Skip to main content
Breaking down an origination decision model

While decision modeling and optimization are employed by top financial services firms in some decision areas, they have not been a standard part of the origination process for most lenders. But this is changing—not only because of economic, regulatory and market pressures, but because technology advances (e.g., SOA approaches) now enable lenders of all sizes to bring advanced analytics into decisioning.

Decision modeling has been a frequent topic of mine on this blog as a fundamental technique for understanding and improving decision strategies. It's probably worth a refresher to answer: what is an origination decision model? In essence, it's a mathematical representation of the relationships between the data you have, the decisions you need to make, the likely outcomes of those decisions and the ultimate goal, such as profitability.

Here's an example of a simplified credit card origination decision model. The yellow ovals represent inputs to the decision, the blue ovals represent predicted outcomes for the action (accept or reject), and the orange ovals represent the revenue or loss outcomes that impact the ultimate objective of maximum profit. The connecting lines represent mathematical equations.

Business users can quickly configure model components to align with their company's portfolio characteristics and business requirements. Typical configuration steps include importing data and scores, and filling in values for variables in action-effect predictions (fee revenues, take-up rates, channel-based costs, operating/maintenance costs for new accounts, etc.).

Business users also specify constraints such as portfolio bad rate, maximum and minimum limits, APR ranges, and so on. This reins in the theoretical ideal of maximum profit to a realistic "best," given the other business considerations that must be taken into account. In essence, it answers the question, "Which origination decision strategy will deliver the highest profit given all our objectives and constraints?"

But this process doesn't just provide an answer, it also quantifies the trade-offs. By adjusting constraints and/or outcome objectives, business users can explore the impact on profit. For example, "How will expected profit be impacted by a change in application risk score cutoff?" "How will it be impacted by an increase in rate?" 

This enables lenders to better understand the profit dynamics and revenue/credit-loss trade-offs for their portfolio. It helps them choose an optimal operating point that is right for their business at the present time. This chosen operating point is output in the form of a decision strategy tree for implementation directly into a business-rules-driven origination process.

Generally the optimized strategy becomes the next "challenger" pitted against the current "champion" strategy in systematic champion-challenger testing. This accelerates the process of improving profitability through champion-challenger methods because it identifies high-performing challengers that would probably otherwise be arrived at only through many more test-and-learn cycles.

High-performing organizations will purposefully design a proportion of challengers to produce “controlled variation.” The goal is to push the design of some challengers outside of the bounds of business as usual, in order to introduce variation into your data and expand what you can learn from it. Think of them not as challengers to replace a champion, but as learning strategies. This cycle, in turn, leads to a new champion and another round of testing.  By exploring a wider range of possibilities, you increase the chances of discovering unique insights that might lead to competitive differentiation.

related posts