How we assisted Belvo in transitioning from a rule-based to a machine learning system for categorizing bank transactions for their customers across various countries and regions.


Transactional banking data are often partial, corrupted and don’t follow a strict syntaxis rule but they are essential for spend categorization, risk underwriting and fraud detection. The traditional approach of rule-based and Keyword matching categorisation heavily relies on manual updates and such solutions become difficult to maintain when scaling volumes.


Implementing a Machine Learning model with an NLP approach for the categorization of banking transactions that can solve this. Training a ML model that “understands” all the possible variants and adapt the rules to new ones based on its previous experience and what has been trained for.


The new ML model helped to increase x5 the speed of API system reply, increased by +20% the accuracy and dramatically reduced the cost of the model maintenance. The more the model learns from new transactions, the better and more precise it becomes.

Revolutionizing Transactional Banking with Machine Learning: Belvo's Improved Categorization Product

Belvo required to find a new way to improve their categorization product with Machine Learning techniques. The challenge was to react in real time to new transaction description exceptions without losing the accuracy and improving the performance of the solution, while analyzing millions of transactional data points in seconds and turning them into accurate and actionable insights for our customers.

Transactional banking data is vital for spend categorization, risk underwriting, and fraud detection. However, traditional approaches to categorization using rule-based and keyword matching systems have proven to be difficult to maintain when scaling volumes. This is because the data is often partial, corrupted and does not follow a strict syntax rule. However, implementing a ML model with an NLP approach for the categorization of Belvo transactions can help solve this problem.

The implementation of a Machine Learning model with an NLP approach for Belvo improved the speed and accuracy of their categorization process, leading to better risk management and fraud detection. This solution was cost-effective and easy to maintain, making it an ideal solution for banks looking to streamline their operations.