Powering mobile fraud detection on Azure at scale
With more than 800 million people using mobile apps today, mobile bank fraud is increasing at a worrying pace, and the financial industry is seeing a 100 percent year-over-year increase in losses from mobile platforms. Selecting the right technology components for a 24×7 cloud-based infrastructure for fraud detection depends on understanding current—and sometimes vague—business requirements.
Technical decision makers are accountable for ensuring that enterprise capacity meets the demands of real-time fraud detection. They must feel confident that the solution they select will perform end-to-end through peak times. And the solution must provide a streaming analytics workflow that handles millions of continuously arriving events. This means an incredible amount of throughput. Technology decisions involving scalability require a modern cloud-based infrastructure to handle the ever-increasing volume. The Microsoft Azure platform is quickly becoming the trusted cloud for the banking industry.
Agile services that are easily configurable and tuned to meet the required latencies are paramount to a scalable system. Four main Azure components help to ensure a scalable fraud detection system:
- Azure Event Hubs can be scaled horizontally to handle the ingestion of multiple versions of concurrent data streams. You can deploy and assign Event Hubs to different consumer groups across the enterprise. This scale-out approach provides flexibility to expand fraud detection from mobile banking, for example, to include the internet banking channel.
- Azure Functions is a serverless architecture that you can use to manage and schedule the detection process flow. Depending on the fraud detection approach (stream or batch), we found that the highest throughput is realized when messages are batched and processed through a single Azure function, rather than processing one message per function call.
- SQL functions are low-latency data processes that handle parsing, preprocessing, aggregations, and storage. The capabilities of in-memory optimized SQL functions are built to meet the most demanding requirements for scalability and concurrency.
- Azure machine learning can handle concurrent requests, scoring each request with yes or no. Azure Machine Learning Web Services provides two options for scaling: 1) select a production web tier to support the API concurrency workload, or 2) add multiple endpoints to a web service if it is required to support more than 200 concurrent requests.
Data volumes will continue to grow. As they do, you can tap into Microsoft’s expertise in cloud-scale data ingestion and real-time analytics. The preceding technologies allow for elastic extension of compute capabilities while scaling cost-effectively. These benefits make moving to the Azure cloud an investment in the future.
Want to know more about how Microsoft addresses mobile fraud technologically and how Azure cloud capabilities can help? Read the Detecting Online and Mobile Fraud with AI use case, followed by the Mobile bank fraud solution guide, which provides actionable recommendations and solutions, and then engage with our coauthors on this topic by reaching out to Howard Bush on Twitter and LinkedIn and Kate Baroni on LinkedIn.
