Unlock powerful insights and forecasts with Predictive AI Functions
Databricks launches Predictive AI Functions in Beta, enabling SQL analysts and business users to run forecasting, anomaly detection, and driver analysis directly in SQL, Genie, or dashboards without advanced ML expertise.
Predictive AI Functions, now in Beta, allow business users and SQL analysts to perform forecasting, anomaly detection, driver analysis, and prediction without requiring advanced machine learning expertise. These SQL-native functions can be called directly in SQL, invoked in Genie using plain-language questions, or embedded in dashboards. By maintaining models centrally, Databricks ensures consistent, high-quality results without users needing to build or manage infrastructure.
The new ai_forecast function projects metrics such as monthly ride volumes in the NYC taxi trips dataset through 2027, revealing seasonal patterns in taxi demand. It uses a transformer-based time series foundation model pretrained on diverse time series data, producing accurate forecasts without tuning. Teams can add business-specific factors like holidays or external covariates such as pricing changes to refine forecasts.
Four new SQL-native functions address common ad-hoc analytics questions, including ai_detect_anomalies, which flags unusual days in a daily revenue time series. These functions run across hundreds of millions of rows and integrate seamlessly into queries and pipelines. Compared to custom model development, they reduce development time and increase accessibility for more users.
Predictive AI Functions deliver industry-leading accuracy, as benchmarked by fev-bench against leading cloud data warehouse functions, with ai_forecast outperforming others in task win rate. The functions enable business teams to answer questions like 'what's next,' 'why did it change,' and 'what's wrong' directly in dashboards, Genie, or SQL without relying on data science teams.