|
With annual revenue in excess of $600 million, Edward Don & Company is privately-held and the largest distributor of foodservice equipment and supplies in the United States.
Edward Don & Company's 41 person credit and collections team is decentralized and consists of one director of credit, five managers and 35 collectors / analysts working out of 5 offices. The team manages 35,000 active accounts with 18,000 unique legal entities. They receive on average 150-200 new accounts each week and each collector manages approximately 1200 accounts which are primarily high volume, low dollar customers.
Challenges faced with bureau data
Prior to integrating statistical scoring into its credit and collections processes, the company was heavily reliant on buying generic bureau scores to manage new account applications, monitor existing credit lines and help drive collection strategies along with aging reports. Their approach was not dissimilar from most companies; for new accounts, they used bureau data and with an internal judgmental scorecard to help determine credit lines and terms. For existing accounts, they used various score cards that were based on a combination of bureau data, financial reports and some internal data.
Inaccuracy in scores creates risk
However, like most companies deploying this approach, it was found to be less predictive than desired. The challenge was that the information was not as accurate and delivered as timely as the company required. Customers, when scored at the "ship to" location rather than at the legal entity, that were low risk were often placed into a higher risk category and customers that were high risk were not always captured as high risk.
Impeding the sales team
This resulted in orders up for credit review placed on a potential hold when the accounts should have been considered low risk. It also impeded the sales team from identifying sales opportunities with low risk customers. Credit managers were spending too much time reviewing accounts and managing the order release process instead of on the phone collecting cash. Essentially, the inaccurate information made the team more reactive instead of proactive.
 Prior to implementing SunGard's Predictive Metric's statistical scoring, 66.2% of our accounts were flagged as high risk resulting in an excessive amount of orders going on to credit hold. Routinely, 98% of those were manually released. The sales team was frustrated and the customers in jeopardy of shopping elsewhere. Statistical-scoring helped us identify that on average only 14.1% of our accounts were truly high risk, and needed to be on hold. This freed up our credit analysts time to focus on those accounts as well as other projects. Additionally, the sales team had more opportunities for up-sell to the low risk customers.
--John Fahey
Edward Don Company
Director of Credit
Introduction of statistical modeling
While it was innovative for business-to-business credit that Edward Don & Company's scoring model considered internal data, the company was not using a statistical model which takes into account payment history, seasonality, economic conditions, order trends, etc.
Validating statistical models
After reviewing what was available in the marketplace, Edward Don & Company chose to integrate statistical-modeling into its credit and collections processes. The company engaged SunGard to conduct a complimentary validation analysis using AvantGard Predictive Metrics, a suite of statistical-based scoring models which leverage a company's internal A/R data combined with terabytes of behavioral trending information to determine the probability-based risk grade. The models help predict the specific probability that an existing "GOOD" paying customer will become a delinquent or a "BAD" paying customer at some point during the six months. The models also calculate how much cash is at risk in the portfolio.
How the Validation Works
Edward Don & Company provided SunGard with files containing 18 months of historical accounts receivable data on its entire portfolio through a secure, encrypted FTP site. This included the original account data file for collections (A/R and billing data), as well as the payment performance information including the types of payments during the validation period.
SunGard statisticians scored the customer file and then attached the performance data (negative payments / transactions). From this retrospective analysis, a summary report was produced showing the model's predictions versus the actual results. The outcome of the validation process proved the models predictive power by producing a report outlining the accounts with the highest propensity to not pay (including expected dollar value). The forecasted results produced by the model were shown to be very predictive in capturing the actual severely delinquent accounts from Edward Don & Company.
Calibrating the models for each portfolio
Edward Don & Company worked closely with SunGard to deploy the Predictive Metrics scoring models. Similar to the validation process, the company provided the SunGard team with monthly historical data on its accounts necessary to produce fresh scores going forward. SunGard turned the scores around within one business day. Edward Don & Company now had fresh data instead of stale data and they used the scores to help manage credit lines and the order hold / release process. Furthermore, they no longer had any hit rate issues with the credit bureaus in this scoring process because bureau data is not required. The company also used the scores to help prioritize collections and focus the collectors on calling upon high risk accounts.
Managing existing credit lines
Prior to implementing statistical scoring, Edward Don & Company's internally developed model showed that 66.2% of accounts were considered high risk which resulted in excessive order holds that the credit analysts had to manually review. In the end, 98% of those orders were released from credit hold. This dynamic not only created extra work for the credit department, but it also fostered resentment within the customer base and sales organization.
Edward Don & Company worked with the SunGard team to calibrate its statistical-scoring models which calculated the probability of accounts that would go bad over the next six months. Adding credit bureau information to the model based on bureau historical payment terms was tested and determined that the lift provided by bureau data was not significant enough to justify the investment in the monthly expense of bureau scores.
Figure A (below) shows the risk distribution after implementing the statistical scoring model. Only 14.1% of the total portfolio is considered high risk compared to the original figure of 66.2%. Edward Don & Company is able to more accurately predict which customers are likely to go delinquent or not, set the appropriate credit limits, or hold orders only on actual high risk customers.
By using the scoring models, most orders do not need to be reviewed by an analyst and are automatically approved based on the risk classification. The credit analysts spend significantly less time manually reviewing accounts and have reduced the number of order holds which ultimately helps the company identify opportunities to increase sales within those accounts.
Statistical modeling also helps Edward Don & Company identify high risk customers early and reduce credit lines to minimize losses and also identify low risk customers early to help increase credit lines for increased revenues.
Figure A
Driving collections prioritization
Prior to incorporating statistical-modeling into its collection processes, Edward Don & Company was contacting aged, high dollar accounts first. This old practice is widely considered inefficient and ineffective. Once they introduced the Predictive Metrics statistical-scoring models into the collections process, they were able to truly prioritize by designating high and low risk accounts. Once determined, the collections team could then identify the appropriate strategy for each one.
Edward Don & Company sends SunGard a file every month and SunGard immediately scores each of the accounts and provides them with two reports. The first report shows the expected probability of accounts going bad and assigns 6 risk grades from Extreme Risk to Low Risk. Accounts are assigned to one of these risk grades. The second report shows the same results as the first report but adds in the dollars at risk. Edward Don & Company can drill down and see which accounts fall into these grades and the cash at risk and utilize the information for not only credit line management but collections prioritization.
SunGard's Predictive Metrics has helped us drive improved collection prioritization and ultimately lower DSO by 5.3 days. Instead of focusing on terms and past due balances, the statistical-scoring models help us focus on risk and tell us which accounts have a probability of going delinquent and the dollars at risk. We are able to be more proactive with those accounts.
--John Fahey
Edward Don Company
Director of Credit
Results
The Predictive Metrics models havehelped Edward Don & Company improve its ability to effectively manage credit lines and the order hold-release processes. The company has also been able to drive improved collections results using superior prioritization techniques with risk instead of aging.
Operational efficiencies have also improved as they have been able to free up the credit and collections team to focus on contacting more of the risky customers.
The team has realized a 25% increase in phone calls and a reduction in DSO by 5.3 days in one year. Edward Don & Company has also been able to decrease headcount during the company's growth by deploying this statistical modeling service. The company works closely with SunGard to calibrate and revalidate the models every year to help ensure that the models sustain a high level of performance and reflect the latest market conditions.
|
