The client, a fintech company offering banking services, including but not limited to virtual cards, debit cards, currency exchange. Their Credit Risk team spent most of the time working on extract, transform and load (ETL) tasks due to a high number of data sources. The team needed to precisely calculate credit risks with accuracy and speed. Datrics provides an easy-to-use infrastructure, collaboration mode within and across teams, and seamless integration of models to production.
A high number of data sources and occasional changes in data structures/schemas led to excessive time spent working on data preparation and pipeline versioning/support. Collaboration within the team and with the end consumers of models' output led to a slow and painful release process requiring significant manual efforts and downtime. Regulations led to excess redundancy and complexity for model governance, versions, and migrations between environments.
Datrics empowers risk analysts to have much easier access to the data sources, build reusable ETL processes and do predictive analytics using state-of-the-art machine learning models. It also provides seamless integration of successful models into the production environment. In other words, data scientists got:
- A single playground for different versions of data and model and environmental governance
- Collaboration capabilities to make relevant assets accessible to the right people in the company
- Proper versioning of datasets and risk models, replicability, and regulatory accordance
- A single-click deployment that enabled seamless integration of the new models into the production through Datrics API
In addition to risk analytics, the marketing team also showed interest in using Datrics to tailor the credit products outreach to the audience more efficiently, doing more experiments internally, and decreasing the load on the marketing analytics team.
Since deploying Datrics, the risk analysts have saved time and increased revenue by building and deploying models that improved the accuracy of the credit risk scoring.
- Better accuracy: faster experimentation by risk analysts (30% less time to build and test new models) without a need to write code leads to better models making it to production
- Improved deployments: quicker time-to-market for new models, hours instead of weeks for moving models between environments for validation or production deployments
- Time savings: reusable data pipelines for ETL and feature engineering. 40% less redundancy and duplication of data pre-processing pipelines