Belfius uses Microsoft Azure Machine Learning to help detect fraud and money laundering

Belfius recognized the opportunity of its cloud transformation to further scale up technologies such as artificial intelligence (AI) and machine learning (ML). Lacking an overview of all features, data scientists struggled with repetitive code. Azure Machine Learning, Synapse Analytics, and Databricks helped improve development time, efficiency, and reliability.

Continually progressing towards a more sustainable society is an essential tenet for Belfius. Belfius offers a full range of banking and insurance products for retail customers, small and medium-sized companies, public institutions, non-profit organizations, and large enterprises. The organization invites every customer—personal, business owner, government agency, municipality, or company—to actively participate in this effort. 

Eager to explore new ground and push boundaries, Belfius approaches all its activities with passion, purpose, and integrity. Customer satisfaction is at the core of Belfius’ mission. This includes ensuring that its products and solutions balance the interests of all stakeholders. Belfius strives to create long-term value for its customers and for the company, as well as for the community and the environment. 

Belfius recognized the shortcomings of its existing systems and the need to increase synergies to further scale technologies such as artificial intelligence (AI) and machine learning (ML). Belfius had been deploying AI tools to help with risk assessment and identifying unusual behaviors. It had also begun its digital transformation, moving key functions to the cloud to adapt to the changing needs of its customers. Its future cloud-based infrastructure will allow a dynamic and flexible use of AI and ML applications within the stringent privacy, security, and compliance requirements of the financial industry.

Lacking an overview of all features, Belfius data scientists were rewriting the same code repeatedly for different data models. “There was no versioning control and no search,” says Thibaut Roelandt, Lead Engineer for the Central AI team at Belfius. “Without versioning control, coding took longer, making it very challenging for us to act quickly to seize new opportunities,” explains Julie Dedeyne, a data scientist on the banking side of Belfius. “The bank was eager to have consistency between its various operational models and its training by using the same feature pipeline for both.” 

An opportunity for transformation

Belfius was keen to improve development time, become more efficient, and gain reliability. To do this, Belfius built on the Microsoft Intelligent Data Platform using services including Azure Machine Learning, Azure Synapse Analytics, and Azure Databricks. An early adopter, Belfius used Azure Machine Learning managed feature store, then in public preview, to operationalize ML features for an end-to-end ML operations workstream. At its core, managed feature store empowers machine learning professionals to collaboratively develop and use features in production. “Azure Machine Learning managed feature store holds a lot of promise,” says Roelandt. “Our data scientists can simply provide a feature set specification and let the system handle serving, securing, and monitoring of the features. This frees them from the overhead of setting up and managing the underlying feature engineering pipelines. They can also perform local development and testing of features.” Feature store can consume features from Azure Machine Learning, Azure Databricks, and more.

Managed feature store increases agility in building models because users can discover and reuse features instead of starting every time from scratch. It encourages faster experimentation with the ability to do local development and testing of new features. Consistent feature definition across the organization increases the reliability of ML models and supports versioning, just as Belfius had imagined. As features can be reused and materialization and monitoring are system managed, feature store reduces costs.

Expecting results with impact

Belfius initially identified two use cases for the new solution: Fraud detection and anti-money laundering. Fraud detection, under the auspices of Belfius insurance company, is an example of the importance of the online feature store, where the company needs quick access to the features to calculate a fraud risk score. Today, this calculation takes place via nightly batches. In the future, by using real-time scoring with the online feature store, the insurance company will be able to detect deceitful claims within minutes. “We are looking to build more models like this every year, gaining efficiency, meeting stringent regulatory standards, and offering more personalization to our customers,” says Roelandt.

Every year Belfius bank processes hundreds of millions of transactions, checking each one for potential money laundering activities. For suspicious transactions, an alert is generated. ML models are used to calculate risk scores on these alerts, allowing Belfius to have analysts focus on high-risk alerts and automatically close false positives. 

With Azure Machine Learning managed feature store reaching general availability (GA), Belfius can take advantage of industry-leading AI and ML technology. The cloud-scale data and app platform allows Belfius to deliver adaptive, responsive, and personalized experiences through intelligent applications built with Azure. As Roelandt says, “We want our data scientists to focus on creating transformative features rather than waiting for data engineering. We’re excited to provide them best practices and standardized processes across the company on our new corporate data platform.”

472 Views 2 Likes Introduce me  Read More