Nobody serious in e-commerce still argues about whether lifecycle marketing works. Lifecycle marketing is the automated emails and texts a brand sends, such as a welcome series. In the controlled tests my agency YOCTO has run, lifecycle marketing has reliably cut the cost of winning a new customer by 15% to 20% within seven days.
In e-commerce, we know the channel works. We do not know how much of the revenue it gets credited with, is actually a result of that marketing. That is a limit of attribution, the method platforms use to decide which message gets credit for a sale.
ATTRIBUTION ANSWERS THE WRONG QUESTION
Attribution only addresses what touched this order most recently. It cannot answer what matters most: Would the order have happened anyway?
Take a browse abandonment email, sent to someone who looked at a product and left. If they buy within the window (the few days after the email is sent) in which purchases count toward it, the email gets the credit. Some of that credit is earned. Some is interception: The customer was returning anyway, and your email was the last thing they touched. Separating the two requires knowing what the customer would have done without the email, and no platform records that.
Klaviyo, a leading email and SMS platform for online stores, sets that window at five days, and a purchase counts “if someone opens or clicks, or if Apple Mail Privacy Protection automatically opens the email.” So a machine opening a message on the customer’s behalf is enough to claim a purchase. Every platform works roughly this way. The number on your dashboard is the maximum the channel could have earned, not what it did earn.
The gap between credited and caused usually flatters the channel. Researchers had a retailer mail 20,000 existing customers 12 or 17 catalogs over eight months. For the retailer’s best customers, the short-run increase in demand was “almost entirely offset by a reduction in future demand.” The mailings did not create purchases, only pulled forward ones those customers would have made anyway.
THE SAME BLIND SPOT, IN REVERSE
We tested a new email for a subscription brand, sent when a subscriber (a customer on a recurring order) tries to cancel. The email let them pause or skip a delivery instead. Some of the customers who tried to cancel received the test email; the rest got the standard email. On day one, the first group’s cancellations were about 60% lower. On any dashboard, that looks like the best result.
By day 30, cancellation rates in the two groups were equal. By the end of the test, lifetime value in the group that got the new email was slightly lower. We had not prevented cancellations. We had postponed them, and while those customers were paused, nobody was winning them back. A save that only moves the date is not a save.
A purchase that was coming anyway gets booked as revenue you produced. A cancellation that was coming anyway gets booked as a save you made. Attribution has no way to represent the word “anyway.” And it looks in days, while a subscription customer’s value plays out over months.
HOW YOU MEASURE DECIDES HOW YOUR TEAM BEHAVES
If a message gets credit for every purchase that follows an open, the sensible strategy is to send more. Every send collects credit whether it caused the purchases or not, and no dashboard ever tells you to stop.
Email teams send two kinds of messages: campaigns (a one-off sends a person plans), and flows (automated sequences that run on their own). If the person running your program is paid on campaign revenue, the sensible strategy is to neglect flows. Flows run whether or not that person shows up. Campaigns are visible effort. So they run campaigns, and the flows that continue earning in the background get whatever time is left.
Here’s my favorite example. A reliable way to grow flow revenue is to remove the discount code from the signup popup and make new subscribers open a welcome email to get it. Flow revenue rises immediately. Nothing about the customer, the offer, or the business has changed. Only who gets the credit.
None of the people planning this are cheating. They are doing what the numbers they are judged on reward. The numbers are the problem.
MEASURE WHAT YOU CAUSED
A holdout test answers the question attribution cannot. One group gets the email or text; a randomly chosen group does not. The difference between them is what the marketing caused. It needs no special software or data scientist.
Klaviyo has a built-in version, but it requires at least 400,000 contacts and tests the whole program at once, not the one flow your team is arguing about.
Any company can build its own version in an afternoon. Pick the flow you are least sure about; the series after a first purchase is a good start. Let 85% of customers through and hold 15% out, so they receive nothing. Label both groups, then leave it alone.
When the test ends, export both groups and compare them on what matters: how many cancelled at 30, 60, and 90 days, how many bought again, revenue per customer, lifetime value. That is a spreadsheet, not an analytics project.
The discipline is in how long you wait. Measure over the life of the behavior you are trying to change, not the five days the platform counts. For a subscription welcome series, one billing cycle is the minimum; three is better.
WHAT TO EXPECT
Some of your flows will come back flat. That is not a testing failure, but the first honest number your program has produced. It frees up the hours your team spends on a sequence that changes nothing.
Ask your team: If you switched off your best-performing flow tomorrow, do you know what would change?
If the answer is a revenue figure from a dashboard, you do not know what would change. You know what would stop being credited.
George Kapernaros is the founder and CEO of YOCTO.