Ours asks five questions. Here is why it is five, what they can and cannot tell us, and where the approach runs out.
A size recommender is a trade. Every question you ask improves the answer and costs you shoppers who abandon the quiz. The whole design problem is deciding which questions earn their place.
The five
- Height, in feet and inches or centimeters.
- Weight, in pounds or kilograms.
- Age.
- Preferred fit: slim, regular, or relaxed.
- Gender.
That is the entire input set. Three of them sit on the first screen, two on the second, and the third screen is the answer. No account, no measuring tape, no photograph.
Why not ask for real measurements
Because almost nobody has a tape measure within reach while shopping, and the ones who do already know their chest measurement and are reading your chart directly. Asking for bust, waist, and hip produces a better recommendation for the small group who will answer and no recommendation at all for everyone else.
Height and weight are numbers people know without getting up. That is the entire argument for them, and it is a strong one.
What the recommendation is actually made of
The recommender reasons against the size chart you published for that specific product, plus the variants you have in stock. It is not consulting a generic industry table, which means a vague chart produces a vague recommendation. Your chart quality is most of the answer.
It also records whether the size it recommended was purchasable. Recommending a large you sold out of three weeks ago is a real failure mode, and it shows up as a number rather than as a mystery.
Where it runs out
Footwear. Height and weight tell you very little about foot length, so we do not pretend otherwise: shoes get the conversion chart rather than a recommendation. The same caution applies to swim, where bust, underbust, and hip drive the fit far more than overall build does.
We would rather say that on the pricing page than let a footwear merchant find out after paying. Category-specific questions are the obvious next step, and until they ship, the honest description of the recommender is that it is built for apparel.
A recommender that answers confidently in a category it cannot read is worse than no recommender.
How to judge one
Not by accuracy claims, which nobody can verify from outside. Look at two numbers instead: how many shoppers who open the chart actually complete the quiz, and how often the recommended size gets overridden at the point of adding to cart. A high override rate is not a broken model. It usually means the underlying chart needs work.