| This post was kindly contributed by Key Happenings at support.sas.com - go there to comment and to read the full post. |
Customer retention is a vital concern to many industries including financial services, telecommunications, healthcare and insurance. Many organizations use data mining and analytics to analyze customer behavior to learn attrition risk factors most likely to contribute to customer turnover. With that knowledge, tailored marketing can then be targeted to those customers deemed most likely to attrite.
Ward Thomas, General Manager and Director of Analytics at Euro RSCG Discovery, presented a paper at the NorthEast SAS Users Group Conference highlighting compelling reasons to use logistic regression modeling to identify customers at risk to attrite, risk factors most significant for each customer and marketing materials most likely to return them to the fold.
According to Thomas, it costs (in resources and funds) between three and ten times more to acquire a new customer than to keep an existing customer. “Attrition modeling can help you minimize customer defection,” says Thomas. “If you identify customers most likely to leave, you can then focus more of your efforts on those customers.”
Thomas says to look to your database as a historical record that will help you predict the future. Churn differs by industry so you will first have to define the attrition you want to address:
- Are you concerned by reduced account activity? (fewer credit card purchases, falling balance)
- Is the defection passive or active? (reducing use or changing providers)
Now that you know the type of attrition you want to stop, it is time to build a loyalty model. According to Thomas, you should consider all available data including transaction data, customer service data, RFM (recency, frequency and monetary data), surveys (mine for sentiment). It’s also important to enhance your in-house data with external demographic data, wallet share data and macroeconomic information. Use this data to build your loyalty model.
Feed this data into your logistic regression model. The model provides attrition probability outcomes for each customer and gives you the ability to prioritize the customer for loyalty marketing. Thomas slices the data into meaningful groups – deciles – to get a stable estimate of attrition risk of the larger sample.
As you can see from the graph, decile 1 is most likely to attrite.
This chart gives a good representation of who the decile 1 customer is.
Not all at risk customers are the same though. Differentiating the customers based on individual risks is derived by dissecting with additional logistic regression variables. Thomas suggests various combinations in his paper, which is available at online.
Do you have other ideas for using logistic regression modeling for customer loyalty?
| This post was kindly contributed by Key Happenings at support.sas.com - go there to comment and to read the full post. |