The Covid-19 pandemic amplified the benefits of digital financial services by reducing the need for physical contact in financial transactions and helping lenders reach clients virtually using credit scoring. Although credit scoring is widely used in some countries, in Tunisia it has been limited largely to a Central Bank reporting system on borrower defaults. With the USAID-funded Tunisia Jobs, Opportunities, & Business Success (JOBS) project, Connexus has supported the development of digital credit scoring tools to make data-driven decisions on loan approval.
Connexus’ objective in developing credit scoring systems and tools was to improve outreach to clients without extensive credit histories, particularly women- and youth-owned small and medium enterprises (SMEs). With new scoring systems, financial institutions can use technology to improve risk management, while reaching new clients and decentralizing decision-making. For borrowers, these digitalized systems reduce the need for multiple trips to the bank to apply for a loan.
Two JOBS partner banks have issued 610 loans valued at $107.8 million using their new credit scoring systems. Connexus’ manager of JOBS’ Access to Finance Team describes the impact that credit scoring has had on partner portfolios: “Our partners have been able to dramatically reduce non-performing loans to just over 1% for SMEs, which is the lowest rate in their portfolios.”
Development of the credit scoring tools began with understanding existing data management systems and mining four years of portfolio data to create risk-based statistical models. From more than 5,000 loan applications, data management experts selected fewer than 2,000 loans as the initial data pool. These loans were assessed for data completeness, quality, and accuracy. A final sample of 200 loans was selected and data on customers, loans, and risk profiles were extracted and cleaned.
Using these data, Connexus’ experts developed statistical models to cross-reference financial, non-financial, and psychometric information, and created benchmarks and scoring matrixes. The data categories for these models included loan and client financial performance, account conduct, business structure, management experience, industry risk, and psychometric assessments of business owners. Within these broad categories, experts developed 39 business performance and psychometric indicators.
The statistical models assessed risk across a range of indicators, cross-referencing them with loan non-performance. Loans were classified by performance, with “bad” loans being more than 90 consecutive days past due. Because these loan data represent only a sample of the portfolio, Connexus used its expertise and that of partners’ risk and credit teams to develop a holistic approach to analyzing risk using the statistical model.
Experts weighted indicators by their impact on the risk of non-payment. The initial weightings were financial performance (22%), account conduct (8%), business structure (20%), management experience (15%), industry risk (15%), and psychometric score (20%). These weightings were intended to evolve over time, increasing the relative importance of psychometric indicators as bank staff grow familiar with the scoring systems.
The sum of these weighted scores was used to determine a loan applicant’s credit score. Statistical data experts recommended a scale for determining loan approval, with cutoff values for loan rejection and acceptance. For example, loans with scores of 170 or lower were automatically rejected, and scores with 225 or higher were automatically accepted. Scores that fell in between were tagged for additional manual analysis.
To ensure that the model reflected partner realities, Connexus experts regularly shared and validated their findings and recommendations with bank management and staff. In addition, they integrated the scoring tools into the banks’ management information systems and built capacity to use the tools. Pending full integration, the consultants developed stand-alone spreadsheets to calculate credit scores. During the pilot period of 12-18 months, banks are monitoring scoring, data entry, and analysis. After the pilot, partner banks will review and adjust their tools based on experience. Once the scoring tools are integrated into the core banking systems, all departments will have access to them.
Although full integration of credit scoring is still a work in progress, Connexus has identified several important lessons. Building awareness and engagement of top management, credit committees, and board directors is vital, as are frequent consultations to validate data and share conclusions with stakeholders. Finally, extracting full value from innovations like credit scoring takes time and requires continuous engagement. For example, the integration process requires monitoring and follow-up to ensure that lessons learned are fed back into the system. It also requires ensuring that bank staff have the capacity and incentive to use the new systems.
