Yes, AI can reduce size-related returns, but the gains come from better data more than clever models. AI size recommendation works by combining a shopper's inputs (usual size in other brands, height, fit preference) with your product's real measurements and what past buyers said about fit, then suggesting the size most likely to fit. It helps most when your size charts are accurate and your return reasons are recorded, and it can't fix products whose sizing is inconsistent from batch to batch.

Fit is the biggest return driver for most apparel and footwear stores, and returns eat margin through reverse shipping, handling, damaged stock and lost resale time. This post covers the main approaches to AI fit recommendations, the data each needs, what virtual try-on can and can't do, how to handle body data responsibly, and a practical way to start.

Why are size-related returns so hard to fix?

Sizing is not standard. A "medium" from one brand is a "large" from another, and even within one brand, fit varies by style, fabric and factory batch. Shoppers compensate by "bracketing": ordering two or three sizes and returning the rest. Every bracketed order is a return you've built into the sale.

Store owners usually try three things: a size chart, a "fits true to size" note, and a generous return policy. Size charts help only if shoppers measure themselves, which most don't. "True to size" is vague. And a generous policy, while good for conversion, makes bracketing painless.

AI's role is to make the size decision easier and more accurate for the shopper, using information you already have but don't present well.

What approaches to AI size recommendation exist?

There are four main approaches, often combined.

Questionnaire-based recommendations. The shopper answers a few questions: height, weight or build, usual size in brands they know, fit preference (snug or relaxed). A model compares those answers with your garment measurements and with outcomes from past shoppers who gave similar answers, then recommends a size. This is the most common approach and works without cameras or photos.

Purchase and return history. For returning customers, what they kept and returned is the strongest signal. If someone kept a medium in one style and returned a medium in another as "too tight", that tells you about both the shopper and the product.

Review and return-reason mining. AI reads reviews and return comments to extract fit signals per product: "runs small", "long in the arms", "tight across the shoulders", "true to size". These can update the product page ("Most buyers say this runs small; consider sizing up") and the recommendation model. This is valuable even if you never build a recommendation widget. See AI customer feedback analysis for the method.

Body scanning and virtual try-on. Some tools estimate body measurements from photos, or show a garment on a generated image of the shopper. These are improving, but they raise the biggest privacy questions and vary widely in accuracy. More on them below.

What data do you need for good fit recommendations?

DataWhy it mattersHow to get it
Garment measurements per sizeThe foundation; charts alone are often body measurements, not garment measurementsMeasure samples of each style and size
Fit type per styleSlim, regular, relaxed, oversized changes everythingTag in product data
Fabric stretchStretch fabrics tolerate size differencesTag in product data
Return reasons by product and sizeShows where sizing goes wrongMake "too small / too big / fit issue" a required return field
Review fit commentsReal-world fit feedbackMine reviews with AI
Shopper inputsMatches the shopper to the productShort questionnaire on product pages

The uncomfortable truth: many stores' size charts are copied from a supplier, or describe body measurements that don't match the actual garments. If the data is wrong, AI recommendations will be confidently wrong. Measuring your actual garments is often the single most useful step.

Can virtual try-on reduce returns?

Partly, and less than marketing suggests. Virtual try-on tools generally fall into two types:

  • Visual try-on shows how a garment looks on a model with a similar body type, or on an image of the shopper. It helps with style and color decisions ("Does this suit me?") but often doesn't accurately represent fit, drape or tightness.
  • Measurement-based fit estimates body dimensions from inputs or photos and compares them with garment measurements. It addresses size more directly, but accuracy depends on photo quality, pose, clothing and the model.

For eyewear, makeup, accessories and some footwear, visual try-on can be genuinely useful. For fitted apparel, a well-calibrated size recommendation based on garment measurements and past outcomes is often more reliable than a visual. Be cautious with any tool that claims to know how a garment will fit from an image alone, and test against your own return data.

How should you handle body data and privacy?

Fit recommendations involve some of the most personal data a store can collect: height, weight, body measurements, sometimes photos. Treat it accordingly.

  • Collect the minimum. Usual size in another brand plus height and fit preference is often enough. Don't ask for weight, photos or detailed measurements unless they improve recommendations measurably.
  • Make it optional and explain why. Tell shoppers what you'll use the data for and let them skip it.
  • Don't reuse it for other purposes. Fit data should not feed ad targeting or be shared with partners without clear, explicit consent.
  • Be very careful with photos. Body images are highly sensitive. If a tool processes photos, understand where they go, how long they're kept and whether they're used to train models. Prefer tools that process and discard.
  • Follow the law where you sell. GDPR, India's DPDP Act and other privacy laws cover this data; some jurisdictions treat biometric data, which photo-based body scanning can involve, with additional restrictions. Check current requirements, and get advice if you plan to collect images.
  • Watch the tone. Recommendations should be neutral and practical. Nobody wants a store commenting on their body.

Our post on AI data privacy and security for business covers the general controls.

What are the benefits and the precautions?

BenefitPrecautionGuardrail
Fewer size-related returnsRecommendations based on wrong size chartsMeasure actual garments; audit charts per style
Less bracketingConfident wrong advice damages trustShow confidence and reasoning ("most buyers with your size kept an M")
Better product pages from review fit insightsOverreacting to a few reviewsRequire a minimum number of fit comments before changing guidance
Shoppers buy with more confidenceCollecting sensitive body dataMinimal, optional inputs; clear consent; no reuse for ads
Product and supplier feedbackInconsistent sizing across batchesFeed return data back to sourcing and QC
Personalized experience for repeat buyersOver-personalization feeling intrusiveKeep it practical; let shoppers edit or delete profiles

How should a fashion or footwear store start?

Start with the data, not the widget.

  1. Make return reasons specific. Require "too small", "too big", "fit issue (where?)" or "not a fit issue" on every return.
  2. Mine reviews and return comments for your top 30 products with AI, and tag each with a fit verdict.
  3. Update product pages with clear fit notes ("runs small, consider sizing up") where the evidence is strong.
  4. Measure your best sellers and correct any size charts that don't match the garments.
  5. Then trial a recommendation tool on a subset of products, with an A/B test.

Measure size-related return rate per product, bracketing rate (orders with multiple sizes of the same item), and conversion. Steps 1 to 4 often deliver a meaningful part of the improvement on their own. For related reading, see AI for D2C brands and our AI for ecommerce businesses guide.

Frequently Asked Questions About AI Size Recommendation

How accurate are AI size recommendations?

It depends mostly on your data: accurate garment measurements, consistent sizing across batches and good return reason records. With weak data, recommendations will often be wrong. Test against a holdout group and track size-related returns.

Do shoppers actually use size recommendation tools?

Many do when the tool is quick and appears where they choose a size. Long questionnaires and mandatory sign-ups reduce use. A few questions about usual size and fit preference is a reasonable start.

Is virtual try-on worth it for a small fashion store?

For fitted apparel, usually not as a first step. Accurate garment measurements, clear fit notes from reviews and a simple recommendation often do more. Visual try-on can be more useful for eyewear, makeup and accessories.

Is it legal to collect customers' body measurements?

Generally yes with appropriate consent and transparency, but privacy laws apply, and photo-based body data may be treated as biometric or otherwise sensitive in some jurisdictions. Collect the minimum, explain why, and check current rules where you sell. This is not legal advice.

Can AI fix inconsistent sizing from my suppliers?

It can detect it, by spotting products or batches with unusually high size-related returns, but it can't fix the garments. Feed those insights back to sourcing and quality control.

Want fewer "didn't fit" returns?

If size-related returns are eating your margin, tell us about your catalog and return data. We'll help you work out what your data can support and what to build first. See how we approach AI automation for online stores.