How We Used Predictive Modeling to Retain At-Risk Customers for RevLocal
Our client, RevLocal, is a digital marketing agency that provides personalized digital marketing strategies for small businesses. Though serving a large, diverse suite of high-profile customers, they were experiencing a higher-than-desired customer churn rate, resulting in significant financial impact.
A Need to Predict Customer Behavior
They were looking for a way to anticipate when and why customers might leave before they actually leave, allowing more opportunities to “rescue” and retain these “at-risk” customers with targeted content and agent interactions.
RevLocal determined a predictive analytics model could help them measure the probability that a customer will leave in the near future. They could use this model to identify at-risk customers and address contributing factors. But to develop this predictive model, they needed the help of a partner. Aware of Centric Consulting’s deep experience with machine learning and predictive analytics, they sought our expertise to help them retain more customers, and in turn, increase their revenue.
Enter Centric: From Proof of Concept to Predictive Model
At the start of the project, we commended that we work closely with RevLocal to develop a Proof of Concept (POC). This would increase the reliability and accuracy of the predictive model, establish an important technical foundation, and help RevLocal further support their investment with measurable data. We followed a number of steps as we put together the POC:
- Workshop. The first step was goal setting. We facilitated a day-long workshop with key RevLocal employees to determine clear objectives for the predictive model. Through survey sessions, we were also able to identify which factors RevLocal employees believe to be predictive of a customer’s departure, based on their own experiences or instincts. Example factors include: the agent the customer was working with and whether they consistently changed, the age of the account, and changes in price.
- Data Inventory. The next step was to take an inventory of all available customer data, and measure its availability, quality, quantity of data to ascertain whether it could feed into the predictive model.
- Predictive Model. In the final step, we developed an initial, time-box predictive model able to identify specific clients and customers at risk for departure.
To know if the POC actually works, we also conducted a series of tests comparing the model’s findings with data reports and client departure rates during the same time period. The results of these tests would further validate the model’s features, accuracy and demonstration of predictive capability.
Using Results to Rescue Customers
After a three month-testing period, the POC Model was able to accurately identify 20 to 30 percent of account attrition. Using these results, we compiled a list of factors that were most likely to contribute to a customer’s departure.
Factors that were more likely to put customers “at risk” for leaving included:
- If they regularly get a new strategist or consultant
- If their original list price changes within a short time period
- How long their account has existed and how many months they are into their contract
- Their industry.
The list provides insight into the types of behavior that can deter customer longevity and retention. RevLocal can use it to improve the factors that may cause an account to leave, allowing consultants to “rescue” customers before they can depart through timely and more personalized interactions, increasing customer satisfaction and revenue as a result.
