What data does AI size recommendation need? Measurements, fit notes and returns
Size recommendation is only as good as the data behind it. A practical guide to the garment, customer and returns data that size and fit models need, and who in the business owns each piece.
KEY TAKEAWAYS Summary by the editors
- AI size recommendation combines three kinds of data: garment data (measurements and fit intent per size), customer data (purchase history, fit feedback and sometimes body measurements) and returns data with reasons.
- Zalando says its size flags and recommendations draw on brand-provided product information, purchase and return behaviour, customer fit feedback and insights from trained fitting models who try on products.
- Zalando reports that its size and fit solutions prevented 8% of size-related returns in 2025 and that its foundational size and fit tools cover about 70% of its assortment.
- Return reason codes are only useful for fit models if they distinguish 'too small' from 'too large' and are captured consistently across channels.
- Brands that supply accurate graded measurements and clear fit descriptions make size recommendation work better on every channel that sells their products, including retailer and marketplace sites.
AI size recommendation needs three things: accurate garment measurements for every size, signals about the customer such as past purchases, fit feedback or body measurements, and structured return reasons that say whether an item was too small or too large. Without consistent product measurements and clean return data, even sophisticated models fall back on guesswork.
Why does size recommendation matter for returns?
Returns are a large cost in online fashion, and size and fit are a major driver. Zalando wrote in June 2026 that European online fashion return rates can reach about 50%, that size and fit account for up to half of those returns, and that return rates for jeans can reach 65%. In the US, the National Retail Federation estimated in October 2025 that online returns would amount to 19.3% of online sales across retail as a whole.
Zalando, which has worked on size and fit since 2018 with an in-house team it says numbers 80 specialists, reports that its size and fit solutions prevented 8% of size-related returns in 2025. In an earlier corporate article it said size advice had reduced size-related returns by 10% compared with items without size advice. These are company-reported figures, but they give a realistic sense of scale: meaningful, not miraculous.
What garment data does size recommendation need?
The foundation is product data that describes how each size of each style is actually cut. Fashion teams often have this in technical packs and grading files, but it rarely travels cleanly into e-commerce systems.
| Data type | Examples | Usual owner | Common problem |
|---|---|---|---|
| Graded garment measurements | Chest, waist, hip, inseam, sleeve per size | Product development, technical design | Stored in PDFs or spreadsheets, not in product data systems |
| Fit intent | Slim, regular, relaxed, oversized; stretch content | Design, merchandising | Described inconsistently or only in marketing copy |
| Size system mapping | EU, UK, US, letter sizes, brand-specific labels | Product data, e-commerce | Conversion tables differ between brands and markets |
| Purchase and keep history | Sizes bought and kept per customer | E-commerce, CRM | Not linked to return outcome |
| Return reasons | Too small, too large, wrong length, style | Customer service, logistics | Free text or vague codes such as 'didn't fit' |
| Customer fit feedback | Ratings of fit after delivery | E-commerce | Low response rates |
| Body measurements (optional) | Measurements from phone photos or manual input | E-commerce, with consent | Privacy and consent obligations |
What customer data improves size recommendations?
Zalando's published descriptions show how customer signals are layered. Its personal size recommendations use purchase history including return reasons, the sizes a customer has entered in a profile, fit feedback on ordered items and reference items that fit the customer well, even if bought elsewhere. Customers can also list brands and products that fit them well, which helps filter the assortment.
Since July 2023 Zalando has offered size recommendations from body measurements: customers take two photos, front and side, in tight clothing, and the system predicts their measurements. The retailer says more than 1.5 million customers have tried its body measurement tool and that it now holds a large anonymised dataset of body measurements from European fashion customers. Body data is powerful but sensitive, so it requires clear consent, data minimisation and transparency under data protection law.
How are return reasons used in fit models?
Return reasons turn individual disappointments into product knowledge. If many customers return a style as too small, a model can flag it as running small and suggest sizing up. Zalando describes size flags that tell customers whether an item runs smaller or larger than expected, using brand-provided measurements, customers' return history and reasons, and findings from a dedicated team of fitting models who try on items. It also says computer vision can flag new items from a brand with known sizing issues even before any purchase or return history exists.
- Use structured reason codes that separate too small, too large, too short, too long and not as described.
- Capture reasons the same way across web, app, store and marketplace returns.
- Link each return to the exact SKU and size, not only to the style.
- Allow an optional free-text comment, and analyse it separately.
- Review reason quality regularly; a high share of 'other' is a data problem.
Who owns size and fit data inside a fashion company?
Size data crosses departmental lines, which is the main reason it is often poor. Technical design owns measurements, e-commerce owns customer behaviour, customer service owns return reasons, and merchandising writes fit descriptions. A workable model assigns one owner for the size and fit data set and agrees which system is the master for graded measurements.
For brands that sell through retailers and marketplaces, the stakes are higher. Zalando says it shares return insights with brands, showing whether size and fit is a recurring issue in their categories. Brands that supply accurate measurements and fit descriptions to partners improve size advice on every channel that sells their products, not just their own site.
Size system mapping deserves particular attention for brands selling across markets. The same garment may be labelled with EU, UK, US or letter sizes, and conversion tables differ between brands. A model that learns from customers who bought a size 38 in one brand and a size M in another needs a reliable mapping to the underlying measurements; otherwise it learns the labels rather than the fit.
How should a brand start improving size recommendation data?
- Pick one high-return category, such as jeans or dresses, and collect graded measurements for every live style in a structured format.
- Standardise fit descriptors and stretch information across the category.
- Fix return reason codes so that size direction is always captured.
- Join returns to SKU-level sales to calculate return rates by size, and look for styles that run small or large.
- Feed findings back to product development and to retail partners, then measure size-related return rates over the following seasons.
Frequently asked questions
How does AI size recommendation work?
It combines garment measurements per size with signals about the shopper, such as past purchases and returns, fit feedback or body measurements, and predicts the size most likely to fit. Retailers such as Zalando also use aggregated return reasons to flag items that run small or large.
What is a size flag?
A size flag is a message on a product page telling shoppers whether an item runs true to size, small or large, often with a suggestion to size up or down. It is typically based on return reasons and fit feedback from previous buyers, plus brand measurements.
Do size recommendation tools reduce returns?
Published results suggest a measurable but moderate effect. Zalando reports that size advice reduced size-related returns by 10% compared with items without it, and that its size and fit solutions prevented 8% of size-related returns in 2025.
Is body measurement data personal data?
Yes. Body measurements linked to a customer account are personal data and need a lawful basis, transparency and secure handling under data protection law such as the GDPR. Retailers should collect only what is needed and explain how it is used.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
