Customer churn represents lost opportunities as customers move their business to one of your competitors. At Aunalytics, we have nearly a decade of experience developing such models for clients in various industries, including financial institutions. For client subscribers to our Daybreak™ Analytical Database, the churn propensity report Smart Feature^TM offers a uniquely valuable dataset for understanding the risk factors that could indicate customers are about to take their business elsewhere.

The algorithm behind this Smart Feature was developed by studying historical customer churn data from many different client institutions to discover common patterns in the typical behaviors that precede account churn. One key approach to developing this algorithm was to redefine churn beyond simply customers who closed an account; instead, churn was defined to include customers who suddenly drew down a high-balance account and never returned that account balance to its previous level after a period of months. By looking at the data in this way, our data scientists were able to understand that the behaviors that predict churn often happen many months or years before an account is actually closed—for example, a customer stops using direct deposit or bill pay services, then withdraws most of the balance a few weeks to months later but doesn’t actually close the account for a few years.

Based on this deeper look at the data, Aunalytics has been able to develop an algorithm that looks at numerous factors such as customer transactional data, the age of the account, the presence of ACH deposits prior to churn, number of debit card transactions per month, online banking login frequency, and length of the total customer relationship, among others. Once this algorithm is trained on a client’s data, we can provide that client with propensity to churn scores expressed both as a standard percentile (e.g. absolute probability in the range of 0-100%) and a decile score (e.g. a score from 1-10 where 1 is the group composed of the 10% of customers least likely to churn and 10 is the 10% most likely to churn). Using these Smart Features, our clients in financial services can understand immediately what the likelihood is that a customer will churn, or build datasets to target customers at risk of churn with offers and incentives that might keep them actively using those financial products.