The apply(risk) conference highlighted four key insights for developing machine learning systems in risk and fraud detection: investing in high-quality features and data is crucial, as well as expertise in handling unique data challenges such as drift and imbalance. Additionally, compliance and data governance can create technical challenges that require careful access control and management. Furthermore, it's essential to keep the business context in mind when evaluating trade-offs between model accuracy and performance, and to remember that a delicate balance must be struck between catching bad actors and maintaining a positive experience for legitimate users. The conference also emphasized the importance of using feature engines and ML observability tools to simplify data engineering and improve model health.