Snowflake’s Cortex AISQL aims to simplify unstructured data analysis

Snowflake is adding generative AI-powered SQL functions to help data analysts and their organizations analyze unstructured data with SQL.

These new AISQL functions will be part of Snowflake’s Cortex, a fully-managed service inside its Data Cloud providing the building blocks for using LLMs without the need to manage complex GPU-based infrastructure. It already includes serverless functions that can be called from SQL or Python to analyze data or build AI-based applications.

AISQL builds on these serverless functions to enable analysis of unstructured data, improve query performance and eliminate the need for data analysts to rely on data engineers and developers, said Christian Kleinerman, EVP of product at Snowflake.

The ability to query unstructured data is important for enterprises looking for more accurate business insights and faster decision-making.

The ability to access unstructured data directly with SQL syntax is not a new capability for Snowflake but these generative AI-powered functions makes the task easier, said Constellation Research principal analyst Michael Ni. Before the introduction of AISQL, enterprises could use various ways to access  unstructured data via SQL — using Document AI to load data in documents, using the TEXT column or creating a table with a FILE column and using SQL to run queries against it, although with some limitations.

Google’s BigQuery ML also allows enterprises to use SQL to write queries on the results of machine learning models prepared on unstructured data

No more waiting for data engineers

But more importantly, Ni said, AISQL could eliminate the need for data analysts to wait for data engineers or scientists.

“By embedding generative AI into familiar SQL syntax, Snowflake enables data analysts to execute tasks like sentiment analysis, image classification, and document parsing without writing Python or managing ML pipelines — operationalizing AI at the query layer, not just in the lab,” he said.

Another benefit of AISQL, according to Bradley Shimmin, lead of the data and analytics practice at The Futurum Group, is how it can help make Snowflake a unified query engine for enterprises to analyze all types of data.

Snowflake is not the only data warehousing software provider looking to merge unstructured data and structured data for analytics, Shimmin said: other data warehouse vendors such as Databricks, Google, and Oracle have either already introduced a way to do it or are developing something.

But they need to do more in the analytics space to drive value for enterprises, especially with SQL, perhaps bringing in retrieval augmented generation (RAG) methodologies or increasing the accuracy and quality of generated SQL statements, he said.

IBM is one vendor doing more in the SQL field, Shimmin said: It recently introduced an update with watsonx.data where the company enhances unstructured data sources destined for RAG pipelines by adding query-able structured data. “Users can then blend SQL and semantic search to optimize data access and accuracy,” he said.

Cortex AISQL uses large language models (LLMs) from Anthropic, Meta, Mistral, and OpenAI among others, to generate SQL functions. On the performance front, Snowflake claims it can reduce query response time by 30-70% depending on datasets and also save up to 60% of cost when filtering or joining data. Cortex AISQL is currently in public preview.