Risk-based Collections Part IV: Collections Prioritization

In Part I of my Risk-based Collections blog series, I described the differences between Judgmental and Statistical Scoring models.  In Part II, I discussed which scores are best for managing existing credit lines.  In Part III, I described which scores are best for new application credit risk evaluation.  In Part IV, I will describe how corporations can use statistical scoring for collections prioritization.

Historically, in most credit and collections departments the prioritization of collections has been based on aging.  The customer who owes the most money for the longest period of time receives the highest priority. But using aging alone to prioritize collections activities may be the wrong strategy; especially if the company’s goal is to optimize collection efficiency, improve DSO and reduce write-offs.

During the past decade, credit scoring has become one of the most powerful tools available for automating risk analysis and evaluating the collectability of a company’s accounts receivable portfolio. The model is designed to predict the inherent risk of a customer, including the probability that the customer will become seriously delinquent, go to write-off or file for bankruptcy.

Statistical-based scoring specifically quantifies specific risk probabilities on your accounts. And it is that capability that most separates it from credit bureau generic scores or in-house judgmental-based scoring. The score produced as the product of statistical-based scoring essentially provides a measure of the risk that a given customer will pay their bill on a timely basis. The standard output from a statistical-based scoring system includes not only a credit score but, but also the probability that the account will go bad i.e. Probability of Bad (PBAD) within a specified period from the scoring date, usually six months, and an estimate of the cash value of the account that is at risk, i.e., Cash at Risk (CAR). These values, when properly applied, will aid you in allocating collection resources to specific accounts such that the return on investment (ROI) from collection operations will be maximized.

To understand how statistical modeling drives collections, let’s look at a typical situation that a company can find itself in and see how this additional information can be applied. Collection resources are limited and every overdue account cannot be directly addressed by a collector so a decision needs to be made as to which accounts to call directly and which ones to handle by a less expensive method – say by a dunning notice of some type.

Let’s assume that you only have the resources available to make one call and there are two accounts in question. Account AAA owes $50,000 and has a Probability of Bad (PBAD) of 10%, i.e., has only a 10% chance of going bad within six months and their CAR is $5,000 (PBAD times AR value). Account BBB owes $20,000 and has a PBAD of 50% so their CAR is $10,000. Who do you call?

From a risk-based collections standpoint you call Account BBB. And from both a statistical-based as well as a common sense based position, here’s why. Account AAA is a relatively good risk, a PBAD of 10% indicates a lower risk account and the chances are that the account will pay in due course and may resent a collection call which could upset otherwise good customer relations. Additionally, they represent only about 50% (100 times 5,000/10,000) of the risk that Account BBB represents.

In other words, a call to Account BBB gets you about 200% more bang for your collection dollar than a call to Account AAA. Account BBB has a PBAD of 50% and is a very high risk account. Accounts in this class should be monitored very closely and called as soon as they are one-day late. In this situation, the action implied by risk-based collections is the opposite of what a historical collection decision would be which would be to call Account AAA, the higher value account, because you would not know that Account BBB actually represents significantly more risk.

In summary, several factors need to be considered when deciding whether to use generic scores, credit bureau reports and data or a judgmental-based model enhanced with credit bureau data, versus a statistical-based model.  The critical characteristic that separates statistical models from judgmental models is their ability to quantify risk. This capability, more than any other, is what makes statistical-based models such a powerful tool for the credit and collection function.

By knowing and using the probability of the occurrence of specific credit and collection events, it is possible to optimize the allocation of the resources available in a given credit and collection environment, thereby developing strategies that mitigate the possibility of negative results, while simultaneously increasing the credit lines of low risk accounts and providing the opportunity for additional revenues.

Are you using statistical modeling to help with collections prioritization? I’d like to hear from you.