Development of an Anomaly Detection System to Achieve Zero Ad Delivery Incidents

At mid-sized advertising agencies, the high number of accounts per employee often led to delayed detection of abnormal advertising performance, resulting in frequent client disputes.
While automated report updates were already implemented, critical issues—such as sudden spikes in specific ad delivery costs or suspensions due to review rejections—could only be discovered if each account manager manually checked the details.
Compensating clients for delivery failures and the increased workload for managers reviewing reports became serious problems.

Our company has provided an anomaly detection service developed in-house.
This service notifies responsible personnel via Slack when anomalies are detected, alerting them to the occurrence and the numerical changes.
Based on delivery data collected by the existing advertising reporting system, it automatically detects issues such as delivery stoppages, review rejections, and sudden performance changes using pre-set thresholds and rule-based logic, notifying each responsible person daily.
This significantly reduces manual report verification tasks and establishes an automated monitoring system that prevents critical issues from being overlooked.

The implementation of an anomaly detection system has eliminated compensation payments to customers due to delivery incidents.
Additionally, the time required for staff to review reports has been reduced by 60%, creating an environment where they can focus on more strategic advertising operations. This has resulted in a significant improvement in customer satisfaction.
Furthermore, the enhanced accuracy of anomaly detection has earned high praise from staff, who report feeling "more confident in their operations."