In a fraud detection system that analyzes transaction data in real-time, it is important to handle out-of-order events to ensure accurate fraud detection. Watermarking should be implemented to allow for a certain degree of data latency and handle out-of-order events effectively. By configuring the solution to handle out-of-order events in real-time, the system can maintain high performance and accurately detect suspicious activities without being affected by the arrival order of the transaction data.