High customer churn, few verifiable reasons why.
Use predictive data science to develop a system that accurately scores customers according to their propensity to churn.
A Data-Driven Look at Customer Attrition
With more than a terabyte of new data flowing into its “Big Data stack” every month – in addition to the vast store of historical customer data in its data warehouse, which includes order, billing, provisioning and financial reporting information – this leading provider of security and automation solutions for homes and businesses does not lack for data.
What it had in shorter supply, however, was a way to make that data useful for the business, uncovering insights about customers and prospects, decreasing customer churn, improving marketing efficiency and increasing customer lifetime value.
To address these needs, the company sought a partner that could nimbly apply data science and predictive data analytics approaches and pivot with them as the data exposed patterns, presented fresh challenges, and unlocked new opportunities. It knew it was sitting on a gold mine of customer data, if it could only make that data operational. Combining a cost-efficient approach, the best techniques and software in data science, a knack for data detective work and discovery, and a powerful platform that the company could use to run models to “score” millions of customers, TROVE proved to be that partner.
Understanding “Propensity to Churn”
Like most companies in the smart home and digital subscription-services space, this leader is constantly wrestling with customer churn and saw in predictive analytics a way to get in front of the problem. With TROVE’s help, it sought to answer, with precision, the following questions:
To begin, TROVE’s Science Squad worked with the company’s IT team to understand the pedigree of the data in its data warehouse, an exhaustive effort to make sense of, clean and normalize a large and unwieldy set of customer data spawned by numerous legacy systems and amassed over several acquisitions and across multiple channel partners. After an intensive effort to attain necessary data quality, TROVE merged the company’s customer data with its own proprietary store of 3rd party data – which includes 650+ consumer and commercial attributes drawn from six reputable sources – to develop a comprehensive, 360-degree view of each customer.
Drawing on “survival modeling” used by epidemiologists to forecast the growth and decline of wildlife populations, the TROVE Solver helped the company identify and flag risk factors that were driving churn – including a customer’s payment method, installation fee, zip code, annual income, number of call center interactions, problem resolution record, and even the number of children and pets they had.
The Solver assigned each customer a score that the company’s team could reference to more accurately identify an individual’s propensity to churn over the next 30-180 days. With this information in hand, the company could start offering personalized treatment plans for at-risk customers or execute larger-scale outbound marketing initiatives, including e-mail offers targeted to the entire at-risk pool.
Dramatic Results, A Desire for More
TROVE’s data-driven approach has provided the company many useful results, some confirming the firm’s hunches, others upending them. For example, the TROVE Customer Retention Solver confirmed that customers on a recurring billing plan, where fees are pulled automatically from a credit card, were more likely to stay than customers who pay by check, and that rural and suburban customers typically churned less than those in the city – conclusions the company had long suspected but wanted to verify.
Yet when it came to understanding the effectiveness of one of the company’s largest promotions, TROVE really showed the clarifying power of data science. One of the company’s biggest promotions featured free installation. While this offer attracted large numbers of new customers, and was widely viewed as a major success within the company, the attrition rate among these new customers proved equally high. A deeper dive into the data provided the company a valuable, unexpected insight: customers investing $400 or more for installation churned at a dramatically lower rate than those customers who paid nothing at all.
Data science also helped the company stop an ineffective marketing campaign in its tracks, saving it hundreds of thousands of dollars. While analysts at the company had helped determine the average income of its “ideal customer,” further analysis revealed that “average income” was actually the dividing line between two distinct customer segments. Each segment bought different products and services and would require separate, tailored marketing campaigns to reach them, not the one homogenous campaign the company had launched.
Insights like these have won TROVE believers across the company and are driving the next phase of predictive analytics there, focusing on customer lifetime value (CLV). By understanding a customer’s propensity to churn and combining that information with a customer’s buying history, TROVE can model the expected lifetime value, or revenue potential, of each customer. The company plans to use this value to optimize its customer portfolio management capabilities, enabling it to better service customers and monetize sales opportunities across its base.
TROVE’s Predictive Data Science Platform is built on four key pillars: Datasynthesis™, Solvers™, Frictionless Functionality™ and Science Squad™. Here’s how each factors into making data useful at this company: