Snowflake is set to revolutionize complex data analytics with the launch of Cortex Analyst, an advanced agentic AI system now in public preview. Announced at the company's Data Cloud Summit in June, Cortex Analyst offers businesses a conversational interface that allows users to engage with their data in plain English, simplifying the analytics process.
Cortex Analyst streamlines self-service analytics by converting user queries into SQL, querying the data, and validating responses, all performed effortlessly by the AI. According to Baris Gultekin, Snowflake’s head of AI, Cortex Analyst utilizes a collaborative approach involving multiple large language models (LLMs), achieving an impressive accuracy rate of around 90%. This level of precision significantly outperforms existing LLM-powered text-to-SQL solutions, including those offered by Databricks, facilitating faster analytics workflows and empowering users to make timely, informed decisions.
Transforming Analytics with Cortex Analyst
As companies increasingly invest in AI-driven forecasting and generation, data analytics remains crucial for business success. Organizations harness insights from structured historical data to inform decisions in areas like marketing and sales. However, traditional analytics often relies on business intelligence (BI) dashboards that visualize data through charts and graphs, creating rigidity. This can hinder users who seek to analyze specific metrics and often require assistance from analysts, which can slow down the entire decision-making process.
Gultekin illustrated this challenge: “When a dashboard reveals something unexpected, users typically have immediate follow-up questions. Analysts may take time to gather and deliver responses, leading to prolonged decision cycles.”
In response to this challenge, Snowflake recognized the limitations of early LLM offerings, which struggled with accuracy. Their internal benchmarks revealed that state-of-the-art models like GPT-4 delivered only about 51% accuracy in analytical insights. In contrast, dedicated text-to-SQL models, such as Databricks’ Genie, achieved 79%. Gultekin noted, “Accuracy is critical when asking business questions. We aimed to double that accuracy to approximately 90% by integrating multiple large language models into Cortex Analyst.”
How Cortex Analyst Works
Cortex Analyst redefines the analytics landscape by enabling natural language queries that undergo thorough processing through various LLM agents. These agents assess the user's intent, execute the SQL query, and ensure the accuracy of the returned data, grounding responses in Snowflake's data cloud.
Snowflake emphasizes providing semantic descriptions of data assets during setup, which significantly enhances understanding and contextualizes user queries. Gultekin explained, “In real scenarios, data can involve thousands of tables with complex naming conventions. By specifying metrics like ‘Rev 1’ and ‘Rev 2’ in semantic descriptions, our system comprehends their meanings.”
Cortex Analyst is accessible via a REST API for easy integration into applications, allowing developers to customize user experiences. Additionally, businesses can utilize Streamlit to develop tailored applications powered by Cortex Analyst.
Currently, around 40-50 enterprises, including pharmaceutical leader Bayer, are piloting Cortex Analyst, with the public preview expected to expand accessibility as companies adopt LLMs cost-effectively. Snowflake plans to introduce new features, including support for multi-turn conversations for a more interactive user experience and enhanced compatibility with complex tables and data schemas.
With Cortex Analyst, businesses can harness the power of language models for analytics without the burdensome implementation costs typically associated with such advanced technologies.
Ranking
