Databricks launches Predictive AI Functions in Beta for SQL forecasting and anomaly detection

Databricks releases Predictive AI Functions in Beta, a set of SQL-native functions for forecasting, driver analysis, anomaly detection and value or class prediction, with no ML tuning required.

The company says the improved ai_forecast function uses a transformer-based time series foundation model, can account for holidays and external covariates, and outperforms other cloud data warehouse providers on the fev-bench benchmark.

Only ai_top_drivers is available now, through Genie, where users can ask plain-language questions such as why revenue dropped last quarter. Databricks says ai_forecast, ai_detect_anomalies, ai_predict_value and ai_predict_class are coming soon.