Transforming Financial Automation with GenAI: A Cost-Effective Transaction-Labeling Solution for Finqware

The Market

According to IDC, the banking sector’s AI spending in Europe is expected to surpass $10 billion by 2025, with a projected 30% compound annual growth rate (CAGR) in the following years. This positions banking as the industry with the highest AI investment, far outpacing sectors like technology, retail, and other lightly regulated industries. Financial institutions are leveraging AI to enhance operational efficiency, reduce costs, streamline workflows, minimize human errors, and improve decision-making processes.

Finqware is at the forefront of this trend. As a provider of real-time treasury and open banking infrastructure, it helps large companies manage cash flows, reconciliation, and reporting across multiple banks and markets. By integrating AI into its operations, Finqware demonstrates that while automation may be complex, it can enable deeper insights and more intelligent financial decision-making.

The Challenge

Tagging and classifying banking or financial transactions across multiple banks and geographies is notoriously difficult. Each bank formats transaction data differently, and legacy rule-based systems can’t keep pace with the volume or variability. For Finqware, this meant:

• High manual effort for defining rules for tagging and

• Difficult and time consuming review processes

To scale effectively and maintain high standards for its unified transactions tagging catalogue, Finqware needed an intelligent, automated solution that could accurately tag transactions with reduced human intervention and without relying on static rules.

The Solution

Zitec, Finqware’s long-time technology partner, developed a GenAI-powered automated solution for transaction tagging, built on a cloud-optimized and modernized Google Cloud Platform (GCP) infrastructure.

In order to ensure scalability and operational efficiency, the solution leveraged advanced services including Vertex AI for model deployment and grounding, BigQuery for powerful data warehousing, Cloud Functions for processing, and Pub/Sub for reliable message queuing.

This way, the system was able to intelligently tag banking transactions by analyzing key attributes, delivering enhanced automation and accuracy at scale.
The project was delivered in just one month as a Proof of Concept (PoC), testing Google’s Google’s Gemini 2.5 Pro and Flash 1.5 models on a dataset of over 17 millions of anonymized transactions, with detailed testing performed on a 500k transaction subset.

The solution was centered on prompt engineering, systematically testing multiple Gemini models to identify the most suitable one for Finqware’s specific business requirements. This approach enabled the AI to autonomously classify transactions by detecting patterns and similarities within the data, moving beyond predefined rules and requiring minimal human oversight.

The results demonstrated strong performance across models. Gemini Pro achieved an average processing time of 3.19 seconds, while Gemini Flash averaged 4.38 seconds. The grounded Gemini 1.5 Pro model reached a peak accuracy of 91.3%, successfully tagging the vast majority of transactions automatically.

Key solution highlights:

• Extracting and processing of banking transaction data from diverse sources
• Utilizing AI models trained on anonymized transaction data for enhanced accuracy
• Integrating seamlessly with Finqware real time banking data aggregator

In essence, the GenAI tools learned directly from the processed data and applied their own logic to classify new transactions, unlocking significant potential for optimizing financial processes.

This PoC also demonstrates that GenAI has the potential to accomplish some tasks in a more autonomous way compared to traditional machine learning models. Also, it shows how various GenAI models compare to each other. For instance, within the tested GenAI models, Gemini 1.5 Flash demonstrated the potential to perform the tagging task at less than 1/16th the estimated cost of Gemini 1.5 Pro for the same volume, highlighting significant cost differences even between advanced models.

“This initiative confirms the potential of artificial intelligence in corporate financial processes. Our tests show that the system can significantly reduce manual work and the risk of errors in transaction classification. The next step is to expand the transaction catalog and further refine the AI models so that companies can benefit from more accurate classification and real-time financial visibility.”

Elena Cosma
Chief Product Officer, Finqware