Fraud in Cards II - Neural Networks and Real time Decisioning.

The most sensible starting point is the implementation of neural network capabilities that achieve a balance between aggressive fraud detection and serving cardholders:

 

A strong neural network helps reduce fraud losses and focuses investigations on accounts and transactions that are most likely to be fraudulent.

 

 

Neural network technology builds cardholder profiles and utilizes predictive models to detect potentially fraudulent card usage.

 

The predictive models are used to determine the fraud potential of each ATM and POS transaction by evaluating it against a complete history of cardholder usage patterns as well as unique transaction characteristics that are known to be fraudulent and legitimate.

 

This process results in the recognition of any uncharacteristic transaction behavior.  

• Transaction
• Industry
• Cardholder 
• Merchant data

Are all used to forecast the likelihood of fraud.

With all neural networks, some level of false-positives will be generated. In a real-time environment, cardholders’ transactions will be denied if the transaction score exceeds a defined scoring threshold. The lower the scoring threshold, the higher the false-positive rate and the more likely that legitimate transactions will be denied.

Conversely, the higher a scoring threshold, the fewer actual fraudulent transactions you will prevent.

Real-time decisioning involves taking the neural network score into account in the authorization process.

While the industry norm has historically been to review the neural network score after a transaction has already been approved or denied, there is increased interest in considering the likelihood of fraud (neural network score) in the transaction authorization process.

Although this practice can drastically reduce fraud exposure, it can also have a negative impact on cardholders’ ability to perform legitimate transactions and potentially impact overall card usage due to a higher rate of false-positives experienced with current technology.

Financial institutions should carefully weigh the level of risk they are willing to accept against the level of cardholder satisfaction they wish to deliver. An appropriate balance of the two is necessary to ensure you are protected against the majority of fraudulent transactions and that cardholders continue to view their cards as a safe and dependable payment method.

Inserting the neural network process into the authorization path provides more information upon which to make approve/deny decisions, enabling you to stop suspect transactions before they are approved.

You will be able to select the criteria for those transactions you deem high risk that should be sent for real-time neural network processing (e.g. dollar amount, international, country codes, merchant category codes).

But even the most robust authorization systems can be enhanced. That’s where a rule authoring capability becomes a differentiator.

When you layer in rule authoring, you are ready to immediately address flash fraud – reducing fraud losses, maintaining consumer confidence and protecting the reputation of your institution.

A robust rule authoring service will provide more availability to fields in the online message and greater flexibility in the number and complexity of rules that can be deployed, automating actions so risk mitigation can start immediately.