By: Ruan Mingwei, Li Xuan, Wu Chunliang, Pablo Cal

Nowadays Artificial Intelligence (AI) and Machine Learning (ML) technologies have been widely used across different industries. If harnessed properly, they could deliver benefits over many business sectors. The online payments industry leverages AI and ML extensively for risk management, so that we can learn from the past and protect customers from fraudsters and financial losses. PayPal has been at the forefront of this, developing innovative and state-of-art risk management systems for years. Accuracy is one of the most important criteria for risk decisioning and is dependent upon complex ML techniques and algorithms. However, these technologies inherently have a barrier of explainability that must be overcome. Additionally, it is equally, if not more, important to enhance the interpretability of risk decisions. Within PayPal, eXplainable Artificial Intelligence (XAI) has been introduced to achieve this goal.

What is XAI?

XAI is an emerging technology that converts machine learning models into transparent and interpretable results, which can be easily comprehend by humans. Although AI and ML can help to make quick, efficient, and accurate decisions in many business domains, it is difficult for users to understand how and why those decisions are derived, due to the ”black-box” nature of ML technology. XAI creates a shared understanding between humans and machines about machine learning process, and how the machine reached its results. This enables AI to be more transparent, interpretable, and trustworthy to humans.

How does PayPal Integrate XAI into Our Risk Eco-System?

At PayPal, a real-time XAI model has been developed, powered by AI and the unique reason code mapping technique, to calculate the importance of model components, and then aggregate components to yield specific reasons. The reasons help explain the risk declines and can be integrated into the risk eco-system.

PayPal has a sophisticated risk management system to mitigate fraud pressure. Thus, integrating XAI into our risk-management system and making it scalable was by no means a simple task. A practical and streamlined paradigm was developed to provide guidelines for XAI intelligence implementation.

1. Integrate with Risk Strategy: Optimize Risk Management Decision

Risk models and strategies are developed based on historical global data. Some reasons (features) might not be applicable for all use cases. With XAI reason codes and risk experts’ intelligence, PayPal is able to identify what parts of the models and strategies are working as expected and what parts need adjustment. This enables us to identify transactions with high-risk concerns as fraudulent, as well as improve decision accuracy when identifying legitimate transactions.

2. Channel with Customers: Maintain Trust and Safety

We keep transparency by providing relevant reasons to customers who are facing risk friction. This enables customers to have a better understanding, as well as build trust with PayPal.

Additionally, the XAI reason helps identify grey area transactions. Previously, these transactions would have been declined, but now we provide these customers with a customized path towards authentication (such as asking the customer to provide certain proofs). This enables legitimate customers to still find a way out and complete transactions.

3. Integrate with Internal Stakeholders: Improve Transparency and Efficiency

Providing decline reasons to relevant stakeholders within PayPal fosters:

· Transparency, which also ensures risk actions comply with regulations and ethics.

· Efficiency, since it reduces the manual work to review every risk escalation.

Summary

Our risk system at PayPal, powered by AI and ML, has evolved and learned from increasingly complex and sophisticated fraud behaviors. This entailed more computations of a paramount level of data points and risk intelligence, which made risk decisions more difficult to understand. Therefore, XAI emerged to serve as a bridge to build trust with humans and made our risk-management ecosystem more manageable and understandable.

With the XAI-reason based paradigm, PayPal streamlines, simplifies, and optimizes existing risk-management systems to become an innovative risk-management metaverse. We now have better performance, maintain greater trust with our customers, and we enable better transparency with internal stakeholders.