AI Size Recommendation System for Clothing and Shoes
Size mismatch returns — largest cost item in fashion e-commerce: 30-40% of returns due to fit issues. AI size recommendation system reduces this by 20-35%, increasing conversion through buyer confidence in choice.
Size Recommendation Architecture
Task has two components: normalization of brand size charts and personalization based on purchase history. Different brands use different standards (EU, UK, US, IT), plus sizes vary by category within brand. System learns continuously: each return with "size didn't fit" reason refines brand offset. Minimum history for personalization: 3 completed purchases without return.
| Metric | Before | After |
|---|---|---|
| Returns due to size | 28% | 18% |
| Product page conversion | 3.2% | 4.1% |
| Confidence > 0.7 % users | — | 65% |
| Coverage (has history) | — | 72% users |