Demo
In today’s world, where technology is developing with increasing momentum, many digital traces (records) are formed while daily operations continue. Digital traces are produced from various sources such as social media platforms, mobile phone applications, internet browsers, navigation systems, shopping platforms, game platforms, ERP systems of stores, and search engines. These traces are stored in many different data hosting environments.
Today, many businesses conduct descriptive and predictive analyses to obtain insights from these digital traces and apply them in business processes such as sales, marketing, supply chain, and customer relationship management.
While different data types are produced from many sources, transactions such as direct contacts, purchases, and complaints between a business and its customers have a high explanatory power. Keeping a record and storing these transactions relationally is relatively easy.
However, in certain sectors such as furniture, durable consumer goods, and automotive, where long-lasting products are sold, direct interactions with customers occur at rare intervals, sometimes spanning many years. This low interaction frequency makes it difficult to calculate key customer relationship management (CRM) metrics such as customer satisfaction, customer retention, and customer churn.
Customer churn analysis is widely applied in businesses that operate under subscription models, such as credit cards, mobile phone networks, digital entertainment, and gaming platforms. One of the main reasons for performing churn analysis in these sectors is that unsubscribe records can be clearly tracked and stored.
In the literature, many customer churn analysis approaches exist for sectors that do not operate with a subscription model. However, conducting such studies is significantly more challenging for businesses where shopping frequency is infrequent.
This study explores predictive customer churn analysis on the customers of a business with infrequent shopping frequency. Additionally, alternative approaches for assigning the lost customer label were tested and evaluated.
