Google BigQuery gets managed AI functions to simplify unstructured data analysis

Google has boosted its BigQuery data warehouse with three new managed AI-based SQL functions to help enterprises reduce the complexity of running large-scale analytics, especially on unstructured data.

The functions — AI.IF, AI.CLASSIFY, and AI.SCORE — are designed to enable LLM usage for analytical tasks for enterprises on both structured and unstructured data directly within SQL queries. These functions work without prompt tuning or external tools, Google wrote in a blog post.

The new AI functions can be used to filter and join data based on semantic meaning using AI.IF in WHERE or ON clauses, categorize unstructured text or images with AI.CLASSIFY in GROUP BY clauses, and rank rows by natural language criteria through AI.SCORE in ORDER BY clauses.

Lowering the barrier to entry for data analysts

Traditionally, integrating LLMs into SQL workflows for AI-based reasoning of data has been a time-consuming, tedious, and costly affair as it requires data movement, prompt engineering, manual model selection, and parameter tuning, analysts pointed out.

The movement of data is typically required due to SQL’s inability to understand nuance and meaning of unstructured data, making advanced analysis, such as sentiment analysis or categorization, of customer reviews, support tickets, reports, etc., difficult, said Bradley Shimmin, lead of the data, analytics, and infrastructure practice at The Futurum Group.

To bypass this challenge, data analysts often had to export data from the warehouse, send it to a data scientist, and await the data scientist to send back enhanced, categorized data suitable for analysis using SQL, Shimmin noted, adding that the new AI functions “can literally collapse that entire workflow into a single query, using standard SQL syntax.”

The new managed AI Functions lower the entry barrier for data analysts as they can now adopt SQL-friendly syntax for AI-based reasoning of data without having to learn prompt engineering, HyperFRAME Research’s practice leader of AI stack, Stephanie Walter, pointed out.

“For enterprises, this means faster time-to-insight, less specialized skills required, and lower operational cost and risk,” Walter said.

Enterprises also gain from the managed nature of these new functions, Walter said, referring to Google’s backend automated handling of model selection, prompt optimization, query plan tuning, and endpoint management for these functions.

“This managed approach addresses the enterprise pain-point of complexity and operational risk: instead of analysts or teams having to decide which model variant to use, and optimize queries for latency and cost, Google abstracts that,” Walter said.

Growing demand for AI-based SQL functions

The integration of AI-based SQL functions in data warehouses is becoming a wider phenomenon. It can be viewed as a highly competitive space with all major vendors making comparable moves at varying stages of maturity.

While Databricks already offers AI Functions that can be used to apply generative-AI or LLM inference directly from SQL or Python, Snowflake provides AI_PARSE_DOCUMENT, AISQL, and Cortex functions for document parsing, semantic search, and AI-driven analytics.

Other warehouses, such as Oracle’s Autonomous Data Warehouse, also support AI workflows alongside SQL.

These integrations, according to Phil Fersht, CEO of HFS Research, point to a broader agentic evolution of data platforms.

“These managed functions act as foundational building blocks for more autonomous systems. Imagine agents that can query data, interpret results, and make decisions in real time, all without leaving the warehouse… giving enterprise data the ability to ‘think’ inside its own environment,” Fersht said.

The three new functions are currently in public preview.