The Personalization Paradox
Case Study: How a Retailer’s Personalization Engine Won Every Metric and Lost Its Best Customers
Author: IMB Editorial Team
IMB Journal – International Marketing Board
Volume 1 | Issue 5
May 2026
How a Retailer’s Personalization Engine Won Every Metric and Lost Its Best Customers
The following case draws on a composite of client engagements handled by IMB advisors. Identifying details have been changed to protect confidentiality, but the pattern is one we’ve now seen across multiple direct-to-consumer retail brands.
The Setup
The company, which we’ll call Halcourt, sold mid-range apparel through its own site and app. Three years in, Halcourt invested heavily in a personalization platform that tracked browsing behavior, purchase history, and even how long a shopper lingered on a product photo before scrolling past it. Every email, every push notification, and every homepage banner was assembled individually for each customer based on that data.
The results, by the numbers the marketing team reported monthly, were outstanding. Email click-through rates nearly doubled within two quarters. Push notification open rates climbed well above industry benchmarks. Abandoned cart recovery, in particular, became something of a showcase metric, recovering close to a third of carts that would previously have been lost entirely.
What the Number Was Hiding
The team measuring these wins was, understandably, looking at engagement and conversion. Nobody on the marketing side was responsible for tracking a slower-moving number: repeat purchase rate among customers who had been with Halcourt for more than two years, the group that had historically driven a disproportionate share of revenue.
That number had been quietly declining for nearly a year by the time anyone noticed, and it was hiding in plain sight because it wasn’t on the same weekly dashboard as click-through rate. When customer service finally flagged an uptick in complaints, the complaints weren’t about product quality or pricing. They were almost all some version of the same sentiment: customers said they felt like the brand knew too much about them and wouldn’t leave them alone. One long-time customer, in an exit survey after unsubscribing, described getting a push notification about the exact pair of shoes she’d looked at for eleven seconds the night before, followed by an email an hour later, followed by a retargeted ad on a completely different app that afternoon.
The Diagnosis Came From the Wrong Team, at the Wrong Time
The initial instinct, once the repeat purchase decline was finally noticed, was to assume the personalization wasn’t targeted enough yet, and to invest further in refining the algorithm. It took a customer research project, commissioned almost as an afterthought, to reveal that the opposite was true. Long-tenured customers weren’t disengaging because the recommendations were poorly matched to their taste. They were disengaging because the volume and persistence of the targeting had crossed from helpful into invasive, and it had crossed that line specifically for the customers Halcourt had the most data on: its best, longest-standing shoppers.
The customers most rewarded by the personalization engine’s own success metrics were, in practice, the ones being pushed away fastest.
What Changed
Halcourt introduced a second reporting line alongside its engagement dashboard: a quarterly relationship health score for customers above a certain tenure and spend threshold, built from unsubscribe rates, notification opt-outs, and a simple survey question asking how customers felt about the frequency of communication from the brand. Marketing leadership agreed to cap the number of personalized touches a single customer could receive across all channels within a rolling seven-day window, regardless of how well any individual touch was predicted to perform.
Short-term engagement metrics dipped modestly after the cap was introduced. Click-through rates fell by roughly a tenth. Within two quarters, however, the decline in repeat purchase rate among long-tenured customers reversed, and unsubscribe rates from that segment dropped to their lowest level in over a year.
The Lesson
Halcourt’s personalization engine wasn’t broken. It was doing exactly what it had been built and measured to do: maximize the click. The problem was that nobody had built a way to see the cost that success was creating, because that cost showed up on a different customer, on a different timeline, and on a metric nobody was watching every week.
Part of a three-part series on personalization and trust. Next: a short insight on the one question that keeps a personalization strategy from quietly becoming surveillance.