In the previous blog post about Standardizing Credit Ratings, we discussed integrating multiple agencies’ scores. In addition to integrating scores, you might need to build and integrate your own custom scorecard that considers your organization risks appetite, compliance, and regulatory requirements. In this article, we discuss how to model and build a custom scorecard.

Scorecards help financial institutions to consistently validate applications to ensure the applicant has the ability to pay back the credit. The credit-checking process contains several steps in which velocity and accuracy are crucial.

When you apply for any type of loan, the lender will analyze your credit score. The challenge is to generate the Credit Score based above-mentioned factors. Because of these reasons, the approval credit card process can be a complicated and lengthy one.

The Approval Scoring Model

In general, a scorecard requires multiple inputs. For instance, creating a Decision Graph that produces a credit card approval decision, requires input data such as annual income, and assets. Based on the Current Date, Annual Income, and Assets it calculates a Standard Card Score and then determines the approval decision based on the Standard Card Score.

In this example we are going to model a Credit Card Approval Decision that uses decision graph to represent business decision around approval or rejection of a credit card request. The Decision Graph frames the approval business decision in to its decision units such as the decisions, inputs, and knowledge sources we require. In this mode:

  • The Inputs are the Annual Income and Assets.
  • The Knowledge Source is Date Validity based on the current date.
  • The Decisions will calculate Standard Credit Card Score and Automatic Approval outcome.

IDRcredicprocess

This Credit Card Approval model depicts in blue round rectangle, it consists of two decision units:

  1. Calculate Standard Card Score
  2. Automatic Approval Result

Calculating Standard Credit Card Score

In this example we use Decision Table to model the Standard Credit Card Score. For calculating standard card scores, in addition to two input values (annual income, and assets) it needs to get the current application date from the system. In this scenario, the Date Validity is pulled in as a condition column.

We have added the current date as a condition to the decision table to decide whether the current date falls under the accepted date range. If the date needs to be controlled more granularly, the decision table needs to reference another input parameter for the date.

Automatic Approval

Based on the calculated Standard Credit Card Score, the Automatic Approval decision should be made. The business rule of this decison unit is very simple. If the score is greater than or equal to 350, the card request is approved, otherwise its rejected. In FlexRule you can use many different ways to model business rules such as Natural Language, Boxed Expression, Expression Language and may other forms.

We used a Natural Language model for this rule implementation as below:

NLif

Or for instance, you can use only a pure FlexRule Expression to model this simple business rule as below:

ExpressionIf

In either case, they will be linked in the Decision Graph to the “Automatic Approval” decision. and the Decision graph is responsible to orchestrate between multipole decision units and execute them as required.

To explore more about this project, visit FlexRule Resource Hub.

Conclusion

Building your own custom scoring is a powerful approach to ensure your organization's requirements around law, regulations, and compliance while ensuring your organization's risk appetite is met. Once the custom score is ready, you can use an orchestration model to create an overall process to include the unification of multiple agencies' ratings with the custom scorecard into a single decision service.

Once your custom scorecard is published as a REST API, then every process and application can execute the decision, and it will act as a single source of truth for approving or declining an application. This approach eliminates inconsistency between agents taking care of clients and makes your process scalable with high accuracy.

Last updated December 16th, 2022 at 10:57 am, Published December 14th, 2022 at 10:57 am