- Why Size Is the Real Returns Problem
- How AI Size Recommendation Works
- Measuring the Impact: Fit Accuracy and Conversion
- Frequently Asked Questions
- Conclusion
AI size recommendation is the technology behind one of the fastest-payback fixes in fashion e-commerce: telling a shopper their size, for that specific brand, before they add to cart. The direct answer is that it works by combining a few self-reported inputs (height, weight, usual size) with the retailer's own size grid, and increasingly with a photo-based morphology read, to output a single recommended size with a confidence level. This post breaks down why size uncertainty is the single biggest return driver, how the recommendation engine actually works, and how to measure whether it is paying off.
Why Size Is the Real Returns Problem
Colour and style disappointment cause some fashion returns. Fit causes most of them. Industry estimates commonly cited by McKinsey and apparel sizing researchers put size and fit issues behind the majority of fashion e-commerce returns, often well above 50%.
The workaround shoppers use is bracketing: ordering the same item in two or three sizes and returning what does not fit. It feels like a free trial to the shopper. For the retailer it multiplies shipping, handling, and restocking cost per completed sale, a cost structure we break down in The True Cost of Fashion Returns.
Why generic size charts fail
A static "S/M/L" chart assumes every brand cuts the same way. It does not. A size 38 dress at one label can fit like a 36 or a 40 at another, which is why shoppers do not trust the chart enough to buy without a safety margin.
How AI Size Recommendation Works
A modern size recommendation flow takes three inputs: height, weight, and the shopper's usual size in a reference brand. Where a full-body photo is available, the system adds a morphology read (hourglass, pear, apple, rectangle, inverted triangle) and adjusts the recommendation accordingly, for example flagging size up on top or size down on bottom.
The output is not a generic label. It is mapped to the retailer's own size grid, because that is the only grid that matters at checkout. This is the same reason a universal chart fails: the fix has to be brand-specific, not industry-generic.
Integration matters as much as the model. A recommendation engine that requires a 3D body scan or a native app download will not get adopted inside a 3-week rollout window. A single script tag on the product page, returning a size in seconds alongside try-on and colour profile, is what makes this operationally realistic for a retail team without a dedicated ML function, the same approach we describe for personalisation in AI Colour Profiling for Fashion E-Commerce.
Measuring the Impact: Fit Accuracy and Conversion
The clearest signal is the return reason code. Retailers running size recommendation should track the share of returns marked "too small" or "too large" before and after rollout, isolated from "changed my mind" or "not as described" reasons.
Cart and checkout behaviour is the second signal. Uncertainty about fit is a known abandonment driver at the size-selector step, and DHL's reverse logistics research on apparel consistently ties oversized and undersized deliveries to higher return-shipment volume and cost per order.
"Fit and sizing issues remain a leading cause of apparel returns, and retailers that give shoppers a confident, brand-specific size answer at the point of purchase see a measurable drop in size-driven return volume." Source: McKinsey & Company, State of Fashion research.
A useful test structure: run the recommendation live for a subset of SKUs or traffic, hold out a control group on the standard size chart, and compare size-related return rate and conversion rate over 4-6 weeks. That is enough volume to separate signal from noise for most mid-size catalogues, and it lines up with the reduction tactics covered in How to Reduce Fashion E-Commerce Return Rates in 2026.
If you are building the business case for size recommendation on your own catalogue, book a 30-minute demo and we will run the return-rate math against your actual size-related return share.
Frequently Asked Questions
How accurate is AI size recommendation compared to a standard size chart?
Accuracy depends on inputs. Height, weight, and usual size alone already outperform a static chart because the recommendation is mapped to the specific brand's grid. Adding a full-body photo for morphology analysis further improves accuracy, particularly for garments with structured fit like blazers or fitted dresses.
Does AI sizing require a body scan or 3D avatar?
No. A quiz-based flow (height, weight, gender, usual size) already produces a usable recommendation. An optional full-body photo adds morphology detail without requiring 3D scanning hardware or a dedicated capture app.
How is this different from a simple size chart quiz?
A quiz alone still asks the shopper to interpret a generic chart. AI size recommendation maps the answer directly to the retailer's own size grid and, when a photo is provided, adjusts for body shape rather than treating every body as a straight scale-up or scale-down of the same silhouette.
Conclusion
Size uncertainty is the largest and most solvable driver of fashion returns, and fixing it does not require a 3D asset library or a new ML team, just the shopper's basic measurements mapped to the brand's own grid. For retailers ready to see the number on their own catalogue, request a pilot and measure the shift in size-related returns directly.
