PARI has implemented a real-time analytics system that processes user events, such as betting actions, within 10–15 seconds. The architecture utilizes Kafka for data streaming and ClickHouse with Kafka Engine and Materialized Views for processing and aggregation.
The system continuously aggregates data into metrics like turnover and betting volume using the AggregatingMergeTree engine. This replaces manual dashboard monitoring, which previously resulted in detection delays ranging from hours to days.
To detect anomalies, the company compares current metrics against a 90-day historical baseline, accounting for cyclical behavior. The detector checks data in 15-minute intervals to balance responsiveness with the reduction of false alerts.
The transition to real-time monitoring was accelerated by a significant data discrepancy identified in April 2026. Since its deployment, the anomaly detector has generated approximately 6,900 alerts, prompting the team to focus on signal prioritization.
The company is currently shifting from descriptive analytics to predictive modeling. Future goals include integrating predictive capabilities into bonus systems and personalization triggers to improve user retention and system stability.
