What problem does it solve?
Sizes mean different things across brands, categories and markets. Zalando says that a size L or 48 can describe two garments with very different measurements, and that every customer has a different body shape and a preference for how clothing should fit. A shopper who is unsure orders two sizes and returns one, or returns the one that does not fit.
Returns cost the retailer shipping, handling and often the margin of an item that cannot be resold at full price, and they add shipments and emissions. Zalando reports that in European online fashion return rates can reach around 50%, with size and fit accounting for up to half of those returns, and that jeans can reach 65%. Retailers also lose the sale when a shopper leaves because they cannot tell which size to pick.
- Zalando says that in European online fashion return rates can reach around 50%, with size and fit accounting for up to half of those returns, and that in jeans return rates can reach 65%.How Zalando uses technology to help customers find the right size (2026)
How does it work?
- Learn how each product fits. The system combines brand supplied measurements, purchase and return behavior, shopper fit feedback and, at Zalando, trained fit models who try garments on. From this it flags items that run small or large.
- Learn the shopper. It uses the shopper's own history, including return reasons, and a size profile where the shopper says which brands and items fit well, even if bought elsewhere.
- Add measurements if the shopper agrees. Zalando predicts body measurements from two phone photos or a video, a tool it introduced in July 2023.
- Recommend a size at the moment of choice. The product page, size chart or profile shows a recommended size or a "size up" or "size down" flag, as Zalando does.
- Close the loop. Purchases, returns and fit feedback feed back into the model, and fit data goes back to brands so recurring sizing problems can be fixed at the source.
- Audience
- Customer facing
- Autonomy
- Assist
- Adoption
- Early adopters
- Channels
- Mobile app
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Error reduction | 4% | 0.4% to 8% | 3 | 1 organization, 2 vendor |
| Conversion uplift | Too few to pool | 1% to 2% | 2 | 2 vendor |
| Revenue uplift | Too few to pool | 6% | 1 | 1 vendor |
| Users served | Not pooled | at least 1.5 million | 1 | 1 organization |
Value drivers: Lower cost to serve, Revenue growth, Customer experience.
Indicative value
An online fashion retailer shipping 5 million orders a year
USD 15,840 to USD 1.7 million
Annual return handling cost avoided per year
How this is calculated
Formula: orders * returnRate * sizeShare * preventedShare * costPerReturn. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Orders per year orders, orders | 3,000,000 | 7,000,000 | Illustrative size for a mid sized online fashion retailer. Editorial assumption, replace with your own. |
| Share of orders returned returnRate, fraction of orders | 0.2 | 0.4 | Editorial assumption, below the "around 50%" that Zalando says European online fashion can reach, since a mixed assortment returns less than a fashion only one. |
| Share of returns that are size related sizeShare, fraction of returns | 0.33 | 0.5 | Zalando reports that one third of its returns are size related and that size and fit account for up to half of returns in European online fashion. |
| Share of size related returns prevented preventedShare, fraction of size related returns | 0.01 | 0.08 | The high end is Zalando's reported 8% of size related returns prevented overall in 2025. The low end is conservative against Fit Analytics' vendor reported return rate reductions of 0.4% at THE ICONIC and 4% at AMARO, which measure the return rate in an A/B test, a different base from the size related share. |
| Cost of handling one return costPerReturn, USD per return | 8 | 15 | Editorial assumption covering return shipping, handling and write down, replace with your own cost per return. |
What it leaves out: Counts only the cost of handling returns that no longer happen. It leaves out the extra revenue from higher conversion (THE ICONIC's A/B test reports 6% more revenue after returns, a vendor figure), the cost of the tool, the margin effect of fewer markdowns on returned stock, and the emissions from fewer shipments.
Who already uses it?
4 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Zalando
Germany · Retail and ecommerce · 2025
Zalando has run an in house Size & Fit team since 2018, now 80 specialists, that combines machine learning, computer vision and 3D technology to give size advice on product pages. Size flags tell a shopper when an item runs small or large, size recommendations use purchase and return history and a customer's size profile, and a body measurement tool predicts measurements from two phone photos or a video.
Zalando reports that its foundational size and fit solutions cover around 70% of the assortment and prevented 8% of size related returns overall in 2025. A Virtual Fitting Room that builds a 3D avatar from the measurements is moving from pilots to a permanent experience, and Zalando says its pilots reduced return rates by up to 40%.
- Error reduction: 8%, in 2025
"In 2025, these solutions prevented 8% of size-related returns overall."
Claimed by: organization - Users served: at least 1.5 million, customers who tried the Body Measurement tool, as of 2026
"More than 1.5 million customers have already tried the experience, and the number continues to grow."
Claimed by: organization
AMARO
Brazil · Retail and ecommerce · 2022
AMARO, a Brazilian digital native fashion brand that designs over 10,000 items a year, put Fit Analytics' Fit Finder size advisor on its website, its iOS and Android apps and in its Guide Stores, localized for Brazil. In an A/B test against a size chart, Fit Analytics reports a 2% increase in conversion rate and a 4% reduction in return rate over two and a half months. The figures come from Fit Analytics' own case study, which gives no date for the test.
- Conversion uplift: 2%, 2.5 month A/B test
"In just two and a half months, Fit Analytics delivered a 2% increase in conversion rate and a 4% reduction in return rate leading to a significant bottom-line boost for Brazil’s hottest digital native brand."
Claimed by: vendor - Error reduction: 4%, 2.5 month A/B test, across all channels
"In just two and a half months, Fit Analytics delivered a 2% increase in conversion rate and a 4% reduction in return rate leading to a significant bottom-line boost for Brazil’s hottest digital native brand."
Claimed by: vendor
THE ICONIC
Australia · Retail and ecommerce · 2022
THE ICONIC, an Australian online fashion and sports retailer with over 700 brands and 45,000 products, tested Fit Analytics' Fit Finder size advisor against a traditional static size chart in a three month A/B test with 250,000 shoppers. Fit Analytics reports that shoppers who used the advisor converted 1% more often, returned 0.4% less and contributed 6% more revenue after returns than the control group. The figures come from Fit Analytics' own case study, which gives no date for the test.
- Conversion uplift: 1%, three month A/B test with 250,000 shoppers
"Shoppers using the Fit Analytics size advisor showed a 1% increase in conversion rate."
Claimed by: vendor - Error reduction: 0.4%, three month A/B test with 250,000 shoppers
"Improved fit-related user confidence also led to a 0.4% reduction in return rate."
Claimed by: vendor - Revenue uplift: 6%, three month A/B test with 250,000 shoppers
"Overall, Fit Analytics visitors contributed 6% more revenue after returns compared to the control group."
Claimed by: vendor
ASICS
Japan · Retail and ecommerce · 2018
ASICS launched True Fit's fit recommendation technology in 2018 across all its brands, starting with footwear, to give shoppers a size recommendation on its ecommerce sites. True Fit's case study says ASICS reduced size bracketing, where shoppers order several sizes to keep one, and that its ecommerce director said the return data let ASICS update its return policy and helped mobile conversion. The case study page gives headline claims without the underlying figures, so no metric is recorded.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Size charts and garment measurements per product and brand, kept current
- Order and return history with return reasons, linked to customer accounts where consent allows
- A way for shoppers to give fit feedback and to enter reference items that fit
Systems to integrate
- Product detail page, wishlist and customer profile in the web shop and app
- Order management and returns system, for purchase and return events
- Product information system, for size charts and garment measurements
- Analytics and A/B testing platform, to measure return rate against a control group
Complexity: Medium
A retailer either buys a size advisor from a specialist vendor and adds its widget to product pages, or builds one as Zalando has since 2018 with a team of 80. Both routes need size charts and garment measurements loaded for the assortment, order and return events sent back to the model, and a privacy review where body photos or measurements are collected.
- 1
Test against a plain size chart
Run an A/B test of the size advisor against the existing size chart on a set of categories with high return rates, as THE ICONIC and AMARO did, and measure return rate and revenue after returns.
- 2
Start where sizing is hardest
Begin with jeans, where Zalando reports some of the highest size related return rates, and with other categories that have high return rates in your own data. Add categories as the model proves itself.
- 3
Fix the product data first
Check garment measurements and size charts for the pilot assortment before launch. A model fed with wrong measurements recommends the wrong size with confidence.
- 4
Make measurement optional and explain it
If you offer body measurement from photos, tell shoppers what is captured, how long it is kept and that they can use size advice without it.
- 5
Report fit data back to buyers and brands
Share products with repeated sizing complaints with buyers and brands, so the brand fixes the garment itself.
Guardrails
- The recommendation is advice, shown with the shopper's free choice of size and a clear return policy
- Photos and body measurements are used only for size advice, kept for a stated period and deleted on request
- The advisor says when it has too little data for an item instead of showing a confident size
- Results are measured against a randomly assigned control group
KPIs to instrument
- Return rate and size related return rate, advisor users against a control group
- Conversion rate and revenue after returns, per category
- Share of product page views where the advisor gives a recommendation
- Size bracketing, the share of orders with several sizes of one item
Human in the loop
The shopper makes the final choice. We suggest that inside the retailer, fit specialists and category buyers review items that the model flags as inconsistent, and decide whether to correct the size chart or raise the issue with the brand. Zalando also uses fitting models, a dedicated team that tries on items to help identify sizing issues.
Common failure modes
- Selection bias in the headline number
- Shoppers who open a size advisor may be more careful buyers to begin with. Compare with a randomized control group.
- Wrong garment data
- Mislabelled sizes and wrong measurements flow straight into recommendations. Check the data for the highest return products first.
- Cold start for new shoppers and new items
- With no history the model leans on averages. Offer a short size profile or measurement step and show fewer recommendations when confidence is low.
- Privacy backlash on body photos
- Photo based measurement can put shoppers off or create a data protection problem. Keep it optional, minimize what is stored and explain it in plain language.
What are the risks and rules?
EU AI Act
Depends on design
A size advisor is not listed in Annex III, and a size widget on a product page is minimal risk. The design decides the rest. Article 50 transparency applies only if the size advice is delivered through a conversational assistant that interacts directly with shoppers, and then they must be told it is AI unless that is obvious from the context. Body measurements derived from photos can be biometric data (Article 3(34)).
Photo based measurement is likely not a biometric categorization system under Article 3(40) while it serves only as an ancillary feature of the sale, for example previewing a product on the shopper, as recital 16 says of body filters on online marketplaces. It would leave that exception if the system inferred sensitive characteristics from the images or reused them for another purpose.
Guidance
- Regulation (EU) 2024/1689 (EU AI Act), Article 50 (European Union, Europe). Article 50 sets transparency duties for AI systems that interact directly with people, which reaches a size advisor only when it is delivered as a conversational assistant.
- EU AI Act, Recital 16, biometric categorization (European Union, Europe). Excludes biometric categorization that is a purely ancillary feature of another commercial service, and names filters that categorize body features on online marketplaces, which fits photo based measurement used only for size advice.
- Regulation (EU) 2016/679 (GDPR) (European Union, Europe). Body measurements and photos are personal data; purpose limitation, data minimization, storage limits and deletion on request apply.
- A guide to the data protection principles (Information Commissioner's Office, Europe). Sets out the UK data protection principles that apply to purchase history, return reasons and body measurements used for size advice.
Controls to put in place
- A documented purpose, retention period and deletion path for photos, measurements and size profiles
- Consent or another lawful basis recorded for each use of purchase and return history
- A data protection impact assessment for any photo or video based measurement
- Regular checks of recommendation quality by category and size range, including less common sizes
Frequently asked questions
- How much can size advice reduce returns?
Zalando reports that its size and fit solutions prevented 8% of size related returns overall in 2025. Fit Analytics reports a 0.4% reduction in return rate at THE ICONIC and 4% at AMARO in A/B tests. The figures measure different things and come from the retailer or the vendor without an audit, so treat the Zalando figure as the best documented overall result and the vendor tests as per test results.
- Is photo based body measurement safe to use?
It is personal data under GDPR, so it needs a lawful basis, a stated purpose, a retention limit and a deletion route. Zalando offers it as an optional tool, with two phone photos in tight clothes, and reports that more than 1.5 million customers have tried it. Shoppers who decline should still receive size advice from their purchase history and size profile.
- Does a size advisor also raise sales?
Fit Analytics reports a 1% conversion increase and 6% more revenue after returns for THE ICONIC in a three month A/B test, and a 2% conversion increase at AMARO. These are vendor reported results from single tests, with the test dates not stated.
How to cite this page
Blits.ai AI Use Case Library, "AI size and fit recommendation for online fashion", last verified 14 October 2026, https://www.blits.ai/ai-use-cases/size-and-fit-recommendation. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 14 October 2026: First published