data science

How a Belgian Bank Used Artificial Intelligence to Manage Asset Liability and Detect Fraud


Our client, a Belgian bank, was looking to introduce new risk modelling approaches that together reduced costs and strengthened balance sheets while detecting fraud. 


Banks are subject to a number of risks, from financial risks to cyber threats.
Successful risk management involves ensuring that these risks are appropriately identified, measured and monitored. What’s more, being able to manage risks and complex financial issues has proved to be a competitive advantage.

Artificial Intelligence has the potential to transform banking and specifically its asset liability management while automating certain processes, in particular screening far more threats than just human teams alone. 

Additionally, big data analysis can decrease risk by providing insights into the products and services to be offered to different customer segments, limiting banks' credit exposure during downturns and preventing capital dilution during recessions while providing greater customer experience.


Our client first contacted us to improve risk assessment using artificial intelligence. Thanks to the existing data, we were able to optimise asset liabilities, resulting in improved balance sheet structuring to support their profitability. 

To do so, we compared market data, from interest rate trends to price changes in asset portfolio, data on customer behaviour and environmental risks, that we combined with other risk calculation models.

We also provided our client with information on the products and services to be offered to the different customer segments. As a result, our client could foresee the needs for each financial product and could invest appropriately over time, while providing an optimal experience for its clients.

Further to our mission dedicated to asset liability and risk management, our client asked us to support them in both reducing manual tracking interventions and increasing the accuracy of real-time fraud detection. 

To reduce manual tasks, we performed automatic client identification and authentication while cross-checking historical data with real-time transaction data to increase the accuracy of the model. This allowed monitoring and early detection of anomalies in transactions and behaviour. 

Real-time identification of potentially fraudulent activity allows to react quickly, thus mitigate risks and problems.  

All in all, a fraud detection model was based on the idea that fraudulent transactions have different forms. Fraud detection algorithms use these differences to automatically predict the likelihood of a transaction being fraudulent before it is completed.


The efficiency of our models rests upon data. Adding external data sources will then make our solutions even stronger.