Scoring has proved very beneficial in credit granting organisations world wide and in many cases has replaced the conventional judgmental decision process. Judgmental in this case refers to the manual process conducted by underwriters to assess the risk of new applications. The main criticisms of this process are the:
•     Consistency of the decision: where different underwriters are unable to make the same decision
•     Ability for management control: It is difficult for management to increase the number of approved applications
•     Lack of automation: Each application is required to be processed by an underwriter

Scorecard based decisions is an automated process whereby certain information from the new applications form is used to produce a score. The various scores rank order according to risk. For instance an applicant that scores 200 would be more risky than an applicant that scores 260. Management now has the ability to set the level or score at which applications are accepted or declined. This is typically called a cut off score. Underwriters still play an important part in the decision process as there are some applicants (typically around the cut off score) that would require more investigation before a final decision is made.

There are two types of scorecard builds, custom and expert.

A custom scorecard is built using past applicants from the organizations portfolio where each applicant’s subsequent performance is evaluated. In the case where past data is not available, due the portfolio being new or where past information was not collected an expert scorecard could still be built. Expert scorecards are also called generic scorecard.

Expert scorecards are not built on the organisations portfolio but rather using the scorecard developer’s knowledge of the industry the organisation operates in. Once completed, a scorecard will be delivered that will use information from the application form to produce a score. The main difference between the expert and custom scorecard developments is data.

Expert scorecards are typically seen as a starting point in the scoring process. Due to this fact the organisation team responsible for the signoff and implementation of the scorecard may not know what to expect or how to get the scorecard up and running. With all developments the final deliver would consist of the scorecard as well as implementation guidelines, but this must be seen as guidelines only, because every organisation is unique. This article is intended to cover many of the issues around the impact the development and implementation of expert scorecards will have.

Organisation Policy Impact
With a new scorecard development, the developer spends time gathering information about the organisation. This ensures that the final scorecard is geared towards the specific requirements of the company. It is important that the key decision makers as well as members from the ICT, risk and marketing departments are involved from the beginning in the development. In many cases the results from the development leads to changes in both the policies and the process flow of the application process. This will ensure that the
environment in which the scorecard operates is optimised.

It is not only necessary for the operations department to give insight on their current scoring system, but also to understand the implications and possible changes to the system.

The applications process flow could change to cater for additional policy rules upfront. These are usually accounts that will be excluded from the scoring process. Examples of these can be applications below the required age, staff accounts VIP’s, minimum income or number of loans.

The process of determining whether an application is scored may need to retrieve information from other systems. The policy may be to allow only one account or loan, and it would therefore be advisable to have a link from the application system to the account system. Applications with existing accounts or loans can then be declined upfront without being scored.

Department Impact
The changes to the process flow may also have an impact on the new accounts department where the turn around time is shortened, for instance scores below the cut-off can be declined automatically. The higher score ranges can be approved automatically and only the middle score ranges can be given to the underwriters for additional action.

The marketing department may need to update the marketing strategies and will need to know how the scorecard will affect the decision made on the through-the-door population. The finance department has to budget for the build and implementation of the scorecard.

The risk department will be required to educate the rest of the organization on scoring. Monitoring and validation of the scorecard will also need to be arranged.

Application Impact
The expert application scorecard consists of a number of known predictive characteristics selected by the developer, based on the specific portfolio it is being developed for. To implement the scorecard successfully, the organization will need to ensure that the process flow includes the capturing of all scorecard characteristics (questions asked on the application form) from the application form.

The organisation will need to review the application form and decide which fields should be mandatory and which will allow ‘missing information’. It is recommended that the fields required for the application of scoring and credit policies are made compulsory with additional fields selected for future marketing, scoring and tracking activities. In addition to gathering data for future benefit, the additional fields ensure that individuals cannot be able to tamper with the decision system to work out how the scorecard assigns
points.

Once the application has been captured on the system, it has to go through a process that will look at certain company policies to decide whether to decline or to continue to the next steps in the application process flow. Certain applicants will not form part of the population the model is designed to work on. When implementing a model, the organisation must decide when a customer will be scored and when a manual or automatic credit policy needs to be applied. Examples of this are under age, minimum income, Critical Credit Bureau information, fraud and capacity to repay.

The application will then go through the scoring step where the model will calculate a score for the application. The scoring system should also cater for the bureau integration if available. At this stage the application can either be declined or it can continue to the next step. The next step can include employment verification policies that will be used to make the final decision.

Scorecard Implementation Impact

Acceptance Rate
When the expert scorecard is implemented on a new portfolio there are no statistics available to determine what the approval cut-off should be. This is because no data was used to build the scorecard. Typically it is suggested that the underwriters manually assess the risk of every applicant (which is also scored) until such time as there is enough information (a minimum of 300 applications, including rejects) available, to produce the statistics required for a cut-off score to be determined. Examining the ratio of the accepted to the declined applicants will determine the approval rate.

A conservative approach is advised when setting the initial acceptance rate. A more aggressive decision can be made once performance data is available to assist in making the decision.

Where the implementation is for an existing customer base, it is recommended that a number of historical applications are scored before deciding on the cut-off score. The period and number selected, depends on the data available. Again, a minimum of 300 applications is considered acceptable. This will provide an estimated score distribution where the common practice is to apply a conservative approach to select the same acceptance rate level as the previous decision process.

Data Storage
It is important that the application data is stored in a centralised warehouse, where the application data is stored as at the time of application and never updated with new information. This will ensure that there is future information available to build custom application scorecards.

Linked to the development of custom scorecards and tracking of current scorecards is the availability of the performance related data.. Usually, the monthly delinquency status is required for each account. Written off and inactive accounts should also be stored.

Overrides
There might be specific cases where the scorecard decision is manually reversed. These are usually referred to as overrides. There are three different types of overrides, namely informational, policy and intuitional overrides. Informational and policy overrides are normally the applicants that have been declined even though they scored above the cut-off score. Intuitional overrides are applicants that were accepted with a score below the set cut-off level. It is advisable to track the overrides by reason as this could be incorporated into the policy decisions to minimise manual intervention.

Setting the initial offering
The scorecard is also used when defining credit limit, loan amounts and terms. Once the risk at which the customer is approved is defined, the terms and conditions of the deal can be set depending on that risk. This is usually carried out by creating a matrix of the score and the income (depending on availability, this may be household income, net income, gross income, etc.) to determine affordability of the applicant. Applicants with higher scores (less risky) and high affordability would get the best product terms.

Tracking
The performance of the scorecard should be monitored as soon as it is implemented to keep management informed on the progress being made. The reports will provide management with information that may be used to modify the application processing in order to adapt to the changing environment and improve the overall data collection process. The performance reports will however only show true performance after a period of about six months.

The portfolio as well as the effectiveness of the scorecard is influenced by internal and external factors. These range from economic and marketing changes to modifications in the collections and write off policies. These changes have to be detected promptly in order to react in time.

There are various tracking reports that are recommended. Reports that are used to analyse how the population is changing and whether the scorecard decision is being applied are the population stability and characteristic analysis reports. The final decision report evaluates the decisions made based on the score. It measures the potential approval rate if no overrides were present.

The override tracking report shows the applications overridden by reason for applications over a certain period of time. Problems within overriding or possible trends can quickly be identified.

The delinquency distribution report brings together an understanding of the applicant’s information at the time of application with their subsequent performance. The vintage analysis report is constructed using the delinquency figures from the quarterly delinquency distribution report. This report identifies delinquency trends that may be caused by ‘life cycle’, ‘new account’ and ‘portfolio’ effects.

Lastly, the portfolio chronology log is an on-going record of the internal and external events that my effect the quality of the scorecard. This log is useful when trying to identify the reasons why a certain shift is observed in the tracking reports.

Summary
In summary, when implementing an expert application scorecard the organisation should be prepared to go through policy and system changes. Various departments will need to come together and agree on decisions around the policies and the scoring system.

It is also important to have the correct application process flow, as it can be costly if not thought through properly. The introduction of scoring as well as the changes to system and processes will require training to ensure that the staff accept the new decision process and adhere to the system decisions.

Having no statistics available for the setting of the initial cut-off will require sufficient data to be accumulated. This will enable the organisation to track the performance and quality of the scorecard as well as to determine the scorecard cut off. It is therefore very important to store all the application and subsequent performance data. Ensuring good quality data will not only assist in the tracking of the expert scorecard, but also ensure that data is available to build future empirical scorecards.

Having said all this, it is worth the initial teething stage to implement an expert scorecard. This is a great starting point to eventually develop custom scorecards that will make the organisation millions in the future.

About the Contributor
PIC Solutions is the leading specialist credit risk solutions company in the EMEA region. With offices in Cape Town, Dubai, Johannesburg, Manama and Nairobi, we deliver integrated analytics, consulting and software solutions to over 150 companies in 30+ countries. We work worldwide with organizations to improve performance, drive strategies and enhance profitability. Analytics | Consulting | Software