Size recommendation engines: when to show them and when to stay quiet
Size recommendation engines raise conversion when they appear at the moment of sizing doubt and stay out of the way otherwise. Showing a full measurement quiz to a returning customer who already knows their size adds friction; hiding guidance from a first-time buyer looking at an unfamiliar cut loses the sale. The rule is simple: personalize the visibility of sizing help based on what you know about the shopper, the product, and the purchase history.
The sizing doubt moment
Every apparel purchase has a moment where the shopper asks, silently, "will this fit me?" It arrives at different points for different people: on the size selector for unfamiliar brands, on the product description for unusual cuts, after a return for a burned buyer. The size recommender's job is to be present exactly then, with an answer, not a quiz.
Most implementations get this backwards by making sizing help a permanent fixture that demands attention. A pop-up quiz on page load interrupts the shopper before doubt has even formed. The better pattern is quiet availability: a "find my size" link near the selector that expands in place, so the confident buyer never notices it and the doubtful buyer finds it in one click.
What you know about the shopper changes everything
For a first-time visitor, you know nothing, so the recommender should ask the minimum: one or two inputs, height and weight or usual size, and a recommendation. Every additional question is a drop-off risk. The data you collect should be limited to what improves this recommendation, not what would be nice for marketing.
For a returning customer with purchase and return history, the recommender should barely appear. If they bought a medium in this cut last season and kept it, preselect medium and say so: "medium worked for you last time." That single line outperforms any quiz for repeat buyers because it uses their own proven data instead of asking them to prove it again.
The product side of the equation
Not every product needs the same sizing help. A relaxed-fit tee in a standard size run needs almost none; a tailored blazer with a European cut needs a lot. Let the product's return rate and size-related support tickets decide how prominent the guidance is. High return rates mean the current guidance is failing, which is an argument for more help, not less.
New silhouettes deserve proactive guidance even for loyal customers. A shopper who knows their size in your classic fit does not know it in the new oversized line. Flag the difference explicitly: "this cut runs large, most customers size down." That one sentence, placed at the selector, prevents more returns than any algorithm.
Designing the recommendation itself
A good size recommendation is specific and confident: "Size M" with one line of reasoning, not a probability distribution across four sizes. Shoppers cannot act on "70% medium, 30% large." They can act on a clear pick with a clear rationale, especially when the rationale references something they told you.
Always pair the recommendation with the escape hatch: what to do if it is wrong. Free exchanges, prominently stated next to the recommendation, convert the remaining doubt into a purchase. The recommender's real job is not perfect accuracy; it is making the shopper feel safe choosing. A confident pick plus a painless exchange policy does that.
Measuring whether it works
The metric that matters is size-related returns, not quiz completion. A recommender that everyone completes but does not reduce returns is entertainment, not personalization. Segment return rates by whether the shopper interacted with the guidance, and by whether they followed the recommendation.
Watch the second-order effects too. If the recommender increases conversion but the new customers return at higher rates, the guidance is overselling confidence. The healthy pattern is conversion up and size-related returns down together. When you see that, expand the recommender to more products; when you do not, fix the logic before expanding the surface.
Reviewed
Published Oct 7, 2026.