Deep learning approaches for customer analytics

In recent years, large language models like ChatGPT have become part of everyday life for many of us. Transformers (the ‘T’ in ChatGPT) are advanced types of neural networks that form the core of these large language models. Although transformers are particularly known for their capability to learn language patterns, their design is not limited to processing language sequences. In fact, transformers can be trained to learn patterns in any type of sequence, and generate new sequences based on these learned patterns. This flexibility opens new possibilities for customer analytics. In our research, we apply modern deep learning techniques (like transformers) to understand shopping behavior. For example, we analyze the scenario of customers filling a shopping basket in an online supermarket. Similar to how transformers can learn to predict a missing word in a sentence given the context, we can also train them to predict a missing product in a shopping basket given the products already in the basket. Those predictions can then be used for personalized recommendations to the customer, which can increase revenue and improve the customer experience. We demonstrate that our deep learning approach outperforms traditional, statistical marketing methods in predicting purchases. The same holds for the approaches we propose to model other types of customer behavior, such as online search and click behavior prior to a purchase. Together, our findings show how advanced deep learning techniques can enable companies to better understand their customers in an increasingly digital marketplace.