Enhancing AI Accuracy with Admin Feedback for Suspicious Transaction Monitoring
Updated
At Expensya, we understand that balancing security and user experience is crucial, especially when it comes to transaction monitoring. Our AI-driven system currently analyzes transactions made with Expensya cards to identify and flag suspicious activities.
Alerts for high-risk transactions are sent to the Expensya cards admin, who can review and flag these transactions as either suspicious or valid.
To improve the system's effectiveness, we've put in place a feedback loop where card admins can provide insights when marking a suspicious transaction as valid. Here’s the process:
When a Transaction Is Flagged as Suspicious: Admins have two options:
- Flag as "All Good": If an admin marks a transaction as valid, they will be prompted to provide feedback to help us improve unusual transaction detection accuracy.
- Flag for Internal Review: If a transaction is confirmed as suspicious, no feedback request is needed, as this input reinforces the AI's detection capabilities for true positives.
Using Feedback and Contextual Understanding to Improve the AI Algorithm
By leveraging the valuable insights provided by admins and integrating essential transaction details, the AI system will become more refined, reducing false positives and optimizing its ability to identify genuine suspicious activities.
Here's how we’re enhancing the algorithm:
- Incorporating Admin Feedback: The AI will learn from admin classifications, improving its ability to distinguish between suspicious and legitimate transactions. Over time, this feedback will enhance the algorithm’s decision-making process, reducing unnecessary alerts.
- Utilizing Budget-Related Information: By integrating budget data into transaction analysis, the AI can more accurately assess the context of each transaction. For example, if a transaction aligns with a pre-approved budget, it will be less likely to be flagged as suspicious.
- Excluding Refunded Transactions: Transactions that have been refunded will be excluded from the suspicious activity analysis. This ensures the AI focuses on relevant financial activities and avoids unnecessary alerts.
Use Cases Illustrating AI Improvements
To illustrate how the enhanced AI will function, here are a few real-world scenarios:
- Scenario 1: Budget-Approved Travel
- Context: A cardholder makes a successful transaction in Tunis under a pre-approved budget titled "Trip to Tunis."
- Improvement: When analyzing this transaction, the AI will factor in the budget reason. The transaction will not be flagged as suspicious, reducing unnecessary admin alerts.
- Scenario 2: Familiar Merchant with High Amounts
- Context: A cardholder makes a high-value purchase with Figma, a merchant where previous high-value transactions have been marked as "All Good" by admins.
- Improvement: The AI will recognize the history of admin feedback for this merchant and consider the current transaction as valid, thus avoiding false alerts.
- Scenario 3: Refunded Transactions
- Context: A cardholder completes a transaction that is later refunded.
- Improvement: The AI will identify the refund and automatically exclude the transaction from suspicious activity analysis, eliminating unnecessary alerts for admins.
Our goal is to create a smarter, more adaptive AI system that reduces false positives and improves overall accuracy in transaction monitoring. By leveraging admin feedback and budget-related data, we’re building a solution that keeps financial operations secure without sacrificing efficiency or trust. With fewer unnecessary alerts, admins can focus on genuine risks, confident that the Expensya card product is working intelligently and effectively.
