Alerting and anomaly detection at Google

Statistical machine learning for alerting and anomaly detection at Google.

Problem: When an ads revenue metric suddenly drops, someone has to find out which product, region or system caused it, fast. At Google that metric was billions of dollars.
User: Engineering product teams at Google, including YouTube advertising, ads targeting, Google Trends forecasting, and SRE alerting.
Decision: I was tech lead for research and data science teams building statistical machine learning models and production systems for engineering product teams. The alerting method compares what we observed with what we expected across every breakdown of the data, then points to the part that caused the anomaly. It ran in production, not just on paper, and it’s described in a granted patent, US 12,353,316 B2.
Outcome: We launched a monitoring system that alerts on billions of dollars of Google ads revenue. I also built the first prototypes in ads targeting and Google’s flu prediction, and I published 9 research papers while at Google.

Patent: US 12,353,316 B2, granted July 8, 2025; assigned to Google LLC. More on the research page.