ISG Software Research Analyst Perspectives

SQL AI Functions Combine Old and New Approaches to Data Processing

Written by Matt Aslett | Apr 9, 2026 10:00:00 AM

I have previously written about the impact that artificial intelligence (AI) and agentic AI are having on the requirements for data platforms. AI and data platforms are intrinsically linked, with the latter providing the underlying data persistence and data processing capabilities that support model development and training, as well as the inferencing capabilities of intelligent operational applications. AI agents that automate the execution of complex business processes and the delivery of business outcomes require data platforms that are agile, adaptive, automated, active and part of the agentic process. In addition to supporting AI workloads, many data platform providers are incorporating AI features into their data platforms products, including AI assistants to accelerate administration tasks, as well as AI functions that extend SQL (Structured Query Language) processing capabilities to automatically process data using AI.

AI has the potential to transform the data processing landscape as providers adapt their products to support generative and agentic AI workloads. I assert that through 2028, data platform software providers will increase their native support for AI and generative AI (GenAI) to enable in-database training, tuning and inferencing. Additionally, GenAI is transforming many data administration and processing tasks. For example, GenAI can be used to automate and accelerate the processing, transformation, filtering and classification of structured data, as well as to extract data from unstructured documents, reports and image files. Adoption of GenAI does not mean throwing away existing technologies and skills, however. This functionality needs to coexist with the established essential components of the technology industry, one of which is SQL.

SQL is a ubiquitous part of the technology landscape, thanks in part to the dominance of relational databases. SQL is not only the primary language used to manipulate and retrieve data in relational databases, but variants of SQL have also been created to manipulate and retrieve data in non-relational databases as well. SQL functions are a key element of SQL, providing a set of pre-defined commands that can be used to perform specific database tasks. Data platform providers offer built-in functions to perform common calculations, as well as enabling database administrators to create user-defined functions (UDFs) that fulfill their own specific requirements.

Since the emergence of GenAI models, data platform providers have introduced new SQL AI functions that combine the advantages of SQL and GenAI by providing a set of specialist SQL functions that invoke AI to automate data processing. These SQL AI functions are not to be confused with general-purpose AI functions, which can be created by users with full control over the choice of model, prompt and parameters. In comparison, SQL AI functions are purpose-built and optimized for specific tasks, enabling the use of standard SQL operators to automatically perform tasks such as filtering and classifying text, ranking data according to quality or similarity and extracting text and information from unstructured documents.

Automatically extracting enterprise data from PDFs, reports and image files has become a key focus for many data and AI providers. A significant amount of enterprise knowledge is contained in unstructured documents that can be unlocked to provide context in support of generative and agentic AI applications. Information from tables, figures, diagrams, descriptions and metadata extracted using SQL AI functions can be converted to vectors and utilized with retrieval-augmented generation to improve trust in the output of GenAI. Existing functions, models and data extraction products can already be used to extract text and information from unstructured documents, of course, but additional processing and classification is required to ensure the extracted information retains its context—such as the relationship between tables, diagrams and written text. This information can more easily be captured using GenAI, providing the context for knowledge graphs and semantic models.

The delivery of SQL AI functions is still relatively new. Slightly less than one-half (49%) of the 51 data platform providers assessed as part of the forthcoming 2026 Data Platforms Buyers Guide graded A- or above for SQL functions that invoke AI to automate data processing, although 57% are offering SQL AI functions. Differentiation between data platform providers in relation to SQL AI functions can be found in the breadth of capabilities that the functions perform. Based on our research conducted for the 2026 Data Platforms Buyers Guide, the most popular tasks being addressed by data platform providers using SQL AI functions are (in order):

  • data classification
  • data summarization
  • chunking and vector generation
  • sentiment analysis
  • the extraction of structured data from unstructured content
  • language translation
  • text generation and completion
  • similarity analysis
  • entity extraction

There is a long tail of potential use cases, however. Other, less widely available uses for SQL AI functions include data masking/redaction, grammar correction, prediction/forecasting, model creation and management, text filtering, image manipulation, database configuration and security tasks. Some providers offer SQL AI functions addressing just a handful of tasks while others are addressing a dozen or more. I recommend that all enterprises assessing data platforms and providers add SQL AI functions to the list of capabilities assessed. When assessing data platform providers and their use of SQL AI functions, it will be important to identify which and how many tasks are addressed.

Regards,

Matt Aslett