Custom Scorecards: What’s The Big Deal?

Custom models that focus on the customers in your portfolio are a great way to accurately predict payment behavior.


In the race to win market share, there are plenty of tools to leverage in your credit policy. You can compete on price, but margins are already tight. You can buy deeper into your score grades, but you must be mindful of loss rates. You can decrease response time to capture more business, but that requires confidence in a high auto-decisioning rate.

What if you could redefine your score grades without incurring additional risk? That would be ideal, right?

Generic scorecards that use large, general populations are good at predicting payment behavior on an average population, but custom models that focus only on the customers in your portfolio are even better. By using the origination and payment data from your credit union’s members, you can identify the distinct characteristics that count most when decisioning YOUR applications.

First Things First

Most credit unions already use a generic scorecard from a credit bureau in their underwriting process. Before you get your heart set on a custom scorecard, you need to understand how your current scores are working. You can accomplish this by completing a score validation, which will tell you how well your current scores are performing by going back to the time of application and measuring the actual payment behavior of your customer. A well-performing score should rank risk as well as the odds charts predicted.  

Add Another Dimension

The generic bureau models only account for data from the credit report, so if your bureau model is working as it should, it might make sense to consider an application model. You are likely already using the application data in a range or capped format. For example, you decline applications with a monthly income of less than $1,500, and you require at least 12 months of employment. An application score identifies the variables in the application that are most likely to predict default. Therefore, you can choose elements and ranges empirically instead of judgmentally and score the application data in seconds. You can use this information to determine whether you want to pull a credit bureau report after the application or automatically pull a credit bureau report and use that score in a matrix to create a total score grade. This allows you to buy deeper without increasing risk because you will be able to approve customers who have lower bureau scores that are offset by higher application scores. 

Take The Next Step

Now that you have a score matrix, you should see more bad accounts in the lower scores and more good accounts in the higher scores. For even more lift, replace the generic bureau model with a custom model. This type of approach requires a layered implementation in which you add one score and observe the process while developing and implementing a second score, if desired. Throughout this process, use your institution’s risk tolerance and approval rate targets to drive the strategy.

Sounds Great! Where Do I Sign Up?

Credit unions must review many details when they consider a custom model. After all, a custom model is not always an appropriate solution for every institution or product. Details to consider include:

  • Data availability — Most scorecards are built on the concept that past behavior is the best predictor of future behavior, so it is critical that there is at least two years of application data and at least one year of performance data available to collect and use in this process. 
  • Data volumes — Because the poor performers are the most telling players in the population, it is important that there are enough of them in the sample. Depending on delinquency rate, this could equate to 2,000 or more applications per month for the time frame listed above. A delinquency rate of less than 1% might require more applications.
  • Credit policy consistency — If your credit union has drastically changed its membership parameters or has significantly changed products or policy, proceed with caution.
  • Cost — Scorecards are usually developed at the product level. Development can cost $25,000 or more per product, so consider the gains you want to achieve with a custom model. 
With more than 23 years of combined experience in decision analytics, Kendall Keeling uses her analytics expertise to serve the clients of CRIF Achieve, the decision management division of CRIF Lending Solutions. To learn more about CRIF Lending Solutions’ automated lending solutions, visit



This sponsored content article is provided to the credit union community for shared insights and knowledge from a recognized solutions provider in the industry. Please note that the views and opinions offered here do not reflect those of Callahan & Associates, and Callahan does not endorse vendors or the solutions they offer.

If you are interested in contributing an article on, please contact our Callahan Media team at or 1-800-446-7453.


July 20, 2018



No comments have been posted yet. Be the first one.