FraudIa
ARTIFICIAL INTELLIGENCE IN THE FIGHT AGAINST CREDIT CARD FRAUD
Quick links:
FraudIA has been originally developed for Transactis the joint venture between two equal partners Société Générale and La Banque Postale specializing in the management of real-time, high-volume payment systems. fraudIA is a software solution to detect credit card fraud in real-time. fraudIA complements the existing fraud detection systems such as expert rules engines.
Plus points
FraudIA™
recall on frauded transactions undetected
by other tools*
credit card transactions
per day
Response in less
than 50 milliseconds
Reduce financial losses
Avoided fraud up to several million euros per year
250 transactions/second
Tested on volume metrics in real time
Speed, security, and simplicity of processing
Simple API
*The values are estimates based on simulations performed on historical data from 2017-2019. The values of recall differ between 5 and 35% depending on the type of transaction (VAD, withdrawal, payment) and the bank. The percentage corresponds to the fraction of reported fraud detected by the FraudIA algorithm with an accuracy of 30%. Reported fraud is defined as fraud reported by the bank’s customers, i.e. fraud that has not been discovered by any other tool used by the bank.
Performance in production may differ and is highly dependent on changes in the configuration of other fraud detection tools and changes in the behavior patterns of fraudsters and customers between the preparation of the model and its use in production.
AI in the service of fraud detection for online payments, face-to-face payments, and ATM withdrawals.
fraudIA helps to detect and block fraudulent transactions in online payments, face-to-face payments, and ATM withdrawals. It reduces the amount of fraud impacting cardholders.
High processing speed is one of the essential features of any credit card transaction processing system. fraudIA must produce an AI score within 50 milliseconds!
In this short time interval, fraudIA performs complex data processing, optimized to handle a load of several hundred transactions per second.
- Analysis of the card transaction history
- Calculation of model features
- Model Scoring
FraudIA™
fraudIA uses real-time machine learning to calculate the probability that a transaction is fraudulent. The calculation is based on :
- information included in a request for authorization
- a history of the credit card related to the transaction
- a history of all past transactions recorded by the bank
- a history of the objected cards
- a history of fraudulent transactions
fraudIA regularly retrains the machine learning models to include new fraud scenarios manifested by fraudulent transactions and objected cards.
fraudIA has been developed by a team of Data Scientists, developers, and architects. It contains a custom model adapted to the specifics of the bank’s data.
The advantages of fraudIA
- Simple API to connect with an existing software stack
- Technical performance adapted to the processing of large volumes of data in real-time
- Scoring performance: based on AI algorithms
- Security: validated by a client security audit
EXAMPLE OF THE FRAUDIA INSTALLATION
at Transactis
Transactis is the joint venture between Société Générale and La Banque Postale. It manages the information systems for e-banking activities, developing new SEPA & international direct debit and wire transfer management platforms.
Context
Transactis is a major payment processor in France. Every day, up to 12 million transactions for more than 3 million cards pass through its servers. Each year, more than 1TB of data is produced by the 15 million credit cards in circulation managed by Transactis.
Implementation / Expertise
To answer these technical challenges, Probayes employs Big Data and NoSQL technologies. The data flow between cluster nodes is managed by Apache Kafka, while in-memory databases such as Couchbase and RocksDB ensure extremely fast response times for transaction data storage and retrieval. To ensure process isolation, high availability, and scalability, services are managed by containers and an orchestrator.
Results
The solution has been deployed and performance data will soon be available.
« A real-time, AI-based solution to detect fraud built by Probayes, is directed towards the banking and insurance industry.
fraudIA manages more than 200 transactions per second with response times shorter than 50 ms per transaction.
The solution complements existing tools and enables the detection of fraud patterns in credit card transactions (withdrawals, face-to-face payments, and online payments) that remained undetected by traditional expert systems. The features contribute to a net reduction in fraud. »
Vincent Maigron
La Banque Postale – Head of Anti-Fraud – Operational Management
Société Générale – Head of the Fight Against Fraud – Carriers & Merchants
Probayes solutions for the financial services industry
FraudIA™
Detect credit card fraud
Churn
Detecting customers that are likely to leave
Chatbot
Giving customers and employees a conversation agent
Copyright 2019 – Probayes – 53 avenue Kuntzmann – 38330 Montbonnot – France – Tel: +334 76 42 64