A repurchase looks like success on every dashboard that tracks it. Someone bought a serum in January. They bought it again in March. The analytics team calls that a win for the recommendation engine.
Here's what the dashboard doesn't show. That same person stopped using the serum in February. Their skin broke out, so they switched to something else for three weeks. They only came back to the original serum once the reaction settled down.
None of that shows up as a data point anywhere.
A dashboard sees two purchases. It never sees the three weeks in between.
What Repeat Purchase Actually Measures
Repurchase rate gets treated as the main proxy for a recommendation being right, largely because it's the easiest thing to measure. Someone buys again, so the logic goes, the product must be working. That reasoning skips over a fairly ordinary reality of skincare, where people repurchase things out of habit, out of not wanting to waste an unopened bottle, or simply because switching products feels like more effort than continuing with what's familiar.
"We had customers reordering a moisturiser for four months before finally admitting in a review that they'd stopped liking it after month one," said an operations lead at a skincare brand reviewing churn patterns.
A repeat order confirms someone clicked "buy" twice. It says very little about whether their skin actually improved, stayed the same, or got quietly worse in ways they didn't bother reporting.
Clicking "buy" again isn't the same as skin getting better.
The Silence Problem
Most skincare brands never hear from the majority of customers a recommendation didn't work for. A product causing mild irritation doesn't usually generate a bad review. It generates a returned product with no explanation, or worse, no return at all, just a customer who quietly stops reordering and moves to a competitor without saying why.
Silence gets recorded as neutral data. It rarely is.
"Only about four percent of dissatisfied customers ever contact a brand directly. The rest just leave," a customer research analyst noted in a study on skincare churn behaviour.
This creates a strange blind spot. A recommendation engine trained on purchase and return data alone is really being trained on the loud minority, the people angry enough to write in or organised enough to send something back. Everyone else, the quiet majority whose skin reacted badly but who couldn't be bothered to explain why, disappears from the dataset entirely.
Why Surveys Don't Fix This Either
The instinct to solve this with a post-purchase survey runs into its own problem. Surveys sent two weeks after a purchase catch people mid-adjustment, before most actives have had time to show a real effect. Surveys sent three months later catch people who've often forgotten which product caused which reaction, especially if they're layering four or five things in a routine at once.
"By the time we asked customers how a product worked, most of them were already using something else and couldn't clearly separate the two experiences," a skincare brand's CX manager said, describing a failed feedback survey rollout.
A single data point collected at one moment in time can't capture something that unfolds over weeks, changes with weather, stress, or a new product introduced halfway through.
One survey, sent once, is a snapshot. Skin doesn't work in snapshots.
What Actually Closing This Gap Requires
Closing this gap needs more than a bigger spreadsheet. It needs a system built to ask about outcomes at multiple points, tied to when actives are supposed to start working rather than an arbitrary two-week mark, and it needs to treat silence itself as a signal worth investigating rather than defaulting to "no news is good news."
This is roughly the ground Crea8 has been building on. Instead of relying on repurchase alone as a stand-in for success, Crea8's approach checks in at intervals mapped to how actives actually behave on skin, and treats a customer going quiet as a prompt to ask a question, not a reason to assume everything is fine.
A recommendation engine that only listens to complaints is only ever getting half the story.
Building a Feedback Loop That Actually Closes
The harder, less flattering truth here is that most personalisation data available today measures behaviour, not outcome. It tells a brand what someone did and why. It rarely tells them how their skin actually responded, which is the only thing that was ever supposed to matter.
Crea8's position on this has been to treat that gap as the actual product to solve, building feedback loops that go looking for the truth, grounded in formulation science, rather than waiting for it to arrive as a complaint.
A recommendation is only as good as a brand's ability to find out, later, whether it worked. Right now, most of that ability doesn't exist yet.