Snowflake: Data strategy essential for AI

During this latest earnings call season, many banks, exchanges, and financial services providers have touted their large language model (LLM) capabilities and their focus on generative AI. But having an AI strategy—and that includes a generative AI strategy—cannot work without a data strategy. 

While there is huge potential for institutions to leverage generative AI and LLMs, setting the data foundations to enable firms to execute their AI strategies is more important, Frank Slootman, chairman and CEO at Snowflake, said during the company’s Q2 earnings call on August 23. 

“Generative AI is at the forefront of customer conversations. However, enterprises are also realizing that they cannot have an AI strategy without a data strategy to base it on. We have a head start in this race, as the epicenter of highly curated, optimized and trusted enterprise data,” he said. 

Particularly when it comes to deploying LLMs, firms need to have highly organized, optimized, trusted, and sanctioned data. 

“If you think you can just drop a model on top of a data lake and just see what happens, that’s not going to end well and that’s what people are realizing. … If you don’t have a good foundation, there’s not much you can build on top of that,” he said.

There are also governance and regulatory issues that firms must contend with. Slootman said determining who can access what data also needs to be translated into the world of LLMs too. 

This is where being “extremely organized in your data” will have a premium. 

Christian Kleinerman, SVP of product at Snowflake, added that the result of traditional machine learning or generative AI must be a function of having the right data, quality, and metrics. 

“The technology will be as good as the data that is fed in, so all the investments we make on data quality and cleansing and pipelines—all of that is very important,” he said. 

On top of that, it is equally important to measure and get feedback to determine how good the solutions are and if there are potential biases in the data or gaps in understanding the model’s performance. 

During the quarter, Snowflake’s product revenue grew 37% year-on-year to $640 million. Customers that are data sharing also grew 20% against the same period a year ago. 

Meanwhile, Snowflake’s Snowpark, which allows developers to write code in their preferred language and run it directly on Snowflake, had more than 400 customers. Slootman said 63% of Snowflake’s global 2,000 customers are using Snowpark on a weekly basis. 

Snowpark exposes new interfaces for development in Python, Scala, or Java to supplement Snowflake’s original SQL interface. 

During the Snowflake Summit 2023 in June, Snowflake launched Document AI and Snowpark Container Services, both of which are available for private preview.

Document AI uses a “leading LLM” to help customers extract information from documents. “With Document AI, customers can use natural language to ask questions of unstructured data. Legal contracts or invoices are now available for inquiry and analytics. This is an early example of how language models are expanding our opportunity,” Slootman said. 

Meanwhile, Snowpark Container Services helps firms bring full-stack applications, LLMs, and other data products to the data securely.