Burger

CRIF Uses Synthetic Data for Credit Scoring Models, Facilitating Data Sharing

Anonos’ synthetic data produces private and accurate credit scoring models that can be securely shared internally and externally

DISCUSS YOUR USE CASE
CRIF
€10,000
saved on average per proof of concept (POC)
0 risk
when sharing datasets internally and externally
95%
compatibility between credit scoring models developed on real and synthetic data
Case Study Summary
Credit scoring helps lenders and banks assess an individual's likelihood to repay credit by evaluating the risk of default. Due to the data involved in determining creditworthiness, these models necessitate an adequate level of data protection.

It's also crucial for the data in these models to maintain its utility and integrity.
When data is not accurate enough, credit scoring models won’t perform adequately to determine risk.

By adopting Anonos’ synthetic data, CRIF, a global company specializing in credit & business information systems, successfully developed credit scoring models that retained the utility of real data, while ensuring privacy.
"Using Anonos' synthetic data has been a transformative step in our internal model development. It acted as a catalyst in enhancing data privacy and utility, helped us expand our use cases and enabled us to proactively prepare for regulatory changes, delivering time and cost savings.”

Enrico Bagli
Data Science and Innovation Manager
CRIF
Goals
  • Improve model-to-model development for credit scoring
  • Safely share data with external and internal providers
  • Ensure a proactive approach to possible future regulatory changes
Challenges
Balancing Quality & Privacy Using private yet granular data to improve model-to-model development for credit scoring.
Data Sharing Achieving an adequate level of data protection to share data with external providers or internally.
Time to data Accelerate data access within CRIF’s subsidiaries and ease internal data sharing among branches.
Data Retention Ensure a proactive approach to possible future regulatory changes to also find a solution for the data retention requirements.
Solution
Data Quality & Privacy
Using Anonos, CRIF developed a credit scoring model based on synthetic data. CRIF worked with their legal and privacy teams to test the synthetic model for both utility in assessing creditworthiness and privacy. The model achieved a GINI coefficient compatible at 95% with the one developed on real data (a metric for credit score model performance).

While other technologies like masking compromised data usability, Anonos' synthetic data ensured both privacy and utility.
Time to data
CRIF faced challenges in timely data access for model development, partly due to regulatory differences related to the high number of countries the Group operates in. Therefore, particular emphasis was placed on the issue of internal data sharing. Since synthetic data contains no Personal Identifiable Information (PII), CRIF could efficiently share datasets internally without compromising customer information.

Additionally, this approach expedited the Proof of Concept (POC) process - a phase where CRIF tests external providers' platforms to integrate with its services. The legal and privacy teams could rapidly assess the synthetic data's anonymity, streamlining approvals and shortening project timelines, resulting in considerable time and cost savings.
Data Sharing
CRIF utilized Anonos' synthetic data to create high-utility datasets for testing with external providers. Following privacy evaluations, CRIF concluded that both the synthetic model and datasets were apt for external sharing.
Data Retention
By converting credit scoring model data into synthetic datasets, which were proven to be anonymous and free of PII, CRIF has adopted a proactive approach to possible future regulatory changes.
Results
Proactive compliance With the use of Anonos' synthetic data, CRIF certified that the data is properly anonymized, and adopted a proactive approach to possible future regulatory changes.
Boosting team efficiency Synthetic data allowed CRIF data scientists to work with relevant data models with fewer barriers, while preserving customer privacy.
Accelerated time to POC The use of synthetic data for sharing with external providers preserved customer privacy, and allowed time to POC to be reduced significantly.
Growing use case portfolio As a result of using Anonos' synthetic data, CRIF decided to attempt to scale the use cases for data sharing, such as for use in insurance, banking, credit scoring, and other data sets.
Expanding services CRIF saw an opportunity in their success with using synthetic data, in that they could explain it and offer it to clients, adding value to those relationships.