Fal de Saint Phalle, VP, Manager, Decision Analytics – Region Americas, De Lage Landen

Over the years there have been many discussions about how to increase the rate of effective automated credit decisioning for commercial small-ticket processing. Many financial companies were making progress until 2008, when the credit crisis intervened that resulted in more applications being decisioned by analysts (users) due to changes in cutoffs and rules. In more drastic situations, lenders shut off their models. However, the business imperatives to improve efficiency and reduce costs and, therefore, to move aggressively to a more automated credit underwriting process have not abated, but in fact have grown.

This article will outline one concept to accomplish 100% automated credit underwriting of small-ticket lease credit applications. While this article does not deal with the implementation of this concept, it would make sense that it be applied initially to micro-ticket lease applications to prove the concept, transactions for which there is clearly a business case for not having these applications underwritten by a credit analyst.

This approach is called “Project No Touch,” due to the fact that once the applications are in the system and submitted for credit scoring and decisioning, there is no manual intervention possible. All decisions are rendered instantaneously.

Advantages
In addition to the obvious reductions in cost and improved customer service through 100% instant decisioning, there are several advantages with this approach:

• It incorporates in a very transparent manner the key considerations used by credit analysts (users) in making manual decisions; but, under this process, those decisions would be made much more objectively by the system.
• It resolves the concern expressed by customers about the scorecard solely making the automated credit decisions, as the analyst input is fully integrated into the process.
• It facilitates targeted approval rates for all strategic channels, partners or other segments, which should be a key motivation for those responsible for managing this commercial business.
• It guarantees faster turnaround time on decisions, which should lead to greater customer satisfaction, higher conversion rates and, therefore, more bookings.
• It allows for implementing fully a financial company’s current credit policy.

Basic Approach and Key Assumptions
The basic approach is to build a credit process flow that uses three sequential scorecards to process each application. These three scorecards would allow full transparency to credit management as to what is happening to each application at each step in the process. These three scorecards are:

1. The probability of default (PD) model, as is used in current scoring and decisioning flows used by most leasing companies,
2. A review rule model, used to evaluate current review rules in such a manner as to avoid using this information to send applications to analysts, and
3. A cut off model, used to determine which set of score cutoffs to use for decisioning, based on that particular PD model score.

The following are key assumptions under this concept:

• Analyst considerations that are used to make manual (including override) decisions need to be incorporated in the 100% automated process.
• The optimal scoring environments would require multiple sources for credit reference agency data, such as commercial and consumer (e.g., sole proprietor) data, as all applications that do not have scorable data will have to be automatically declined.
• Data used to currently run logic for analyst referrals will be fully incorporated into the automated system to completely avoid manual intervention.

Concept Flow
The concept has five main components: PD models/scorecards, decision rules, a review rule scorecard, a cut-off scorecard and a set of five alternative cutoffs per model, all of which are more fully explained below:

1. The PD models/scorecards would be the same as those used currently, and therefore this new process would require no changes to those models.
2. Decision rules would be invoked for those applications that should be automatically declined (such as bankruptcy, prior loss, etc.) or approved. Those decision rules would be normally defined by credit policy.
3. A review rule hybrid scorecard would be initially developed statistically, assigning points for the severity of each rule that any business user would wish to invoke, with the most severe rules being assigned higher points. Severity refers to the degree to which the rule either correlates with default or is determined by expert judgment to be “bad.” Each application would be classified into three buckets based on the sum total of weighted points: low, medium and high. This output would be one of the variables in the cut-off scorecard, which is described next.
4. An expert scorecard customizable per asset, region or product type would be developed and used to determine which set of cut-offs (among five alternatives) to use in order to process each application. This scorecard would take into account those factors commonly used by the credit analysts to make manual and override decisions, with the following elements serving as examples for this article:
   1. Type of customer
   2. Type of vendor
   3. Industry/region risk level
   4. Overall level of vendor profitability
   5. The review rule score
   6. Lease type

   The highest scoring applications would be routed to the most accommodative cut-offs, with a high expected approval rate, and the lowest scoring applications routed to the most restrictive cut-offs with a low expected approval rate. This would be consistent with the preference expounded by most companies, to approve lower risk applications to existing customers from strategic vendors/segments, etc. An example of such a model is shown in Figure 1.
   

   

5. A set of five alternative cut-offs (per PD model) would be provided, ranging from very accommodative to very restrictive. Each would have only one score delineating between accept and decline. An example of the net result of processing these applications is provided in Figure 2.
   

   

Conclusion
As a result of this process flow, an automated decision of accept or decline is immediately rendered, either by decision rules or cut-offs, for every application submitted. There are no referrals to analysts. This process would also allow credit managers to target both overall approval rates and corresponding risk cost levels. The improvement relates to their ability to target higher approval rates at the same risk cost, or keep approval rates constant and lower risk cost, or a blend of those two approaches. It is recommended that this process be introduced to micro-ticket applications to prove the concept, and fine tune the process, while later increasing the average ticket size of the applications as experience is gained.

Fal de Saint Phalle is VP and manager of Decision Analytics – Region Americas for De Lage Landen, located in Wayne, PA. He joined the company (formerly Tokai Financial Services) in 1996, and since that time has worked in the credit administration area with particular focus on supporting members and vendors with respect to automated credit decisioning, scorecard management, analysis and reporting, as well as predictive model development, validation and monitoring. Prior to this assignment, he spent 24 years at Fidelity and First Fidelity Banks in a variety of commercial credit and lending positions. He holds a Bachelor’s degree from Bowdoin College in (Economics) and a Master’s degree in Business Administration from Columbia University (Finance).

de Saint Phalle is retiring this year after 16 and ½ years with DLL, but is expecting to continue supporting the company on various projects in 2012 and beyond. He intends to remain active in this field. From September 1, he can be reached at [email protected].