Unleashing the power of churn prediction algorithms

Churn is undesirable. It indicates the rate of attrition or how many customers (or employees) are leaving an organization. It is estimated that, in the insurance industry in the UK churn in customers for home and auto insurance only results in a loss of £3.3B. Hence, companies focus on reducing churn. It would be easier to reduce churn if one were able to predict it in a timely manner.
“Churn prediction” is one of many tasks that are successfully solved by machine learning (ML). Big companies, such as banks, telecoms are already successfully implementing churn prediction systems. Many executives are curious about churn prediction and if it is suited for their business. Some are familiar with ML but not with its churn prediction capabilities. In this article, we describe how churn prediction works and its uses for various industries. We also briefly outline the steps needed to adopt churn prediction for your organization.
Depending on the context, various types of “churn” can be defined. In the telecom sector, “customers churn”, or customers signing up for a competitor service, is often a concern. Some researchers in telecom sector describe churn types based on customer behaviors e.g., “credit card churn” means about 3% of customers churn each month as their credit cards expire. In banking or insurance sector, “employees churn” or the overall turnover of employees is an important factor. Losing employees is costly as hiring and training activities incur expenses.
Employee churn prediction results in better organization of culture
Churn prediction is done using predictive modeling. This is a type of ML algorithm that is generally developed in three steps. Initially, historical customer data that include information about churned customers and retained customers are collected. Once the data are prepared, a predictive model is built using R, Python or similar languages. Finally, churn predictions are made along with its confidence interval.
Benefits of smart retaining of a customer are numerous. Churn prediction enables the companies to take necessary steps such as customer incentives, new loyalty programs etc. to minimize churn rate. Some companies make targeted and non-targeted (mass) efforts. For instance, sending email reminders to a customer who has not made a purchase in last two months. Or non-targeted effort such as providing all potential churn customers with a discount code.
Having employee churn prediction capabilities results in an overall better organization of culture. Churn prediction enables employers to see patterns of hiring (and firing) of employees. If a particular department of the organization is not able to retain employees, churn prediction is able to detect it. This allows for timely actions such as employee training, improving hiring practices etc.

In the retail sector, attracting new customers is an expensive proposition. Hence, businesses in retail industry focus on retaining existing customers. In this industry, churn is determined by a number of days lapsed between purchases and amount of purchases. For instance, if a customer does not buy items worth $20 in last 3 months then she would be considered churned.
For telecom sector, the predictors of customer churn are related to usage patterns. For example, number of calls made, international calls, daytime calls, number of voicemails etc. Telecom sector includes services such as cable operators, wireless services, and internet service providers. These providers use predictors accordingly (e.g., data used, channel package subscribed etc.).
When it comes to the banking sector, a study conducted in 2012, revealed that about 50% customers changed or wanted to change their bank. The banks need to turn this around by creating a positive and consistent experience for their customers. Churn prediction identifies customers that are may potentially churn (based on predictors such as number of calls to helpline, account balance etc.) and allows banks to take steps to retain them. For instance, first contact to resolve a complaint was identified as a trigger to churn. So many banks focus on resolving complaints at first contact. Additionally, churn prediction is also useful in detecting customers that may default a loan.
Organizations need to explore effective retention options
The willingness to adopt churn prediction for your organization is the first step towards a positive change. Next the companies need to take some concrete steps such as defining business goals. This requires a tight collaboration between the management and their analysts. To accomplish those business goals, a data science team is hired, which prepares and analyzes existing data. The data science team builds and tests churn predicting models. The new insight gained from these models allows to further explore effective retention options, which may need analytics, such as A/B testing.
A good data science company partners with their client company every step of the way to reach ultimate goals. Data science companies such as BitRefine group provide both the services – consulting and building ML solutions. A continued collaboration with a data science company helps the client organization to meet its goals of churn prediction and thereby revenue generation and employee retention.

October 17, 2017