In today’s highly technological world, the decision making processes and risk management strategies in loan servicing often rely on computerized systems. They build predictive relationships between the data available for each individual customer and the probability of a satisfactory loan repayment. Credit scoring systems can significantly boost efficiency and performance; however, they are not intended to completely eliminate human involvement by covering the functions of loan officers, managers and credit committees.
In general, financial organizations use credit scoring systems as an additional source of information for decision making or as a basis for developing custom credit policies and loan thresholds, enabling them to automate a number of tasks within the decision making process. However, they still rely heavily on the experience and expert opinions of loan officers and credit committee evaluation for borderline cases.
Avoiding implementation pitfalls
Regardless of the purpose, the implementation process for credit scoring includes a number of basic stages:
- Data collection and mining: Determining required data types (demographic data, personal contact info, job info, financial capability and business efficiency, personal characteristics, etc.)
- Data storage maintenance and audit: Reviewing and benchmarking data in the database, checking information accuracy and relevance, implementing real-time workflows
- Mathematical modeling: Choosing an appropriate method of predictive analysis with regard to the desired outcome and input data peculiarities, “re-teaching” the system, if needed
- Rules and restrictions development: Defining and setting a principal loan amount, interest rate, repayment and cut-off terms to be used for risk management
- Scoring system rollout and maintenance: Evaluating performance of the implemented scorecard within the real-time workflows, monitoring and analyzing outcome, implementing updates
Each of the implementation stages may involve a number of questions that need to be answered. So, let’s have a closer look at the possible pitfalls of the implementation process.
Input data-related aspects
As a scorecard deals with numerous clients over a time period, the number of the attributes it processes is also large. Therefore, there might be a need to select relevant items from the collected data, building a custom data set to be used as an input for creating analytic reports.
To create a robust scorecard system, you need to have a test input of at least 1,500 successful application examples, 1,500 default cases and at least 1,000 applications that were rejected or collected within a 12-month period. In case the scorecard was built with a smaller amount of example input data, the loan decision processes may require more expert input from loan officer’s side to ensure reliable results.
So, in what way does the outcome of the analysis correlate with the characteristics of the input data?
The quality of the decisions made with the help of the scoring system largely depends on the quality of the data analyzed. If the level of data quality is poor, a decision making process will have to rely, to a large extent, on the experience and expert opinion of credit officers to ensure an accurate result.
To reduce the involvement of the human factor, quality of the data will need to be revised and improved.
Assuring quality of the predictive analytics results
To effectively implement the scoring systems into practice, you will always need to estimate the accuracy of the prediction. As the efficiency of a credit scoring solution largely depends on the quality of the input data, it is extremely important to perform a set of statistical and significance tests through training and test data prior to system implementation.
Another important point to consider when speaking about the quality of analytics is the classification method the scoring system is based upon. The majority of credit scoring systems use the logistic regression method as it is considered a standard, for its simplicity, transparency and clear results. However, in case the decision making process almost fully relies on the credit scoring system and human involvement is minimal, a substantially higher level of analytics accuracy can be obtained with the help of hybrid models that combine several classification methods.
Setting credit score threshold limit
To be able to adjust the credit scoring system to specific business requirements, it is important to have a clear picture of how many potentially opportune credit applications you are ready to reject for the sake of avoiding a default case.
While estimating and comparing the amount of profit gained from a good credit application and the cost of the default case, a financial organization can specify and set up the acceptable value of the credit scoring system threshold. The threshold limit set appropriately can create a sustainable competitive advantage by helping the microfinance institution differentiate the ”good” borrowers from the ”bad” ones more effectively, offering the corresponding terms and conditions associated with the risk level and therefore establishing a robust decision making process outperforming market competition. This approach can improve financial sustainability and help maintain an effective risk management process.
Maintaining scoring systems up to date with the ever-changing external environment
Keeping a finger on the pulse of the latest environment changes and emerging financial trends will help to ensure the result provided by the scoring system is reliable and always relevant. An ongoing analysis of macro- and micro-economical, demographic and other external factors can help to discover if any of the system elements requires an update and whether it is an input data or an analytical approach that needs to be changed.
Real value and benefits of scoring systems
- Reducing cost and enhancing productivity of loan officer’s work. As far as credit scoring provides information on the probability of a client failing to pay a debt or meet financial obligations of a loan, credit officers can address such cases in advance, preventing them from turning into arrears.p
- Managing default with implementation of behavioral scoring. Default management is among the most important tasks that can be solved using credit scoring systems. One of the main benefits credit scoring brings to default management is an opportunity to perform retrospective analysis, comparing the impact the credit policy changes had in the past and building up appropriate rules and regularities that can be applicable for similar future cases. This means that a scorecard can help regulate the default rate, even at the stage when the implementation process hasn’t been fully rolled out. One of the key competitive advantages the credit scoring introduces to financial business is that it enables the stakeholders to see the effect that a scorecard will have on the default rate, even before the system has been fully rolled out. After the scorecard is developed and implemented, managers can run a historical test to generate a clear picture of how the changes in credit policies can influence their performance and arrearage rate.
- Giving customers a stronger background for more efficient financial decision making. Credit scoring can help lending institutions to prevent their clients from over-indebtedness, as, when combined with tailored financial education, this practice can help clients make more determined and sensible financial decisions.
- Improving pricing strategies and creating targeted marketing campaigns. With the help of credit scoring, financial organizations can develop smarter pricing strategies considering the level of risk and all the specific loan terms to ensure an efficient customer retention process is in place.
Another example of improvement introduced into the financial practices with the help of a credit scoring system is a reference-based approach to customer loyalty programs. The idea behind this approach is to motivate clients who have shown stable financial behavior within the loan period and are considered to be profitable for an organization. Offer to suggest or refer to a new potential customer in exchange for bonuses like a custom loan interest rate or extended credit limit. This kind of reference is also taken as a basis of creating a score of a newly engaged client.
Viacheslav Basov is Data Science Stream Lead at ELEKS, a global provider of software engineering and quality management with focus on data science, mobility, digital production and finance solutions. With seven years of experience in big data analytics, Viacheslav leads the development of complex business intelligence data-driven solutions at ELEKS. An active member of the data science community, he is passionate about sharing his knowledge through blogging. You can contact him at Viacheslav.Basov@eleks.com.