How do online returns influence the retailer’s performance?

Online shopping is becoming more and more popular. Consumers clearly like the convenience of shopping from anywhere. A downside of this development is the increasing number of returned products. Although online shopping has many benefits for consumers, they can only physically inspect the product they ordered after receiving it at home. And when it then doesn’t satisfy expectations, the product gets returned. In some industries, the return percentage reach almost 50%!

Returned products are often eligible for resale, and can thus be added to the current inventory. This also means that a retailer has to take possible returns into account when making decisions about inventory management. Not considering these potential return flows would result in excessive inventory with corresponding high costs. However, predicting when and how many products will be returned is not easy! What we do know is that the return flow depends on the timing and quantity of past sales. Using this information to predict the return flow and use it to improve inventory management is however more difficult than you might think. Figuring out how to do this then also requires state-of-the-art data science techniques, building on machine learning and operations research methods.

In a new study, we focus on this problem for a brick-and-mortar store that uses their store inventory to fulfil both in-store and online demand. The retailer receives product returns from the online sales. We model the inventory decisions as a Markov Decision Process that maximises the retailer’s expected profit. However, such a mathematical modelling approach becomes very difficult to solve when it has to keep track of all the detailed online sales data to be able to consider possible returns. Therefore, we also develop a solution approach based on Deep Reinforcement Learning to find an approximate solution. We observe that our Deep Reinforcement Learning algorithm scales well to the studied problem and outperforms other methods. The new algorithm also allowed us to derive several interesting managerial insights. For instance, it turns out that high return rates for online sales have a negative impact on the service experienced by customers shopping in the stores, as retailers decrease store inventory since online sales become less profitable. Also, in our experiments, the length of the return window turns out to have less influence on the retailer’s profit than we expected.

More details on the study can be found in the full paper available on the publisher's website:

• Goedhart, J., Haijema, R., Akkerman, R. (2022), Influence of returns for an omni-channel retailer, , available online in open access format at https://doi.org/10.1016/j.ejor.2022.08.021.