Fraudulent activities have become increasingly prevalent everywhere, such as online booking, gaming apps, gambling sites, money exchange sites, e-commerce portals, online payment gateways, online banking solutions, insurance claim workflows, and so on. This wide range of activities makes it essential for businesses to have a robust, auditable, explainable model to detect and prevent fraudulent transactions automatically; hence any type of fraud can result in financial loss, reputation damage, and legal implications for victim businesses. Furthermore, fraudulent activities can negatively affect customer trust and loyalty toward business. A fraud detection model that is robust, auditable, and explainable can help mitigate risks of businesses significantly.

This blog post provides a decision model that utilizes a Fraud Rating Score to determine the potential for fraud during an online product booking process. It will involve human domain experts to intervene if the transaction cannot be clearly flagged.

Detect Online Booking Fraud

Assume a booking website with different services like reserving a hotel, renting a car, and booking a flight ticket. The fraud score model generates a risk score, parallel with the booking process, to recognize whether a transaction is safe or fraudulent.
The risk score of a particular transaction is used to make a decision about whether or not to approve or reject it. The decision is made automatically if the risk score is in clear range (low or high risk). However, if the risk score is in the buffer range, specialized human interactions are needed to make the decision about the transaction.

Create a Model

When evaluating the possibility of fraudulent transactions during online product booking, the process typically involves the following steps to determine a fraud rating score:

  1. Data Collection: The first step involves collecting data on both the product being purchased and the customer profile. This may include information such as the transaction amount, the type of product being purchased, the number of previous orders made by the customer, and whether the customer has any active disputes.
  2. Fraud Rating Score Calculation: Once the necessary data has been collected, a fraud scoring calculation is performed to assign a numerical value to the transaction. This value, known as the fraud rating score, reflects the level of risk associated with the transaction, which a higher score indicating a greater risk of fraud.
  3. Identification of Booking Fraud: Based on the fraud rating score, the decision model can either approve or reject the transaction. Transactions with scores falling within a predetermined acceptable range are approved, while those falling into a buffer range are flagged as potentially fraudulent and may undergo further review by a fraud analyst. Transactions with scores exceeding the threshold for acceptable risk are rejected outright.

Automated Decision

The process of determining whether an online booking transaction should be approved, rejected, or flagged as potentially fraudulent is a critical decision based on the calculated fraud rating score. In this context, State Through Processing used to automatically calculate the Fraud Rating Score and determine the Fraud decision.

Fraud Scoring (STP)

The Fraud Automated Decision consists of several sub-decisions, including:

  • the assignment of fraud scores for past orders,
  • active disputes,
  • product types,
  • and transaction amounts.

Domain Expert Intervention

To enable human interaction in cases where a fraud analyst's decision is required, a workflow is utilized, and task assignments can be made to groups or individuals for parallel or individual action, respectively.

Booking Fraud workflow

To explore more about this project, visit FlexRule Resource Hub.

Deploy as a REST Service

When a model is ready, either the automated decision model or the workflow with the human domain experts' intervention, you can publish them as a REST API service. There are multiple deployment options as a serverless model in the cloud or on-prem. With any of them, the REST API service deployment allows applications and processes to interact with either workflow or decision model using a REST call.

Once your decision service is ready, you can start testing by passing the JSON payload as in input, and the server will send the response as the JSON payload.
Fraud Scoring Fraud scenario

As a side note, any of the deployment options can be fully automated and integrated in your CICD pipeline using CLI toolkit. Alternatively, the deployment can be completed with a single click from the authoring platform.

Conclusion

As we continue to witness a surge in fraudulent activities in online booking, gaming apps, money exchange sites, e-commerce portals, online payment gateways, and so on, it has become increasingly important to implement an efficient automated fraud detection solution. The solution can cover the full workflow scenario to enable human domain experts intervention integrated with an automated decision that enables the straight-through processing (STP). This approach is designed to automatically determine fraudulent transactions and take action based on Fraud Rating Score, and if needed, it requests help from a human domain expert.

By utilizing this fully automated decision model with ability to intervene by domain experts, online businesses can prevent fraudulent transactions effectively and efficiently i.e. ensuring consistency on the decision on transactions, reducing time on processing, and explaining exactly why a transaction is flagged fraudulent.

Last updated March 10th, 2023 at 02:55 pm, Published March 6th, 2023 at 02:55 pm