The first step in optimizing credit processes using analytics was in the 60s when the first scorecard was introduced in a credit card management process. After which, for a long period, credit scoring was used mainly to manage simple actions, for example accept/reject. Now the combination of market competition, new regulatory compliance and the 'never ending' economic crisis impacts are pushing on scoring and credit processes to evolve in order to maximize profit and reduce costs with a 360° view of the customer.
Let’s get a perspective from Davide Capuzzo, Analytics Director of CRIF, on how predictive analytics is transforming credit risk management processes.
Q. What is the role nowadays of scoring in credit processes?
A. The role depends on process complexity. In application processes, scoring can cover from 80 to 100% of the process, while in collection usually no more than 20%.
Q. Why does the performance definition affect scoring?
A. The performance definition is crucial in any scoring and process design. When the definition is purely operational, for example bad & good clients, usually scorecards are focused on optimizing a reject/accept cut-off: the quality of the score below cut-off is not relevant. When the definition is financial, for example defaulted versus not defaulted customers, the scorecards are aiming towards optimizing the entire score distribution as it becomes relevant to discriminate between good and very good counterparties.
Q. In which way is the performance definition affecting scoring technology?
A. Especially when the performance definition has a financial outcome, for example PD with RWA or provisions, maximizing scoring system performance is crucial: 100 basis points of Accuracy Ratio can brings tens of millions of euro of savings on Capital Requirements. The way we maximize scorecard performance depends on the kind of project. IRB and compliance projects are based on consolidated scoring technology, so the only way to increase model performance is to enlarge the information sources used. In this area CRIF has a wide experience, specifically in integrating credit bureau data into the internal rating system with proven advantages for the banks’ capital requirements.
Operational projects like early warning, collection, marketing, fraud detection, etc., leave more room to experiment with new technologies or combine new technologies with new kinds of data sources: it is the case of “not traditional data” combined with machine learning, or vector machine, techniques. In this field, often the problem is not necessarily maximizing performance but to provide an input for driving credit processes even when there is poor, little or no data available. It is the typical case of the “no-hit” credit score: “no-hit score” technology consists of providing scoring systems based on pooled and “non-traditional” data sources which can provide a credit assessment when structured financial services information (e.g., from credit bureaus) is absent.
Davide Capuzzo, Analytics Director CRIF
