Retailers are constantly trying to figure out not just how to sell their current inventory, but also what to offer to consumers for the next season. Thanks to new research from a University of Maryland marketing professor, predicting what consumers want could get easier.
Jie Zhang, Professor of Marketing and the Harvey Sanders Fellow of Retail Management at the Robert H. Smith School of Business, worked with former Smith PhD student Min Kim, now an assistant professor of marketing at the National University of Singapore. Their research, in the Journal of Marketing Research, develops a new machine-learning model that takes into account the specific attributes of products a consumer likes. It uses those attributes to predict what other sorts of products consumers will want in the future to help retailers with large and frequently changing store assortments decide what to stock next and how to personalize product recommendations.
“This methodology is able to predict that out of the future products that a retailer may sell, which ones would offer the closest fit to what customers want,” Zhang says.
Zhang and Kim used data from an online deal marketplace that sells an ever-changing assortment of brand-name clothing, jewelry, handbags, and shoes at discounted prices. The researchers looked at individual shoppers’ browsing and purchase data to figure out what product attributes – category, brand, color, price, depth of discount, etc. - drive their browsing and purchase decisions in the entire store.
“As long as those attributes have existed in an individual product in the past, that preference can be picked up by our model and utilized to make preference predictions for new products that are not in a retailer’s existing assortment or currently in the data they analyze,” Zhang says.
These insights can help retailers improve merchandise planning or make better personalized product recommendations for shoppers, she says. Many online retailers already use proprietary algorithms to show shoppers other products they might like, but most of those algorithms are applicable for products that already exist. The new algorithm developed by the researchers is capable of helping retailers come up with products a customer might want that they don’t even offer yet.
It works, she says, because even though consumers hardly ever buy exactly the same apparel item twice, the reasons why they like certain items tend to stay consistent.
“Preferences at the product attribute level – favorite brands, favorite styles, favorite colors – are more enduring and stable than a shopper’s preference for individual items,” she says.
Having that information readily available would be very valuable to retailers who are constantly changing their assortments.
Right now, retailers are using a combination of art and science to figure out how to predict what consumers will want next, says Zhang. They analyze purchase data to draw insights about what consumers like, along with the gut feelings of buyers and observations from social media, pop culture, celebrities and other sources.
“A lot of retail and marketing analytics are built on data of existing products,” Zhang says. “We provide a new approach for making predictions about products that haven’t even been made yet, let alone been seen by anybody. Our methodology draws insights from browsing and purchase data of existing products to predict consumers’ future product preferences.”
The research is also one of the first papers to delve into the reasons why shoppers are attracted to online deal marketplaces like TJMaxx.com and NordstromRack.com – or any similar retailer that acts as an industry scavenger to procure close-out, surplus, irregular, and out-of-season merchandise from other retailers and manufacturers at deeply discounted wholesale prices. They can still offer low prices to customers while making big profits.
Zhang and Kim looked at what draws consumers to a deal marketplace – the premium brands, deep discounts, or absolutely low prices? Consumers tend to be either prestige-oriented shoppers looking for premium brands, or die-hard bargain-hunters looking for the best deals. But what draws people to browse might not be what they end up buying, Zhang says: “When it comes to the purchase phase, most consumers actually end up buying things that are of moderate initial prices with moderate discounts.”
The takeaway for these retailers is that they need to have the full gamut of merchandise to draw people in, she says, but they also need to have enough middle-of-the-road value merchandise to entice shoppers to make purchases.
The research method was designed and tested in the e-commerce field, but even brick-and-mortar retailers that have frequently changing assortments could use it, Zhang says.
The model works particularly well for online retailers because it uses both browsing and purchase data.
“In general, people browse much more than they make purchases at any e-commerce website,” Zhang said. “If you only use the purchase data, some methodologies become less reliable because they don’t have enough observations of an individual’s repeat purchase decisions.”
All of that extra data makes the model so much more precise at being able to predict what consumers will go for in the future – the more accurate, the more beneficial for consumers on the receiving end of personalized product recommendations.
“It’s not easy – especially for apparel products,” Zhang said. “If retailers utilize their data and analytics tools right to provide better products and services for consumers, it’s good for everyone.”
Read the research, “Discovering Online Shopping Preference Structures in Large and Frequently Changing Store Assortments,” in the Journal of Marketing Research.
Media Relations Manager