What is a Size Prediction Algorithm?
Last updated 2026-06-15
Size prediction algorithms tackle one of online fashion retail's most costly problems: sizing uncertainty. With no universal sizing standard across the fashion industry — a medium at one brand may correspond to a large or small at another — consumers frequently order multiple sizes to find the right fit, returning what does not work. Size-related returns account for an estimated forty percent of all online fashion returns, costing retailers billions annually and generating significant environmental impact through shipping and processing. These algorithms employ several complementary approaches to predict the right size. Collaborative filtering analyzes the purchasing and return patterns of millions of customers to identify sizing relationships between brands — if users who wear a size thirty-two at Brand A consistently order a size thirty-four at Brand B, the algorithm learns this mapping and applies it to future recommendations. Body measurement-based approaches use customer-provided measurements or body scan data to match individual proportions against detailed garment specification data, accounting for ease allowances and intended fit. More sophisticated implementations incorporate fit preference modeling, recognizing that the right size is not just about physical measurements but also about how an individual prefers their clothing to fit. A customer who prefers relaxed-fit tops but slim-fit trousers has different sizing needs within a single wardrobe, and advanced algorithms learn these nuanced preferences over time through explicit feedback and implicit signals like return reasons and repeat purchases in specific fits. The accuracy of size prediction algorithms has improved dramatically as datasets grow and machine learning models become more sophisticated. Leading implementations report accuracy rates of eighty-five to ninety-five percent — meaning the recommended size is the one the customer ultimately keeps — compared to the approximately sixty percent accuracy of consumers selecting sizes based on traditional size charts alone. Fashion retailers implementing these algorithms typically see return rate reductions of twenty to thirty percent, making them one of the highest-ROI technologies available to online fashion businesses.
An online shopper browsing a new-to-them denim brand is unsure whether to order their usual size twenty-eight or size up based on reviews mentioning the brand runs small. The retailer's size prediction algorithm analyzes the shopper's profile — previous purchases from five other denim brands, the sizes kept versus returned, body measurements from a stored body scan, and a stated preference for a relaxed rather than skinny fit. The algorithm recommends a size twenty-nine in this brand's relaxed-cut jean, noting that this size provides the relaxed ease the shopper prefers based on their body measurements and the specific cut's garment dimensions. The shopper orders the recommended size and finds it fits exactly as hoped, avoiding the need to order two sizes and return one.
How TRY helps
TRY suggests outfit combinations from the clothes you already own. Upload your wardrobe, pick an occasion, and get ideas that fit your style—including staples and formulas that work.
Questions, answered.
Why do clothing sizes vary so much between brands?
Clothing size inconsistency exists because there is no legally enforced universal sizing standard in the fashion industry. Each brand develops its own sizing based on their target customer demographic, brand positioning, design aesthetic, and manufacturing specifications. A luxury brand targeting a different demographic than a fast-fashion retailer may assign different measurements to the same size label. Vanity sizing — where brands label garments with smaller size numbers than actual measurements would indicate — further compounds the inconsistency. Size prediction algorithms exist specifically to navigate this fragmented sizing landscape and translate between brands' different sizing systems.
How do size prediction algorithms handle different fit preferences?
Advanced size prediction algorithms separate the concepts of physical fit (does the garment accommodate the body) from preferred fit (does the garment fit the way the wearer wants it to). They learn individual fit preferences through explicit inputs — such as asking whether you prefer a loose or slim fit for different garment categories — and through implicit signals like which sizes you keep versus return and your purchasing patterns across slim, regular, and relaxed cuts. This allows the algorithm to recommend a smaller size for someone who prefers body-conscious fits and a larger size for someone who prefers relaxed silhouettes, even if both customers share the same body measurements.