns.
Add lifecycle and behavioural segments once these four are working. Starting with a sophisticated structure that nobody maintains is how segmentation projects quietly stop.
Frequently asked questions
What is the single most valuable segment?
Engagement recency. It improves results and deliverability simultaneously and requires no additional data.
Should inactive subscribers be deleted?
Suppress them from regular sending after a re-engagement attempt. Retain the record for suppression purposes rather than deleting it outright, so they are not re-added by a later import.
How often should segments be reviewed?
Quarterly. Segment definitions drift as the product and audience change, and a rule written two years ago may now capture almost nobody.
Does segmentation reduce total sends?
Usually yes, and that is generally the point. Fewer, more relevant messages typically produce more revenue than a larger volume of generic ones.
