Leveraging the Power of Images

Using image processing techniques and including the product image in a predictive model can reduce retail return rates and increase profits by 8%, according to new research from Frankfurt School of Finance & Management.

The researcher stated that in the $500 billion fashion industry, return rates of over 50% are the norm, and vary greatly by item. Being able to accurately predict return rates prelaunch is a real game changer in online retailing. This could not only save clothing retailers millions of pounds a year, but also decrease the environmental effects of clothing returns, therefore having benefits for companies as well as wider society.

Many current return rates predictive models simply use the characteristics of a piece of clothing, such as its product category, price, and season. However, the researchers experimented with leveraging the power of the product image, to include the visual features of an item as well. In doing so, they found that there was a much more accurate prediction of how many items will be returned, which resulted in an 8.3% increase in profits.

These findings come from research by Siham El Kihal, Assistant Professor of Marketing at Frankfurt School of Finance & Management, who investigated, together with a co-authors team from MIT Sloan, Emory, and New Economics School, how to make the clothing industry more efficient whilst reducing the number of returns companies experienced.

To do so, the research team used a large dataset from a European apparel retailer to observe purchases and returns and developed a machine learning model that incorporates the product image, using state-of-the art image analytics methods. In the study, the researchers found that return rates of items bought online are an average of 53% of clothing – meaning many items are not profitable. This is in contrast with the 3% return rate in the same retailer’s offline channel, with the same set of items. 

“The fashion industry has a huge problem with sustainability,” says Professor El Kihal, “Newer technologies like machine learning models can be implemented to make the company drastically more efficient, and companies should be looking to invest in this type of innovation”.

Professor El Kihal suggests these findings are not only applicable to the clothing apparel industry, but to others too such as hospitality, furniture, real estate, and even groceries.

The paper was published in Marketing Science on 24 July. You can read the full study here.