Part 1C: How Do I Better Market to my Existing Customer Base?

New Call-to-action

Part 1C: How Do I Better Market to my Existing Customer Base?

 Propensity Scoring

 Similar to pricing gap analysis, this third tool (propensity scoring) in our belt takes attributes about each customer and specifically assesses how likely each person is to adopt an additional loan product from each category that our institution offers. However, pricing gap just allows us to filter by what we thought we should be offering. The propensity scoring model gives us more marketing power as we can calculate real probability-based estimates.

 Using a touch more of statistical savviness, we employ a Random Forest machine learning algorithm or logistic regression. Without getting too bogged down in the details, these models simply allow us to identify attributes that correlate most strongly with people who open loans under given conditions. This way we gain a better perspective of our customer base - who’s doing what and when.

 We can use this powerful tool in one of two ways.

 First, we can identify propensity scores for all loan type categories at the individual level. For example, below we see a customer from our sample data set, John Doe, and his associated borrower attributes:


Here we see that John has no mortgage with any institution, he has a credit score of 715, we know his age and his income. So what would we predict John to be most interested in? Employing one of these propensity models, we could predict that John would be most interested in a mortgage loan, the loan type with the highest propensity score.  From these scores we could also see that John would be least interested in an auto loan, as we note he has two auto loans currently.


This propensity scoring tool is powerful because at the individual level, we know what John wants but also what John needs. So if John walks into the branch, we don’t waste our time trying to talk him into an auto loan. Instead we could offer him a mortgage loan on favorable terms, a marketing “win” on all fronts. 

 Second, propensity scoring can also be used at the aggregate level, giving us a holistic view of our entire customer base. Suppose our institution wants to start a targeted campaign for auto loans. However, hoping to ensure that none of our resource is wasted, we only want to market to customers who are most likely to open an auto loan. Propensity scoring at the aggregate level provides us just the tool. Because we assign a propensity score for every single customer in our database, we could then go into a customer list and sort by propensity score. This view isolates existing customers who have the highest probability to open a new auto loan. We then target our marketing efforts at these customers with confidence that our application hit rates will be higher than that of broader campaigns.

 With a better understanding of wallet share and potential leakage, implementation of pricing gap analysis and expected interest rates, and finally propensity scoring for real likelihood of new loans, we have a better handle on our customer base and marketing strategy.

 But now that we have our own customers within targeted marketing reach, how do we branch out to those outside our customer base? Stay tuned for part II where we explore the outer regions of Customer Analytics.

Keaton Baughan

Product Manager