AI Size Finders That Recommend the Wrong Size
A size finder that recommends from the wrong data sends the customer the wrong size with confidence. The return costs the store more than the sale was worth.
A customer enters her height, weight, and usual size into an AI size finder. It recommends a medium. She orders the medium, it arrives too tight across the shoulders, and she returns it. The finder was not wrong about her measurements. It was wrong about the garment, because it had never seen this brand's actual fit data, and it recommended from generic sizing tables instead.
Size finders fail quietly. The customer blames her body, or the brand, and the store eats a return that cost more than the margin on the item. The fix is data the store already has, if it is willing to use it.
Where the recommendation comes from
Most AI size finders work from one of two sources: the brand's own return and fit data, or generic size tables. The generic tables are the problem. A medium in one brand is a large in another, and a finder that has never seen this brand's garments cannot know that. It recommends the average, and the average is wrong for the customer who needed the tool in the first place.
The second problem is the input. Height, weight, and a usual size are weak predictors of fit across the shoulders or the waist. A finder that does not ask about fit preference, loose or fitted, or about the specific garment's cut, is guessing from the weakest signals available, and presenting the guess as a recommendation.
What good fit data looks like
The stores that get size finders right feed them their own data: the measurements of the actual garment, not the size chart, and the return reasons for that style. A style with a high return rate for tight shoulders needs a recommendation that sizes up for broad-shouldered customers, and only the return data says that.
The honest finder also shows its working. A recommendation that says medium, with a note that customers with your measurements usually keep the large in this style, is more useful than a bare medium. The customer can decide, and the store has shared the uncertainty instead of hiding it behind a confident label.
Measuring whether it works
The metric is the return rate for size reasons on orders that used the finder, compared with orders that did not. If the finder is working, the rate should be lower. If it is the same or higher, the finder is not helping, and it is probably hurting, because customers who trusted it and got the wrong size are less likely to reorder than customers who guessed.
Run the comparison by style, not in aggregate. A finder can be right on t-shirts and wrong on jackets, and the aggregate hides it. The styles where it fails are the ones where the fit data is thinnest, and those are the styles to fix first, or to exclude from the finder until the data exists.
Related reading: Glossary, About, AI Review Summaries That Contradict the Reviews Underneath Them.
Frequently asked questions
Why do AI size finders recommend the wrong size?
Most work from generic size tables, not the brand's own garment measurements and return data. A medium varies between brands, so a finder that has never seen this brand's fit data recommends the average, which is wrong for the customer who needed help.
What data makes a size finder work?
The measurements of the actual garment, not the size chart, and the return reasons for that style. A style with high returns for tight shoulders needs a recommendation that sizes up for broad-shouldered customers, and only return data reveals that.
How do I know if my size finder is helping?
Compare the return rate for size reasons on orders that used the finder against orders that did not, style by style. If the finder rate is not lower, the finder is not helping, and customers who trusted it are less likely to reorder.
Should a size finder show its reasoning?
Yes. A recommendation with a note, such as customers with your measurements usually keep the large in this style, lets the customer decide. A bare size label hides the uncertainty behind a confident guess.