How Tipalti Uses Snowflake Cortex AI to Democratize Data Insights

Tipalti sits at the intersection of finance and scale, automating payables for companies that collectively move $75 billion per year. When your customers depend on you to keep their payments running, data reliability and access speed aren’t operational preferences—they’re existential requirements. The company’s data team understood early that spreading the value of that data across the organization meant more than building dashboards for analysts. It meant giving sales reps and marketers the ability to ask questions themselves.

Before deploying Cortex AI, Tipalti’s non-technical teams faced a familiar barrier: they could see the data existed but couldn’t reach it without writing SQL or waiting for data engineering support. The company’s sales and marketing teams were particularly affected—they needed timely, consistent insights to evaluate opportunities and coordinate campaigns, but the gap between intent and insight created delays and inconsistencies between teams.

Tipalti’s engineering team built what they called an “AI prompt store” on Snowflake’s Cortex AI platform. This library of pre-defined LLM prompts lets non-technical users query data directly from the tools they already use, including Salesforce. Under the hood, the solution uses Cortex AI’s AI_COMPLETE function to execute LLM queries within SQL workflows, the response_format feature to eliminate hallucinations and enforce structured outputs, and TRY_COMPLETE to prevent single failures from collapsing batch processes. A custom Snowpark function enables multiple queries to run in parallel, multiplying throughput fivefold.

Since the system went live, over 50 users have run more than 600 ad-hoc analyses and accessed ten or more new features—all without data team involvement. Teams describe the experience as logging in and simply getting answers. The parallel execution architecture, built entirely within Snowflake without external inference infrastructure, has made the system faster, cheaper, and easier to maintain than equivalent third-party solutions would have been.

Tipalti’s next phase involves evolving the framework into a fully autonomous AI assistant layer using Cortex Semantic Models, RAG via Cortex Search, and improved embedding-based retrieval. For a company whose product promise is to simplify financial operations for businesses of every size, building the same simplicity into its own internal data systems is a logical extension of the mission.