Targeted email marketing is one of the few channels where the theory and the practice have drifted almost completely apart. Every platform sells segmentation. Every agency deck has a slide about "the right message to the right person at the right time." And yet the median ecommerce or B2B programme still runs on a handful of broadcast sends to the full list, with a suppression rule for people who unsubscribed and not much else.
The gap is not knowledge. Most marketers can describe RFM segmentation from memory. The gap is that building real segments costs engineering time, exposes data problems nobody wants to own, and produces results that take two quarters to become visible in revenue reporting. Broadcast sends produce a number this afternoon.
This guide covers which segments are actually worth building, in what order, what the data requirements really are, and how to tell whether the targeting is producing lift or just reshuffling revenue you would have earned anyway.
What "targeted" means in practice
Targeting is not personalisation. Putting a first name in a subject line is a merge tag, and its measured effect on open rates has been somewhere between negligible and mildly negative for the better part of a decade. Real targeting changes one of three things:
- Who receives the send at all. The most valuable targeting decision is usually exclusion, not inclusion.
- When the send fires. A trigger tied to a behaviour beats a calendar date almost every time.
- What the offer is. Not the copy tone, the offer: category, price point, urgency, incentive depth.
Everything else — dynamic hero images, weather-based subject lines, countdown timers — is decoration on top of those three. It can help at the margin. It cannot rescue a programme that sends the same offer to everyone on Tuesday.
The practical test for any segment you are considering: would you write a materially different email for this group? If the answer is no, it is not a segment, it is a filter, and it will add reporting complexity without adding revenue.
The four segments that earn their build cost
Most programmes need four working segments before anything more exotic is worth attempting. In rough order of return per hour of setup:
1. Engagement tiers
Split the list by recency of engagement: opened or clicked in the last 30 days, 90 days, 180 days, and beyond. This is the single highest-leverage segmentation available, and it is available on day one because the data is already in your ESP.
The reason it matters is deliverability, not relevance. Mailbox providers weight recent positive engagement heavily when deciding whether your mail reaches the inbox or the promotions tab or nowhere. Sending to a 400,000-address list where 340,000 have not opened anything in a year does not produce a small amount of extra revenue at zero cost. It degrades the sending reputation that determines whether the 60,000 engaged addresses see anything at all.
Cut the disengaged cohort out of routine sends. Run them through a short, honest reactivation sequence two or three times a year, then suppress permanently. Programmes that do this typically see total list size fall by a third and total email revenue rise, which is a difficult conversation to have with anyone who reports on list growth as a KPI.
2. Purchase or lifecycle stage
For ecommerce: never purchased, one-time buyer, repeat buyer, lapsed. For B2B: no trial, trial active, trial expired, customer, churned.
These four or five states justify genuinely different emails. A first-time buyer needs product education and a reason to come back. A repeat buyer does not need a 15% welcome discount and will happily take it if you send one, which is how discount programmes quietly destroy margin. A lapsed customer needs a reason to reconsider, and the reason is rarely price.
The one-time-to-repeat transition is where most of the value sits. The economics are worth working out before you design the campaign, because the amount you can profitably spend to convert a second purchase depends entirely on what a repeat buyer is worth over their lifetime — run the numbers through an LTV calculator and compare against your customer acquisition cost before setting incentive depth. Teams routinely offer a 20% second-purchase discount without checking whether the second order is even profitable at that rate.
3. Category or product affinity
Derived from browse and purchase history: which product category has this person actually engaged with? For B2B, the equivalent is which product line, use case, or content topic they have consumed.
This is where the data work starts. Affinity requires either an ESP with native ecommerce integration that tracks browse events, or a pipeline that pushes behavioural data from your site into the ESP. It is the first segment on this list that will require someone technical to build, and it is worth it: category-targeted sends consistently outperform generic ones by a wide enough margin to survive most attribution disputes.
Keep the taxonomy coarse. Three to six affinity buckets that map to how you actually merchandise is better than forty micro-categories that each contain 200 people and cannot support a send.
4. Value tier
Top decile by spend, middle, bottom. High-value customers should not be receiving the same discount ladder as bargain hunters, and they should generally receive fewer, better emails rather than more.
This segment is the one most often skipped and most often regretted. The people who spend the most are the ones most easily annoyed into unsubscribing, and losing them costs more than any campaign gains.
Behavioural triggers: where the revenue concentrates
Segments define audiences. Triggers define timing, and timing is where per-email revenue is highest by an order of magnitude.
The standard set, roughly in order of value:
Abandoned cart or abandoned checkout. Still the highest revenue-per-send in most ecommerce programmes. Fire the first message within an hour, the second at 24 hours, and stop. Adding a discount to email one trains customers to abandon deliberately; hold any incentive back to the third message, if you use one at all.
Browse abandonment. Lower intent, lower conversion, much larger addressable volume. Worth building after cart abandonment is working, not before.
Post-purchase sequences. Order confirmation, shipping, delivery, and then a genuine follow-up at the point the product has been in use long enough to have an opinion. Transactional emails get opened at rates campaign sends never approach, and most brands waste that attention on a bare receipt.
Replenishment. If your product has a predictable consumption cycle, a trigger timed to roughly 70% of the median reorder interval is close to free revenue. Calculating that interval per category rather than as a single site-wide number is what separates a good replenishment programme from an irritating one.
Price drop and back-in-stock. Requires wishlist or browse data plus inventory integration. High conversion, low volume, and largely self-selecting.
Win-back. Fires at a defined lapse threshold — typically 1.5 to 2 times the median repeat interval. Design it to be answerable: ask whether they want to keep hearing from you, and honour the answer.
The build order matters. Cart abandonment and post-purchase cover most of the achievable trigger revenue for a fraction of the total effort. Teams that start with a fourteen-branch lifecycle map usually ship nothing. There is more detail on sequencing and structure in our guide to automated email campaign strategies.
The data you actually need
The honest prerequisite list is shorter than vendors imply and longer than most teams have:
- Identity resolution. A stable customer ID that links email address, order history, and on-site behaviour. Guest checkout and multiple email addresses per household break this constantly.
- Event data in the ESP. Product views, add-to-cart, purchases, with product and category attributes attached. Most platform-native integrations (Shopify, BigCommerce, WooCommerce) handle this adequately out of the box.
- Consent state per channel. Separately tracked for email and SMS, with a timestamp and source. This is a legal requirement in most Tier-1 markets and a practical requirement for suppression logic.
- A profile schema you can query. If you cannot express "bought category X, has not bought since date Y, engaged in last 60 days" as a single audience definition, your segmentation ceiling is set by your platform, not your strategy.
That last point is the one that determines platform choice more than pricing does. The practical differences in how far segmentation can go across the major platforms are covered in our Klaviyo vs Mailchimp comparison.
Measuring targeted email marketing honestly
The measurement problem specific to targeted sending is cannibalisation. A segment that converts at 6% against a list average of 2% looks like a triumph, but if that segment is simply the people who were going to buy anyway, the targeting has reallocated revenue rather than created it.
Three practices keep the numbers honest:
Holdout groups. Withhold 5–10% of each triggered flow's eligible audience and measure their conversion rate over the same window. This is the only method that reliably separates incremental revenue from revenue you would have received regardless. Almost nobody does it, and the ones who do are usually unpleasantly surprised by the first result.
Revenue per recipient, not open rate. Apple Mail Privacy Protection made open rates structurally unreliable in 2021, and they have not recovered as a metric. Revenue per recipient normalises across segments of different sizes and cannot be inflated by sending more often.
Per-segment unsubscribe and complaint rates. A campaign that lifts revenue 8% while raising the complaint rate is borrowing from next year. Track it at segment level, because aggregate complaint rates hide the specific cohort you are irritating.
Conversion improvements from targeting compound with on-site improvements, so it is worth modelling both together rather than in isolation — a conversion rate calculator is enough for a first pass, and the site-side work is covered in our guide to ecommerce conversion rate optimisation.
A realistic 90-day build order
For a programme starting from broadcast-only sending:
- Weeks 1–2. Engagement tiers and suppression. Immediate deliverability benefit, no engineering required.
- Weeks 3–6. Cart abandonment and post-purchase sequences. Verify the event data is firing correctly before writing any copy.
- Weeks 7–10. Lifecycle stage segmentation applied to the regular campaign calendar. First-time buyer versus repeat buyer as the minimum split.
- Weeks 11–13. Category affinity, plus a holdout group on the highest-volume flow so the next quarter's decisions have real data behind them.
Anything beyond that — predictive churn scores, send-time optimisation, AI-generated subject lines — should wait until these are running and measured. They are refinements on a working system and worthless without one.
FAQ
How small can a segment be before it stops being worth sending to? As a working rule, below about 500 recipients the statistical noise makes performance unreadable, and below roughly 100 the send is unlikely to justify the setup time unless the customers are individually high-value. Enterprise B2B is the obvious exception, where a 40-account segment can be worth more than the rest of the list combined.
Does targeted email marketing reduce total send volume? Usually yes, and that is generally the point. Programmes that adopt engagement-based suppression typically send 30–50% fewer emails while holding or increasing revenue, because deliverability improves and the remaining sends land in the inbox rather than the spam folder.
How often should segments be rebuilt? Dynamic segments that re-evaluate on every send are the default in modern platforms and should be preferred. Static lists exported once and reused go stale within weeks and are a common cause of sending to people who have since unsubscribed or purchased.
Is discounting necessary for reactivation campaigns? No, and defaulting to it is expensive. A meaningful proportion of lapsed customers respond to a preference update or a genuinely new product rather than a price cut. Test the non-discount version first, since it costs nothing to find out and the discount is always available afterwards.