Google adds six AI analytics functions to BigQuery

Google has added six augmented analytics table-valued functions to BigQuery. They are also available in BigQuery Conversational Analytics.

The new functions automate several common analytical tasks inside BigQuery, including identifying the causes of metric changes, measuring the effect of business interventions, detecting change points in time-series data, and examining correlations, trends and recurring seasonal patterns.

The suite includes AI.KEY_DRIVERS, AI.CAUSAL_EFFECT, ML.CORRELATION, ML.DETECT_CHANGE_POINTS, ML.TREND and ML.SEASONALITY. The functions produce structured SQL outputs and run where the underlying data is already stored, allowing analysts to use them without moving information into separate tools.

The release reflects a broader push among cloud data providers to bring more statistical analysis and machine learning tasks directly into core data warehouse products. Rather than requiring users to stitch together separate software for exploratory work, vendors are increasingly embedding higher-level analytical methods into SQL-based workflows.

Six functions

Each function addresses a distinct line of inquiry. AI.KEY_DRIVERS identifies the main factors behind rises or falls in a metric between two periods or groups. AI.CAUSAL_EFFECT estimates the impact of an action or event against an expected baseline.

On the machine learning side, ML.CORRELATION measures the direction and strength of relationships between numeric metrics. ML.DETECT_CHANGE_POINTS flags dates or intervals where a pattern shifts. ML.TREND separates underlying movement from shorter-term volatility, while ML.SEASONALITY identifies repeated cycles across time periods such as days, weeks, months or quarters.

The functions can also be chained together, so the output of one query becomes the input for another. That allows users to move from spotting a structural break in a metric to identifying likely drivers and then testing the size of the resulting business effect.

Austin example

To show how that process works, Google used the public Austin Bikeshare dataset in BigQuery. The example first applied ML.DETECT_CHANGE_POINTS to daily trip totals to find statistically significant shifts in usage over time.

The analysis highlighted a structural shift in February 2018. Google linked it to the day the Austin City Council passed the Dockless Mobility Pilot Program, which changed the city's transport landscape by integrating shared electric scooters and bikes.

Google then used AI.KEY_DRIVERS on a period around that break to compare trip patterns before and after the change. The analysis examined dimensions including start station, end station, subscriber type and bike type.

According to the published results, overall trip count increased by 374.7%, or 40,159 trips, between the reference and interest windows. The increase was concentrated in U.T. Student Memberships, which rose by 7,167.1%, and trips ending at the 21st & Speedway @PCL station, which rose by 20,739.1%.

Those findings matched Austin Bikeshare's response to the policy change, including a promotional partnership with the University of Texas that offered free annual memberships to students.

A third step used AI.CAUSAL_EFFECT to estimate how much of the increase exceeded the level that would otherwise have been expected. Using a counterfactual model, the analysis concluded that the programme led to a 358% surge above organic baseline projections, equal to an estimated 89,775 incremental trips, with a stated 99.9% probability of causal effect.

Conversational use

The functions are also now integrated into BigQuery Conversational Analytics, a tool that lets users query datasets in natural language. In practice, that means an agent can call multiple functions in sequence from a prompt and return findings through a conversational interface.

One example used the public Chicago taxi trips dataset. A prompt asked which metric correlated most strongly with drivers receiving tips, then requested an attribution analysis to identify which categorical dimensions most strongly drove that metric.

Using ML.CORRELATION and AI.KEY_DRIVERS together, the analysis found that credit card payments were the main positive driver of trip distance, adding 1.65 million miles because of longer routes and digital tip tracking. Trips originating from O'Hare International Airport were another positive factor, contributing an additional 1.10 million miles among tipped credit card rides, while cash transactions were a negative driver at 652,960 miles.

Another example used the Iowa liquor sales dataset and combined ML.TREND with ML.SEASONALITY. The output indicated long-term growth in bottle sales from 1.3 million to 1.5 million in 2012 to about 2.6 million in recent years, with recurring peaks in October, December, May and June and weaker sales in January and February.

The additions also point to a growing overlap between analytical databases and generative AI tools. By returning compact SQL outputs, the functions can serve as callable skills for software agents that conduct structured investigations rather than simple keyword searches.

That could make BigQuery more useful for analysts who want to move beyond dashboards and standard reporting into repeatable diagnosis of why a metric changed, whether a trend is sustained, and how much a specific intervention contributed to the result.