25x faster AI/ML model processing with ClearScape Analytics™
- $650K Potential annual cost savings
- 29% Increase in limits consumption
- 32% Increase in clients with income prediction
Sicredi is in the top 10 financial institutions in Brazil. It consists of a group of credit unions offering loans, credit cards, checking, savings, investments, payment services, social security, foreign exchanges, and consortium plans.
Sicredi’s mission is in valuing relationships and offering financial solutions that add income and improve quality of life for its members and society. In the last five years, Sicredi has seen rapid growth in its customer base, growing 15% annually and approaching 8.4 million customers as of August 2024.
“Our goal is to be close to our members and provide the best financial solution possible,” says Matheus Pierozan, senior credit risk analyst at Sicredi. “To do that, we need to understand the client’s behavior. This is why we’re using models to understand behaviors, predict credit risk and delinquencies, and provide the best product that meets their needs.”
An opportunity to manage preapproved credit offers and reduce operating costs
Offering preapproved credit requires knowing the approximate income of the potential borrower. A credit limit or interest rate that’s too high in relation to a client’s income may lead to delinquencies and default. If it’s too low, the client’s needs go unmet and Sicredi misses out on revenue generation. However, for new prospective customers, offering preapproved credit becomes much more difficult. Banks often don’t have the income details of new prospective customers. This is especially true in Brazil, the seventh most populated nation in the world. Without income information, it’s more difficult to accurately identify appropriate credit limits. For Sicredi, this challenge is compounded by over 2,000 credit products with varying limits and rates.
“The income information is important when you’re offering preapproved credit or a credit card limit, as you base your credit policies on the income information,” explains Pierozan. “When you open your account, you probably will provide proof of income. If you don't have this information in your registry databases, the bank needs to purchase this information. This represents a cost at the end of the month.”
Income and credit history details are often available through specialized bureaus, such as Experian. However, as Pierozan highlights, this represents an operating cost. And, in some cases, information on prospective clients might not be available.
AI model processing in Teradata predicts income classification 25x faster than Databricks
“To reduce the dependency of the specialized bureaus, we’re using income classification modeling, which also helps us save money," says Pierozan.
Sicredi stands to save hundreds of thousands of dollars in bureau data purchasing costs by using artificial intelligence and machine learning (AI/ML) models to predict income classifications.
Trust is of utmost importance when dealing with customer financial information, credit products, and customer financial decisions. Credit and lending institutions typically have more expertise than consumers in assessing borrowing capacity and repayment ability and managing credit limits and interest rates. Therefore, banks must act ethically and transparently. Trusted data becomes the foundation for any organization to build accurate, responsible, and customer-focused models.
Teradata VantageCloud on Amazon Web Services (AWS) provides Sicredi with an AI platform built to run its most critical enterprise workloads—unifying, harmonizing, and securing Sicredi’s data. The platform is Sicredi’s foundation of reliable, accurate, and governed data that fosters better decision-making.
“We’re using the power of data to offer the best financial solution to our clients,” says Pierozan. “Trusted AI at Sicredi means providing trustworthy information to create the most precise model that is useful for the business area and most valuable to the client.”
However, a broad ecosystem of tools and technology has led some teams to explore alternative tools that don’t leverage data management and model development best practices. In its original model processing, Sicredi previously used Databricks to perform income classification modeling. This required significant data movement and data wrangling—which is both time consuming and expensive.
“In the old version of our model with Databricks, we had a problem extracting data from VantageCloud and bringing the data to our data lake,” says Pierozan. “This created a problem in terms of processing speeds and data wrangling in our data lake, which isn’t productive. The process was taking more than two hours.”
Faster data processing creates cost savings
By leveraging Teradata’s ClearScape Analytics for its Bring Your Own Model (BYOM) approach, Sicredi analysts and data scientists continue to use the modeling tools of their choice (in Sicredi’s case, Python machine learning models). Sicredi leverages modeling best practices that bring the analytics to the data, minimizing data movement and ensuring trust and reliability.
“Now, we’re running everything inside VantageCloud,” says Pierozan. “The entire analytics process is within VantageCloud, with no data movement. We prepare the data, create the variables, and train the model. We use CatBoost, similar to an XGBoost machine learning model. After training the model, we export it into ONNX format to deploy the model and run at scale directly inside VantageCloud using BYOM."
According to Pierozan, data scientists should consider doing the same and run analytics in database.
“Processing everything in VantageCloud is safe,” says Pierozan.
VantageCloud’s multi-parallel processing and BYOM capabilities have substantially reduced the model processing time previously experienced with Databricks.
“By processing everything with VantageCloud, we’re taking just six minutes—25 times faster data processing compared to the old approach using Databricks,” says Pierozan.
Processing improvements allow Sicredi to increase competitive offerings with more personalized and targeted financial solutions through faster time to market of offerings. Now, Sicredi is increasing clients with internal income scores, which, in turn, is increasing limits consumption.
“By rethinking all our model processing, we’re able to improve our model,” says Pierozan. “Within the first month, we saw an increase in limits consumption by 29% and 2.3 million clients that are being scored in our database. We’ve also saved around 83% in data transferring. Finally, we’re estimating potential savings of around $650,000 per year.”
연결하자
테라데이타 VantageCloud가 비즈니스 성과를 가속화하고 필요한 비즈니스 민첩성을 제공하는 데 어떻게 도움이 되는지 알아보십시오.
영업 담당자가 도와드리겠습니다.