What luxury ecommerce taught me about customer segmentation
Generated by Klaviyo AI
Nanxi Fan, director of paid media and CRM at luxury publishing maison Assouline, explains why standard ecommerce segmentation advice (like a 60-90 day win-back flow) didn't fit her brand once she found their average time between orders is 8 months. She walks through the five ways Assouline now segments customers — browse behavior, product category, gift buyers vs. self-purchasers, RFM as a percentage of the database rather than a fixed day count, and tone — to avoid mislabeling engaged, high-AOV customers as lapsed.
- Browse behavior beats purchase recency: Assouline layers engagement signals — repeat product views, email opens/clicks, on-site activity — against RFM tiers, since long buying cycles mean recency alone can't flag who's actually at risk.
- Category segmentation comes before cross-selling: Because different product categories (e.g., design books vs. travel books) attract genuinely different collector types, Assouline builds affinity groups by category and checks cross-category overlap (12% of design-book buyers also buy travel books) before cross-selling.
- RFM should track as a percentage of the database, not a fixed day count: Assouline uses Klaviyo's Marketing Analytics to rank customers by percentage within their own distribution rather than a static "days since last order" threshold, so "At risk" and "Champions" segments shift automatically with the business.
I'm the director of paid media and CRM at Assouline, a French luxury publishing maison founded in 1994. We're known for collectible, art-directed books on culture, fashion, art, travel, and design, and we've grown from books into boutiques and library-inspired spaces in major cities worldwide.
When I started building out our lifecycle programs, I ran into a problem fast: none of the standard segmentation advice fit us.
Every template, every peer conversation, every blog post said the same thing: set up a win-back flow at 60 or 90 days. That's the industry default. So I checked our own data.
Our average time between orders is 8 months.
That one number changes everything. If I'd followed the standard playbook, I would have flagged someone as lapsed at the exact moment they were enjoying their purchase: displaying a $1,000-plus art book at home, giving it as a gift, telling friends about it. They hadn't churned, but they were on a longer clock.
If your buyers are collectors, executives, or gift-givers rather than habitual repeat purchasers, purchase recency alone won't tell you who's actually at risk. And if your products aren’t consumable goods like beauty products, but rather a piece that lives with the shopper for decades, you need to think about purchasing behavior differently. Here's our approach.
Segment on browse behavior, not just purchase recency
Because our buying cycle is long, we treat engagement signals as more valuable than most brands would. Email opens, clicks, product views, and on-site activity all feed into our engagement tiers, separate from how recently someone bought.
We also layer this against recency, frequency, and monetary value (RFM) tiers, Klaviyo's model for ranking customers by purchase behavior.
- Repeat product views: Someone who's looked at the same piece 3x is showing real intent, not idle browsing.
- Overlap with an RFM tier: Someone in our “Champions” segment who's been browsing recently gets treated differently than someone in that same segment who's gone quiet.
We build these as layered segments in Klaviyo: someone can qualify as an RFM “Champion” and separately qualify, or not, based on recent browse depth. The overlap is often more useful than either signal alone.
Segment by category before you cross-sell
Most DTC brands can assume a customer who buys one product might buy almost anything else in the catalog, but we can't, and I think that’s probably true for a lot of luxury brands.
Someone collecting design books is probably not also buying books about boxing. Our categories attract genuinely different types of collectors, from interior designers building out a home to executives who want a subtle signaling of prestige.
So we use Klaviyo segmentation to build affinity groups by category, and we look at cross-category behavior before we cross-sell. For example, 12% of people who buy design books also buy travel books. That tells us where a design buyer might welcome a new travel title, and it tells us not to send that same person a promotion for something outside their interests just because they're active on our list.
This keeps our sends personalized instead of blasted. Customers get content that's actually relevant to them, and we avoid the fatigue that comes from over-messaging an audience that's only interested in a fraction of what we sell.
“Building affinity groups by category is the difference between a relevant recommendation and a random one,” says Adam Hutton, senior manager of global ecommerce at Benchmade Knife Company, a premium American manufacturer of high-performance knives.
For example, “understanding the relationship between customers who buy hunting vs. fishing knives lets us cross-sell in a way that actually matches their lifestyle, not just their last purchase,” Hutton explains. “Layering in website engagement data sharpens that signal even further, so we know not just what category someone’s in, but how genuinely interested they are in it.”
Building affinity groups by category is the difference between a relevant recommendation and a random one.
Segment gift buyers separately from self-purchasers
A large share of our orders are gifts for housewarmings, weddings, graduations, and birthdays. If we lumped gift buyers in with self-purchasers, we'd miscategorize a big part of our customer base and send the wrong message to the wrong person.
We flag a gift order using signals like:
- Gift message added at check-out
- Shipping address different from billing address
- Timing that lines up with a gift-heavy season
- Gift wrapping added to the order
- First-time purchase of a giftable item, like a boxed set
Once we identify a gift order, we hold it out of our standard self-purchase flows and route it into a dedicated post-gift sequence instead. That sequence gently invites the gifter to come back and buy something for themselves. Then, around the one-year mark when the gift anniversary gives us a natural reason to reach out again, we re-engage.
Track RFM as a percentage of your database
We rely on Marketing Analytics, and I think it's one of the more underrated tools for a brand like ours. The reason it works for us: it ranks customers by percentage within our own database, not by a fixed number of days.
That matters because purchase cadence shifts. Our average time between orders might be 8 months this year and something different next year, depending on the season or on how our average order value (AOV) moves.
Tracking that ourselves would take real analysts quite a bit of time every quarter. Advanced personalization updates our RFM analysis against our current customer base in real time, so we can put that time elsewhere.
In practice, we use the “At risk” and “Needs attention” segments as our entry point into reactivation, rather than triggering off a hard day count like "6 months since last order."
And on the other end, our highest-value customers (what Klaviyo's RFM model calls “Champions” and “Loyalists,” are simply the top slice of our own distribution. That slice rises and falls automatically with our business.
Match your tone to your segmentation
Tone is the other half of segmentation. None of this works if the message itself feels wrong. We follow brand voice guidelines on every send:
- Never rush the customer.
- Never push.
- Never sound like a sale.
The goal is for our communications to feel like client relations, not marketing automation. Getting the audience and the timing right only matters if what lands in someone’s inbox actually sounds like it was meant for them.
The takeaway for high-consideration brands
If your AOV is high and your purchase cycle is long, the standard playbook will mislabel your best customers as lapsed. Before you build another win-back flow at a fixed day count, ask:
- What does my actual average time between orders, not the industry default, look like?
- What behavior, besides a purchase, tells me someone is still engaged?
- Do my categories or products create genuinely different types of buyers, and am I segmenting for that?
- Are gift buyers hiding inside my self-purchase numbers?
- Am I using a fixed day count when a percentage-based model would adapt better?
Client quality matters more to us than client volume. Getting our segmentation right is how we act on that, instead of just saying it.
Related content
