BrazilFinance and insurance2024 – ongoingAME Digital
Machine learning fraud detection for digital payments
AME Digital's fraud models reached 90% accuracy after it moved its machine learning pipeline to Databricks
Problem
AME Digital, a Brazilian digital wallet and payments business with more than 33 million customers, had to extend fraud detection to more use cases as its data grew from 1TB to over 400TB. Its on-premises setup was slow, and many suspicious transactions needed manual review.
Approach
AME Digital, working with the partner Eleflow, moved its data and machine learning work to Databricks and managed model cycles with MLflow. An algorithm flags possible fraud in real time and blocks suspicious transactions, which cuts manual reviews. The team also centralized more than 700 datasets.
Results
Takeaway
The speed gains came with the move off on-premises systems, so they cover the whole pipeline and not only the fraud models. The company did not say how the 90% accuracy was measured.
Categories
- Industry
- Finance and insurance
- Business function
- Finance, risk and legal
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
Databricks reports that AME Digital runs fraud detection models on its platform and manages them with MLflow. It gives 90% model accuracy, a 3.2x faster pipeline, job times cut from 5 hours 30 minutes to 50 minutes, and 38% lower data management costs and 34% lower operational costs.
Full details are in the original sources.
