To become a Vogue Business Member and receive the Technology Edit newsletter, click here.
Sizing inconsistency across brands is a common point of frustration for customers, who know all too well that a size 8 dress or size 12 jeans likely won’t fit the same depending on where you shop. Amazon Fashion wants to solve that problem using artificial intelligence to standardise sizing, with a new tool available for brands selling on Amazon that can help online shoppers better guess their best fit.
The Fit Insights tool uses large language models (LLMs) to extract and aggregate insights from customer feedback relating to an item’s fit, style and fabric. It contextualises customer reviews with returns and size charts analysis and identifies defects in size charts. Brands can then parlay this information into future designs or manufacturing processes.
The insights provided to brands by Amazon include products categorised by “return health”, a comparison of the product return rate against similar products with low returns, a summary of positive and negative customer feedback and size chart analysis. The tool and insights are available for free to US apparel and shoe brands that are enrolled in Amazon Brand Registry and have sold at least 100 units in the last 12 months.
To deal with online uncertainty, customers have taken to “bracketing” or ordering multiples of one item to figure out the right size at home and return the rest. Guessing the right size is not only a pain for customers, it’s extremely costly for brands and retailers — returns have ballooned into a massive expense. Bracketing “erodes margins and profitability” for retailers, says Neil Saunders, managing director of retail at research firm Globaldata. Almost a quarter of clothing bought online is returned, according to Coresight Research, with the top reason being fit.
“Consumers’ bedrooms have become the new dressing room,” says Amena Ali, CEO of returns technology company Optoro. In the US, returns have amounted to $743 billion in inventory and the typical cost of processing a $100 item is $30 or 30 per cent, she adds, with steep discounting and transportation contributing to the bulk of expenses. “The longer it takes for a retailer to receive a return from a customer, ship it to their warehouse and then process it, the less likely it is to be resold at full price.” In 2022, total online returns accounted for $816 billion in lost sales for US retailers, according to the National Retail Federation.
Fit tech has been a tough nut to crack, and brands have resorted to a myriad of technologies such as body scans, avatars, selfies and augmented reality in attempts to recreate the in-person dressing room function online. In certain instances, such tech has reduced returns by 5 per cent and diminished bracketing by 40 per cent, Ali says.
Google recently introduced a feature that enables online shoppers to see clothing items on a range of different models, using AI to digitally dress them. Khaite and Balmain are utilising tech from Bods, which borrows from the gaming world to digitally dress avatar versions of customers. VF Corporation and Reformation have used tech from 3DLook, whose new “Mobile Tailor” makes complex body measurements based on people’s submitted photos. Amazon has even tested custom-made garments and offers Prime members a try-before-you-buy feature that sends physical products for people to try at home for seven days, paying for only what they keep.
“The main solution to this issue is to provide the customer with a lot more information about how products fit and, ideally, help them understand how the products may look and feel on their body type,” Saunders says, adding that AI can help automate some of this. But it can only go so far. “Sizing software can help shoppers make more informed purchase decisions and reduce bracketing, but this behaviour won’t disappear entirely. At the end of the day, shoppers value trying on goods for themselves before deciding whether to keep them,” Ali says.
Using AI to reduce this problem stands to save brands considerable money on free returns. “It’s well known that customers struggle to find the right fit when shopping for fashion online,” says Jenny Freshwater, VP of Amazon fashion and fitness. “Learning the complex relationship between the size and type of a garment and how it will best fit a customer is a significant scientific challenge, which is magnified by our wide assortment and the millions of customers we serve.”
Combining reviews with AI
Amazon believes personalised, adaptive size recommendations can fix this, gleaned by considering the sizing relationships between brands and their size systems, a product’s reviews and a customer’s own fit preferences. The algorithm anonymously clusters similar customers and products with a similar fit. It then combines what it knows about product details with customer purchases and the sizes that were kept (not returned) by similar customers. Amazon can then use this information to recommend the best size and other styles that might fit.
According to the company, 90 per cent of customers who buy the recommended size are happy with their purchase. Amazon also analyses reviews to provide a personalised summary related to size accuracy, garment fit on specific body areas and fabric stretch; for example, it can recommend if the shopper should size up or down based on reviews from similar customers. (This is an extension of its new AI-generated customer review summaries, which were introduced last August and are already gaining positive feedback from customers and sellers.)
Old-school size charts — the tables designed to translate body measurements into a corresponding size — also received the AI treatment. Amazon developed tech that is able to extract and clean size chart data from multiple sources and then convert it into standardised sizes. This includes removing duplicate info and auto-correcting missing or incorrect measurements. Amazon is experimenting with other ways to show relevant size and measurement details, different from the traditional, full measurement table (such as grouping measurements for a shopper’s recommended size).
Combining reviews submitted by customers that address sizing with sophisticated AI is the approach used by reverse-logistics experts Stitch Fix (on personal styling) and Rent the Runway (designer rental). Both are dependent on shipping items that are likely to fit, and rely heavily on customer feedback to become smarter over time. Rent the Runway gets size and fit information from nearly every person who rents from them, this data then informs its algorithm, according to the company. When people pick styles that Rent the Runway’s model recommends, the risk of fit issues decreases by 45 per cent. Stitch Fix gathers nuanced details, including how people like their jeans to fit and if certain fabrics are too tight in certain areas. A client might say they’re a US size 6 or 8, but Stitch Fix’s AI can assign a more specific size, for example 6.4 or 8.3.
This data visibility translates into a business opportunity. Both Stitch Fix and Rent the Runway now work directly with brands to create new items that incorporate recommendations on fit and style, as well as using this data to inform their own house brands. Rent the Runway has manufactured exclusive designs with those including Jason Wu, Rosie Assoulin and Derek Lam using this data, which have resulted in high fit and “love” rates, according to the company. Denim brand Pistola developed a denim line exclusively to sell via Stitch Fix and found that the fit of certain styles varied based on fabrics and wash; it then expanded the available sizes and saw a 60 per cent increase in trouser sales in 2022.
Any insights Amazon gains on fit are likely to have an impact not only on its sellers, but also on its own house brands. For now, expect more AI-driven personalisation, Freshwater says. “We will continue to leverage AI to further personalise the shopping experience, making it easier and more enjoyable for our customers.”
Comments, questions or feedback? Email us at feedback@voguebusiness.com.