Product returns are a serious problem for online retailers. According to a survey by the National Retail Federation, online return rates are 21 percent higher than return rates from all other channels. And the financial hit from restocking fees and the disposal of un-resellable items is substantial, as much as 30 to 40 percent of a company’s expenses. In 2024 alone, that amounted to an industry-wide total of $890 billion.
Back in 2018, Youngeui Kim, a PhD student at the Lubar School of Business at the University of Wisconsin-Milwaukee, began discussing the problem of returns with a few of her professors, including Abhijeet Ghoshal, then an assistant professor at Lubar.
“It struck us that people actually return a lot of products because they don’t find the product to be satisfying for whatever reason,” Ghoshal says now. “And often product reviews say the reasons for that. We wondered if there was something we could find out by studying the product reviews.”
The project continued after Kim graduated and took a job as assistant professor of computer information systems at the Walker College of Business at Appalachian State University and Ghoshal moved to Gies College of Business at the University of Illinois Urbana-Champaign, where he’s currently an associate professor of business administration. Ghoshal, Kim, and their third collaborator, Yang Wang of Lubar, noticed that some product reviews were extremely descriptive, especially if something had gone wrong with the product. Some reviewers even posted photos.
“So we wondered,” Ghoshal says, “is there some information in the pictures beyond the text of the review? And will that information help consumers make better decisions? Because if they make better decisions, the return rate should decrease.”
Why Real-World Product Photos Help Reduce Returns
This seemed to make sense. After all, retailers stage photoshoots to show off their products to their best advantage. But they may not give customers an accurate sense of what the product may look like when it’s not worn by a professional model under studio lighting. A swimsuit, for instance, should look good when it’s brand-new and worn by a model. But how does it look on someone with your particular body type? How does it hold up once it gets wet?
The researchers obtained a dataset of product reviews and sales data from an outdoor speciality online retailer, collected between 2013 and 2015, with a two-month window for returns. Using IBM Watson’s object detection function, they analyzed the reviews. Specifically, they were looking for review photos that showed the product in context, the way it would be used by a real person. Many of these review photos demonstrated the best environment where it should be used (like a parka in a snowy winter landscape) or how someone actually used it (like a rope for a rock climber).
Then, by employing econometric modeling to compare the number of product review with contextual information with sales and return data, they determined that customers were less likely to return an item if they saw reviews that also contained photos of the item in context, rather than just text or a photo of the item fresh out of the box. The exception was text reviews that mentioned how well an item fit (or didn’t).
But why was that happening? The researchers decided to conduct an experiment to find out what customers were thinking.
In the experiment, subjects were asked to read a review of a hiking jacket and then answer some questions about how warm they thought the jacket would be. One group received a text with contextual cues, another received an image with contextual cues, and the final group received no contextual information at all. Then the subjects were asked to look at two reviews, one with text only and one with an image, and asked to determine which was most helpful. The results showed that customers found that reviews with contextual cues of any kind were useful in evaluating how warm the jacket was but that images were more helpful than text.
“We found that the more information coming in, the fewer returns for the companies,” Ghoshal says. “Pictures are useful for figuring out the functionality of the products and whether they meet their needs. Text is also useful for determining functionality, but the effect is larger for the pictures.” This is because reading is a more cognitively complex task than simply looking. “The mind can process pictures much faster than reading text,” Ghoshal explains.
What Retailers Can Do to Encourage Better Product Reviews
In order to cut down returns in the future, the three researchers recommend that online retailers create an environment where it’s not only easy to upload photos in reviews but also encouraged. The more information people have, the more likely they’ll be able to pick out the right product on the first try.
Sellers should also promote honest reviews. As the researchers write in the paper, “Focusing solely on collecting 5-star ratings may not yield long-term benefits for the product’s net sales. Instead, sellers should encourage reviewers to discuss the environment or situation in which they use the product. These contextual cues can help shoppers evaluate product aesthetics more effectively, thereby reducing returns.”
The effectiveness of some products, like computers or dietary supplements, for example, may be more difficult to demonstrate through photos. But Ghoshal encourages customers to think of other visual forms, like graphs, which are still easier for brains to process than text.
Given the age of the dataset, Ghoshal says the researchers have no further plans to continue this line of research. But they feel confident that their recommendations will be helpful for reducing returns.
“The happiness of customers is a very important thing,” he says. “All companies want their customers to be happy.”