Atlas

A SaaS company hired Atlas to figure out why their email engagement was uneven. Open rates were healthy on average, around 24%, but their user base behavior was strange. Some sends ran 35%, others ran 14%, with no obvious pattern in subject line or content quality.

Atlas pulled the engagement data and started looking for patterns. Within a few hours he noticed something odd. A cluster of users in one specific zip code (02139, which he immediately recognized as Cambridge, Massachusetts) were opening emails at exactly 7:14 AM Eastern, every Tuesday. Same minute. Different users, but the same timestamp.

He thought about it for a few minutes, then walked over to a window and laughed.

The Boston commuter rail's red line train arrived at Kendall Square station at 7:13 AM on weekday mornings. The cluster of users in 02139 were biotech and software workers commuting to Cambridge offices. They pulled out their phones during the 90-second window between arriving at the platform and reaching their building. Whatever email arrived in the inbox during that window got opened. Whatever didn't was usually missed for the rest of the day, because the working day swallowed their attention.

Atlas built a behavioral segment for the 02139 cluster and shifted their send time to 7:13 AM Tuesday. Open rates in that segment jumped from 22% to 41%. He found three other commuter clusters in the same dataset (San Francisco, Seattle, Atlanta) and applied similar adjustments.

The lesson he teaches now: the patterns are already in your data. You're just not looking.

---

A small DTC fashion brand was sending one weekly newsletter to their full list of 22,000 subscribers. Open rates were 12% and dropping. Unsubscribe rates were creeping up. They thought they needed to redesign the newsletter.

Atlas told them to keep the design. Just send three different versions instead.

Segment one was new customers, defined as anyone who'd made their first purchase in the last 90 days. The content focused on care-and-styling tips for the items they'd bought. Educational, not pushy.

Segment two was repeat buyers. Two or more orders, last purchase within the last six months. Content focused on new arrivals, restocks of items they'd shown interest in, and customer stories.

Segment three was lapsed buyers. No purchase in the last 6+ months. A different tone entirely, with a softer cadence and an occasional re-engagement offer.

Three sends, three segments, same week. Open rates after the change: new customers 28%, repeat buyers 35%, lapsed buyers 18%. Total revenue per send across all three segments doubled. Unsubscribe rates dropped by half.

The brand asked if they should split into more segments. Atlas told them to leave it at three. The first three are usually 80% of the value. Adding fourth and fifth segments is good, but it's diminishing returns and it's harder to maintain. The discipline is the discipline. Three good segments executed consistently beats fifteen sloppy ones.

The lesson he teaches now: the simplest segmentation often wins. Don't chase complexity. Find the three cuts that matter most for your audience and serve those well.

Atlas's intro:

Look at this. Two senders. Same product, same price point, same audience size. One sends one email per week to everyone. The other sends three emails per week, each to a different third of the list. Same total volume. The second sender's revenue per send is 2.4x the first. Why? Because the second sender stopped pretending all their subscribers were the same person. That's segmentation. Here's how it actually works.

---

A flat email list is a list where everyone gets the same message at the same time. It's the simplest mental model and it's the way most senders start. It's also the way most senders stay, even when their list grows past the point where it makes sense.

Segmentation is the practice of recognizing that the people on your list aren't one audience, they're many. Different subscribers signed up for different reasons. They live in different places. They engage at different rhythms. They're at different points in their relationship with your brand. Sending the same email to all of them treats them as if they were one person, and they're not.

When you segment, you split your list into groups based on something meaningful, and you send different content (or the same content at different times, or with different framing) to each group.

That's the simple version. The reason it matters is harder.

Why segmentation works

Three reasons.

First, relevance compounds. A subscriber who consistently receives emails that match their interests becomes more engaged over time. They open more. They click more. They reply more. They forward more. Each engagement signals to mailbox providers that this sender is welcome, and your deliverability rises with it. The opposite is also true. A subscriber who consistently receives emails that don't match their interests quietly disengages. They stop opening. They start ignoring. Eventually they either unsubscribe or, worse, mark as spam. Mailbox providers see the disengagement and tighten the screws.

Second, fewer wasted sends. When you send to everyone, every email costs you a bit of attention from subscribers who didn't need to see it. A new customer gets a "welcome back" email that confuses them. A lapsed buyer gets a "new arrival" email that's not enough to win them back. A loyal repeat customer gets a generic discount that they didn't need. Every misalignment is a small drain on attention. Multiplied across a list, it's a real cost.

Third, you stop competing with yourself. A flat list forces you to write content that works for everyone, which usually means content that works for no one in particular. The general send becomes the lowest common denominator. With segments, you can write content that's specifically right for one group, and then write something different for another group. The total quality of your sending goes up because each piece can be sharper.

The three axes everyone should think in

Segmentation has many possible dimensions. Three are foundational, and most senders should think in all three.

Geographic. Where the subscriber is. Time zone matters more than country here for most senders. A subscriber in Tokyo opens email at a different real-clock time than a subscriber in Detroit. If you send at 9 AM Eastern, you've reached the Detroit subscriber at a useful moment and the Tokyo subscriber at midnight. Time-zone-segmented sending recovers a meaningful chunk of attention you'd otherwise lose.

Behavioral. What the subscriber has actually done. Opens, clicks, purchases, page visits, abandoned carts, replied-to emails. Past behavior predicts future behavior better than any demographic. Someone who's clicked on three product-update emails wants product updates. Someone who's only ever clicked on storytelling content wants stories. The data already tells you. Segmentation is the act of listening.

Lifecycle. Where the subscriber is in their relationship with your brand. New (first 90 days). Engaged (regular interaction). At-risk (interaction declining). Lapsed (gone quiet). Each lifecycle stage needs different messaging. New subscribers need orientation. Engaged subscribers need value. At-risk subscribers need re-engagement. Lapsed subscribers need either a win-back or a graceful goodbye.

Most senders use one of these axes. Many use none. The best use all three at once, as overlapping layers.

What segmentation isn't

Segmentation isn't personalization. They're related but different. Personalization is about making a single send feel tailored to one person ("Hi [first name], here's a product based on what you bought last"). Segmentation is about making a single send go to a defined group of people instead of everyone. You can do segmentation without personalization. You can do personalization without segmentation. Doing both well is where the real performance lives.

Segmentation isn't slicing the list into a hundred tiny pieces. The biggest mistake I see is over-segmenting. A sender will read about segmentation, get excited, and try to build twenty-three segments based on some clever combination of behaviors and demographics. Then they can't maintain it. The segments drift. Some get sent to with the wrong content because the segment definition was off. Others get neglected because nobody has time to write content for a 47-person micro-segment. The whole effort collapses under its own weight. Don't do that.

Segmentation isn't a one-time setup. Subscribers move between segments constantly. A new customer becomes engaged. An engaged subscriber becomes at-risk. A lapsed buyer makes a purchase and re-enters the engaged segment. The segments themselves need maintenance. The cadence I recommend is a quarterly audit. You go in, look at the segments, fix what's drifted, kill what's not earning its keep, and add new ones if the data is showing patterns you missed.

How to start

If you've never segmented before, here's the cheap way to start.

Pick three segments. Not ten. Not twenty. Three. The classic three are: New (signed up in the last 90 days), Engaged (clicked or opened in the last 90 days, not in New), and Lapsed (no engagement in 90+ days).

Send three versions of your next campaign. Or three different campaigns. The New segment gets onboarding-flavored content. The Engaged segment gets your regular newsletter. The Lapsed segment gets a soft re-engagement attempt or skips this round entirely.

Measure separately. Don't just look at "open rate of campaign X." Look at "open rate of New cohort, Engaged cohort, Lapsed cohort." You'll see distinct patterns. The Engaged segment will outperform the flat-list baseline. The New segment will outperform too. The Lapsed segment will underperform, which is the signal that they need a different relationship with you (or none).

Do this for one full quarter. Don't add new segments yet. Don't get fancy. Just run the three-segment cut for ninety days and see what changes.

By the end of the quarter, you'll either have gotten more signal than your flat list ever gave you, or you'll find that your specific list doesn't benefit much from segmentation (which is rare but possible). Either way, you'll know.

The ship has eighty-seven different reasons to be aboard. Send eighty-seven different emails if you can. If you can't, send three. Three is infinitely more than one.

- Atlas

Atlas's intro:

Every subscriber on your list is at a specific point in the relationship with your brand. They're not all in the same place. The new subscriber who signed up yesterday is not the same as the loyal customer who's been with you for three years, who is not the same as the person who hasn't opened anything in eight months. Treating them as if they were the same is the most common segmentation mistake there is. Here's the lifecycle map.

---

A subscriber's relationship with your brand moves through stages. The exact boundaries vary by industry and brand, but the four-stage model is the cleanest one I've used and it works for almost any sender. Memorize the four stages and the transitions between them, and you have most of segmentation handled.

Stage 1: New

A New subscriber is anyone who joined the list in the last 90 days, regardless of engagement so far.

What's true about them: they just chose you. They're paying more attention than they ever will again. They're forming first impressions about what your emails are worth. They have not yet developed inbox-blindness for your sender name. They are also the most likely to unsubscribe or complain if your early sends don't match what they signed up for.

What to send: orientation content. Welcome series. Educational pieces that explain who you are and what they get. Genuine value early, not pitches. The first three weeks of the relationship determine the next three years. Welcome them properly (Grant wrote a whole article on this).

How to detect: signup date within the last 90 days.

How long they stay: 90 days, then they auto-graduate to Engaged or At-Risk based on their behavior.

Stage 2: Engaged

An Engaged subscriber is anyone past the New stage who has opened or clicked an email in the last 90 days.

What's true about them: they're actively reading you. They want what you're sending. They're the lifeblood of the list. They're also the segment most likely to convert, recommend, or reply. They've decided you're worth their inbox space, and they're sustaining that decision.

What to send: your regular content cadence. The good stuff. Don't over-pitch them, but don't under-pitch either. They've earned the right to hear about your products, your stories, your offers. Treat them as the audience you're actually writing for.

How to detect: opened or clicked any email in the last 90 days, AND they're past the 90-day New window.

How long they stay: as long as they keep engaging. The moment 90 days passes without an open or click, they shift to At-Risk.

Stage 3: At-Risk

An At-Risk subscriber is anyone who hasn't engaged in 90 days but hasn't yet hit the threshold for Lapsed (usually 180 days).

What's true about them: they're drifting. Something changed. Maybe their interests shifted, maybe their inbox got crowded, maybe they got busy. They haven't formally left, but they've stopped paying attention. The longer they stay At-Risk without re-engaging, the lower the chance they'll ever come back.

What to send: a re-engagement attempt. One email, not a series. A clear note that says "we noticed you haven't been around. If you'd like to keep getting our emails, click here. If we don't hear back, we'll cut the cadence way down." Then mean it. The ones who click move back to Engaged. The ones who don't move toward Lapsed.

Some senders run a more elaborate "win-back" sequence on this segment. That works for some industries (especially DTC retail with discount mechanics) and fails for others. The single direct ask is the safer default.

How to detect: no engagement in 90+ days, but engagement at some point in the last 180 days.

How long they stay: about 90 days. By 180 days without engagement, they shift to Lapsed.

Stage 4: Lapsed

A Lapsed subscriber is anyone who hasn't engaged in 180+ days.

What's true about them: they're effectively gone. They didn't unsubscribe, but they stopped paying attention long enough that mailbox providers are now treating your sends to them as a sign of an unhealthy list. Continuing to send to them at full cadence damages your domain reputation and increases the chance that even your engaged subscribers' emails go to spam.

What to send: very little. Or nothing. Some senders run one final "we're going to stop sending unless you click here" message. Others move directly to suppression. The choice depends on your industry and risk tolerance.

What you don't do: keep blasting them with your regular cadence. Lapsed subscribers in the active list are deliverability damage, full stop.

How to detect: 180+ days since last engagement.

How they leave: through suppression. Once suppressed, they stay on a do-not-send list permanently. If they ever re-subscribe through a website form, they reset to New.

The transitions matter

The lifecycle isn't four buckets, it's a flow. Subscribers move between stages constantly. A purchase or a click can move someone from At-Risk back to Engaged. A long quiet stretch can move someone from Engaged to At-Risk to Lapsed. The transitions are where the work happens.

The mistakes I see most often:

Treating New as a static segment. Some senders define New as "anyone who signed up in the last 90 days" and then never let them age out of the segment, even after they've been on the list for two years. New is a stage, not a label. After 90 days, the subscriber is either Engaged or At-Risk. They're not New anymore.

Skipping the At-Risk stage. Some senders move directly from Engaged to Lapsed at 180 days, missing the 90-day window where re-engagement still works. The At-Risk window is your one shot to catch a drifting subscriber before they're gone.

Not actually suppressing Lapsed. Some senders identify Lapsed subscribers and then continue to send to them anyway because they're afraid to "lose" the addresses. That's not loss prevention, it's deliverability damage prevention prevention. Suppress them. Their addresses don't disappear. They go on a do-not-send list, available for future re-permission if they ever come back through a website form.

How to set this up

Most modern ESPs let you define segments based on engagement behavior. The work is mechanical:

New segment: subscribers whose signup date is within the last 90 days.

Engaged segment: subscribers who have opened or clicked any email in the last 90 days, AND signup is older than 90 days.

At-Risk segment: subscribers whose last engagement was 90-180 days ago.

Lapsed segment: subscribers whose last engagement was 180+ days ago, OR who have never engaged AND signed up more than 90 days ago.

Run those four segments as your foundation. Send your regular content to Engaged. Send onboarding content to New. Send a single re-engagement email to At-Risk every 30 days. Suppress Lapsed.

That's lifecycle segmentation. It's the most underrated cut in email marketing, and it's the easiest to set up. Most senders' biggest gains come from this single change. The four segments tell you where each subscriber is. The work is just making sure you treat each stage like it's actually different from the others.

The relationship has stages. Send to the stage, not to the list.

- Atlas

Atlas's intro:

A SaaS company hired me to figure out why their email engagement was uneven. Open rates around 24%, decent on average, but some sends ran 35% and others ran 14% with no obvious cause. Same subject lines. Same content quality. Same list. I spent three hours in their data and found a Boston commuter train. Then I found three more in San Francisco, Seattle, and Atlanta. Here's what I look for, and why most senders never see it.

---

The patterns in your subscriber data are real. They're not noise. They're not random. They reflect actual human behavior, and the behavior is the same kind of thing that's been happening for centuries (people commute, people lunch, people decompress on Sunday evenings) just digitized into timestamps.

Most senders never look for the patterns because they don't know what they're looking for. The patterns are easy to find once you know what shape they take.

The commuter pattern

The original. Open rates spike at specific minutes corresponding to mass-transit arrivals or commute boundaries. The Boston example: a cluster of users in zip code 02139 (Cambridge) all opening at 7:13-7:14 AM Eastern on weekdays. The 02139 zip code includes Kendall Square, which is the MIT/biotech corridor. The Boston commuter rail's red line train arrives at Kendall Square station at 7:13 AM. Those users were checking their phones during the 90-second walk from the platform to their offices. Whatever email arrived in their inbox during that window got opened. Anything that arrived later got buried in their working day.

The fix is obvious once you see it: shift the send time for that segment to 7:13 AM Tuesday. The send-time-segmented version of that segment ran 41% open rates against the 22% the same content was getting at the regular send time.

The commuter pattern isn't unique to Boston. I've found versions of it in:

- San Francisco: BART arrivals at Embarcadero and Montgomery stations, 8:34 and 8:46 AM Pacific - Seattle: Sounder train arrivals at King Street, 7:55 AM Pacific - Atlanta: MARTA red line arrivals at Lindbergh and Buckhead, 8:11 and 8:24 AM Eastern - London: Tube line peaks at 8:30 and 8:45 AM (less precise, more diffuse) - Tokyo: Yamanote line peaks at 8:00, 8:15, 8:30 AM (very precise)

The pattern shows up wherever there's a critical mass of subscribers in a specific transit-served zip code. You don't need to know transit schedules. You just need to look at the open-time distribution by zip code and see if there are tight clusters at specific minutes.

The lunch pattern

Open-rate spikes at 12:00, 12:15, 12:30, and 1:00 PM in the subscriber's local time zone. The cluster is wider than the commuter pattern and less precise, but it's reliable.

What it means: subscribers checking their phones during lunch break. The 12:00 spike is people who eat at exactly noon. The 12:15-12:30 spike is the people whose meetings ran over. The 1:00 spike is the second-shift lunch-takers and the late-eaters.

What to do with it: schedule sends to land just before the lunch window opens, around 11:45 local time. The email is sitting in the inbox when the subscriber pulls out their phone for lunch. You catch the wave instead of arriving too late.

The Sunday evening pattern

Open-rate spikes Sunday between 7 PM and 10 PM local time. This is the "Sunday scaries" inbox cleanup pattern. People sit down to mentally prepare for the work week, and one of the things they do is clear out the weekend's email backlog.

What it means: this is the only window where you can reach people on a Sunday with high engagement. Most senders don't send on Sundays at all because the volume is low. That's exactly why the inbox is uncrowded during this window. A well-timed Sunday-evening send can have outsized open rates because there's less competition for attention.

This doesn't work for every audience. B2B audiences resent Sunday-evening sends from vendors. B2C audiences are fine with them or even prefer them. Test before you commit.

The 2:30 PM slump

Open rates dip across most segments between 2 PM and 3 PM local time. This is the post-lunch energy drop. Subscribers are at their desks but not engaging well with anything that requires attention.

What to do: don't send anything important during this window. If your engagement-tracking shows a 2:30 dip, move sends earlier (to catch lunch) or later (to catch end-of-day). Don't fight the slump.

The end-of-quarter pattern (B2B specific)

For B2B audiences, open rates drop sharply in the last two weeks of every fiscal quarter. Subscribers are heads-down on quarter-end work. Marketing emails are noise during that window.

What to do: schedule major B2B campaigns in the first or middle of the quarter, not the last two weeks. If you must send during quarter-end, expect lower engagement and don't draw conclusions from it.

The hidden geographic clusters

Beyond major cities, there are smaller geographic patterns that show up in some lists. A cluster of subscribers in a specific industrial park. A cluster around a university campus. A cluster in a retirement community. Each cluster has its own rhythm.

The way to find them: pull a heat map of open-rate by zip code, weighted by subscriber density. The unusual zip codes (high or low engagement compared to baseline) are the ones to investigate. Sometimes the cluster is enough to justify a dedicated segment. Sometimes it's just interesting and not actionable. Either way, you learn something about your audience.

The personal pattern

The hardest patterns to find are the per-subscriber ones. Each individual subscriber has their own micro-rhythm. Some only open on weekends. Some only open on the third email in a sequence. Some only open if the subject line uses a specific word.

This is where send-time personalization tools earn their keep. Tools like Klaviyo's smart send time, Salesforce's Einstein, and others run per-subscriber send-time prediction. They learn each subscriber's pattern over time and send to each one at their personal best moment. The lift over batch sending is usually 8-12% in opens.

Per-subscriber send-time isn't free (the tools cost money, and they need 30-90 days of data to calibrate). But for senders at scale, the math works. Sending the same email to 50,000 subscribers at 50,000 individually-optimized moments outperforms sending all 50,000 at one universal best time.

What to do with all this

If you've never looked at your data this way, start with one thing: pull your open-rate heat map by hour-of-day and day-of-week. Just that. Then look at it. You'll see your audience's rhythm in the heat map. The peaks and valleys are the patterns to follow.

If you have more time, segment the heat map by zip code or country. The geographic clusters will jump out.

The patterns are already there. They've always been there. The data is keeping a record of every commute, every lunch break, every quiet Sunday evening, every end-of-quarter slump, every personal rhythm. Look at the record.

The cartographer's job is to chart the world that already exists. The patterns aren't invented. They're observed. Go observe.

- Atlas