ClearScape Analytics™ accelerates model development for credit risk portfolio

Increasing a customer base and personalizing credit offers demand managing credit risk profiles using analytics

Sicredi is the first and largest financial cooperatives institution in Brazil. This institution consists of credit unions across the country offering credit, credit cards, checking and savings accounts, certificates of deposit (CDs), investments, payment services, social security, and more. Its market presence includes over 7M clients, managing over R$311B ($64B USD) in assets and R$192B ($39.5B USD) in credit. By valuing relationships and offering its financial solutions to improve the quality of life of its members and society, Sicredi is focused on sustainable growth.

Like other global economies, Brazil is facing economic pressures, such as inflation, slow gross domestic product increases, wage stagnation, and unemployment. These pressures create challenges for financial institutions tasked with predicting and managing delinquency rates of consumer credit offerings.

“Our gross domestic product (GDP) is increasing, but it’s increasing at a slow pace. The result is an increasing delinquency rate in the financial market. To control delinquency, especially in credit, is really hard,” shares Matheus Daniel Pierozan, senior credit risk analyst at Sicredi.

For Sicredi to maintain sustainable growth, it must continue to grow its customer base and increase personalized credit offers while managing payment delinquency risk. Data from multiple sources must be integrated and harmonized to gain a greater understanding of customer profiles and evaluate a customer’s credit risk.

“Our customer base is increasing around 13% a year and our credit portfolio is increasing around 33% a year in the last five years. This is a big increase, and we need to control the delinquency,” Pierozan elaborates. Controlling delinquency affects profit and growth.

Sicredi’s rapidly growing customer base means nearly 40% of its customers are new since 2019. Delinquency rates are at risk of increasing when credit risk models are inaccurate or when data scientists can’t model on all available data.

“Because we have new customers, they bring together new types of behaviors. One of the main problems of credit risk modeling is that you need to update your models often. You need a history period to train models and capture those behaviors,” explains Pierozan. 

Many of the biggest challenges in credit risk management involve analytics and data. Credit risk analysts often struggle with accessing the right data. Information on new customers can be limited to existing credit scores, income, net worth, outstanding loans, and other liabilities, while existing customers have prior institutional history. To create appropriate personalized credit offers that minimize delinquency risk for new customers, credit risk analysts need large amounts of data. Unfortunately, data quality and consistency can pose a challenge. Moreover, adapting to market changes means models must continually be updated to reflect outside credit risk factors.

Teradata VantageCloud on Amazon Web Services (AWS) and ClearScape Analytics accelerate time to value

Until recently, Sicredi’s model development was time consuming and costly. Roughly 80% of time was spent developing models. Data scientists and analysts relied on several disparate databases, as data was siloed across more than 100 credit unions. Sicredi lacked data integration and harmonization. Before implementing VantageCloud and ClearScape Analytics, Sicredi required two years to develop six models. 

“We need to develop models and deploy them into production quickly. VantageCloud on AWS helps us to better manage data effectively and organize the data pipeline, which is really important in model processing,” explains Pierozan. 

Sicredi ingests data into VantageCloud on AWS and uses ClearScape Analytics to perform feature engineering, select relevant features and attributes, and run credit risk models. VantageCloud is the complete cloud analytics and data platform for AI, and it integrates and harmonizes data across disparate data sources from Sicredi’s 100+ credit unions. This empowers data scientists and analysts with universal access to customer and registration data and confidence in the data’s quality.

“Our data pipeline starts with extract, transform, and load (ETL). We collect the relevant information from data sources, perform data quality, and then select the features to run our models. Since we’ve rebuilt this pipeline, data is organized and harmonized and our model processing is accelerated,” Pierozan shares.

By building a data pipeline in VantageCloud and using ClearScape Analytics, Sicredi has reduced development time to build, test, and deploy a model by 40%. Sicredi is now able to develop more models more quickly, resulting in over 17 new models in less than one year. 

Analysts and data scientists are now able to spend more time on modeling techniques and approaches that will improve credit risk analysis and reduce delinquency rates. Sicredi is shifting its client modeling approach to become more personalized and relevant based on the product and individual.

“In the past, we had models per client. One client had one credit risk and one probability of default. Now, we have models per portfolio. One client may have two, three, or four different credit risk measures, and that’s really important to be able to offer the best financial solution,” explains Pierozan.

By moving to models per portfolio, Sicredi better estimates client credit risk on a per-credit offer basis. 

“The same client may have different credit risks according to the loan. If you have a mortgage loan, the client may have a probability of default of 2%. But in revolving credit, such as a credit card, they may have a 10% probability of default. These two products are very different, and the behavior of the clients, in the moment that they are going to pay the bills or loan, is different. This is what we gain by converting models per client to models per portfolio,” describes Pierozan.