Case Study

Synthetic Data and Computer Vision to enhance anti-fraud services

Reply has supported Agos in the digital transformation of the financial sector, introducing a high-performance system for fraud prevention.

#Computer vision
#Synthetic data
#Automatic frauds detection



Automate the detection of financial fraud in consumer credit, incorporating new data sources into anti-fraud processes.


New anti-fraud solutions

Agos, a leading financial company in the consumer credit sector, has faced a series of challenges related to the identification and prevention of financial fraud. With the ever-increasing threat of fraudsters using false documents or stolen identities to obtain illicit loans, the manual analysis of identity cards remains a point of vulnerability.

To counter these threats, the company has adopted an innovative and automated approach that integrates new information sources and advanced data analysis techniques. Agos was thus able to strengthen its system for the detection and prevention of financial fraud and to protect the integrity of the company and its customers.


Identify counterfeit documents with synthetic data and ML

To respond to Agos' need to detect fraud, Target Reply has proposed an innovative approach that effectively combines elements of Computer Vision (through the use of Convolutional Neural Networks), Machine Learning algorithms and synthetic data in different phases of the data analysis process. Convolutional Neural Networks, due to their considerable efficiency in image processing, enable the extraction of visual characteristics used as input for Machine Learning-based classification algorithms. This automation facilitates document analysis for funding requests and ensures an effective identification of counterfeit identity cards. At the same time, the integration of synthetic data has resulted in a significant improvement in the quality and quantity of the available data, helping to increase the overall performance of the model.



The key role of synthetic data for a large and balanced dataset

As part of the project, synthetic data was used to effectively address the challenges related to fraud detection.

Detecting fraud requires a large and balanced training dataset. The conventional approach faced limitations due to the lack of authentic data and the need to safeguard privacy. Leveraging synthetic data generated through Generative AI models has overcome these challenges, has made it possible to use highly realistic artificial information that accurately reproduces the characteristics of counterfeit cards. This use of synthetic data has improved the ability of the model based on Convolutional Neural Networks (CNN) to detect fraud with high accuracy.

Particular attention has been given to the explainability of the analysis model. The ability to understand the system's decision-making process, namely how counterfeit identity cards are classified and which document attributes influence those decisions, is crucial for instilling greater confidence in Agos regarding the choices made by the system.

The results

Efficiency and precision, the Reply model

The cross-use of the different methodologies has proven to be highly effective: on the one hand, the use of CNN has made accessible a non-relational data source previously underused by algorithms, while the enrichment of the dataset by 25% through the use of synthetic data has led to an increase of 4.5% in the identification of false documents, thus reducing the number of false negatives detected. Thanks to this solution, Target Reply has therefore helped Agos to take a further step in the prevention of financial fraud.


Increased efficiency

Automating the document verification process has eliminated the need for manual verification by industry experts, reducing associated efforts and errors.


Introduction of new tools

The automation of document analysis has introduced an innovation in the control process, providing new tools and techniques to identify financial fraud in a more timely and efficient manner.


Reduction of time to market

The use of synthetic data has made it possible to speed up application development and reduce release times. Therefore, it is possible to implement the solution quickly and obtain benefits in a shorter time.


Increased performance

The enrichment of the dataset with synthetic data has improved the consistency and representation of the data, increasing performance in identifying false documents.


Agos is a financial company that deals with the implementation of its customers' projects and supports the sales of partners in different markets through a wide range of products and services: personal loans, finalized financing, credit cards, the sale of a fifth of the salary/pension, leasing and a wide range of insurance services.


Target Reply is the Reply group company specialized in the creation of Big Data and Advanced Analytics solutions. Target Reply supports companies from identifying needs to designing and implementing solutions through data integration, data modelling and predictive analysis technologies, using the most innovative tools in the field of Business Discovery and Big Data. Target Reply has gained significant experience with major Italian and foreign business groups and is able to operate in all major markets: Telco, Finance and Manufacturing.