Assurant Solutions sells insurance policies on credit card payments, for example if you lose your job or become disabled. Their customer service center takes calls from people who see the monthly add-on fee on their bill and call to cancel. Metrics are simple. If a customer does not cancel, the customer is “saved.” If a customer is saved by downselling or upselling to a different plan, a percentage of revenue is saved. Assurant’s objective was to increase the percentage of saved customers and to increase the percentage of saved revenue. To do this, they agreed to abandon their assumptions about how the customer service center should operate in favor of what data mining revealed. MIT Sloan Management Review published this article detailing their astounding results.
Assurant was routing calls to customer service representatives (CSRs) based on availability and their assessment of CSR knowledge of the product in question – most of which was anecdotal. They had also embraced some industry “best practices” and implemented guidelines so that 80 percent of calls were answered in 20 seconds and customer hold times were kept to a minimum. A key difference in the data-driven approach would be the ability to base decisions on successful outcomes such as customers and revenue saved, rather than on efficiency measures such as time to answer and customer hold time.
Assurant benefited from being an insurance company with actuaries and mathematicians they could call upon for help in doing the data analysis. They also benefited from having large, comprehensive databases on customers, customer calls and CSRs, which allowed them to analyze detailed information about customers and CSRs as well as every customer call from the previous five years.
As it turns out, some CSRs are more successful with lower monthly premium accounts. Others can’t save low monthly premium accounts but have great success with higher monthly premium accounts. Based on a number of different factors such as monthly premium, credit score, amount financed, and percentage of credit used, the data mining exercise began to reveal patterns in the data. The result was a predictive model that scored CSRs on their likelihood of saving a particular customer. The model was complex and comprehensive. For example, it considered how valuable a customer was, the difference between the “best” and “next best” CSR success rates, estimated call durations for specific CSRs and customer tolerance for hold time. Assurant likened the process to a “sales matchmaking” between CSRs and customers. They didn’t know exactly why some CSRs did better with certain groups of customers, nor did they care. They simply knew that it was true because it was evidenced by the data.
At the beginning of the project, the customer save percentage and percentage of revenue saved were both 16 percent – already at the top of the industry standard. From the very first day the model was implemented, Assurant watched their results improve. Even better was that the greatest improvements were made with customers who paid the highest monthly premiums. Overall, the customer save percentage rose to about 32 percent and the percentage of revenue saved jumped to about 50 percent. As more valuable customers stayed customers for longer, this also translated into an impact on future revenue using a customer lifetime value model.
Assurant already had the knowledge about how to best route customer calls to CSRs in order to save the most customers and revenue, but it was hiding inside their databases of calls, customers and CSRs. What knowledge do you think is hiding inside your databases that you have yet to discover?
Subscribe to our blog to stay informed on data mining and other data analytics topics.
Spotfire Blogging Team