WHOOP is a human performance company on a mission to unlock human performance and healthspan, with an ambitious goal of adding 1 billion healthy years to humanity. Founded in 2012 as a tool for elite athletes, WHOOP now helps millions of people better understand their sleep, recovery, activity, and overall health, while expanding into areas such as health records and blood biomarkers.
Behind those experiences is an enormous variety of data. WHOOP’s centralized analytics team works across marketing, product, growth, and other functions, helping teams understand how to reach members, which features they use, and how those experiences affect retention. dbt and Snowflake provide the primary data infrastructure supporting that work.
WHOOP wanted to make trusted data easier to use across the business, including through AI. But when the team began connecting agents to its well-governed dbt tables, agents still hallucinated, metrics didn't always align, and analysts spent time reconciling conflicting answers. “One bad metric, one confused definition, one poorly targeted table doesn't produce just one wrong answer. It can quickly cascade to thousands,” says William Tsu, Staff Analytics Engineer at WHOOP. WHOOP needed a governed way to give agents the same business definitions and relationships its analysts relied on.
The team built an open-source tool that connects its dbt models to Snowflake semantic views, bringing consistent definitions, relationships, metrics, and business context into the same development workflow.
Bridging governed data and business context for AI
WHOOP chose Snowflake semantic views to provide the business context its AI agents needed, but much of the structure and metadata behind that context already lived in dbt. No YAML-based bridge existed between the two, so teams were left translating definitions by hand, for example with the dbt semantic view package, or bringing in an external tool like the dbt MCP server.
“If you're a data team mainly using dbt for data curation, you couldn't readily leverage that work for Snowflake AI readiness,” Tsu explains. “We realized there needed to be a more delightful way for developers to build Snowflake semantic views.”
Rather than wait for industry standards to converge, WHOOP built Snowflake Semantic Tools, an open-source Python package that connects dbt artifacts with Snowflake semantic views. The initial version took just a few weeks and a fraction of a full-time employee to build with help from AI coding agents. The team also codified its development process into reusable workflows and began using agents to automate some of the steps, making it easier for analysts to build semantic views consistently without becoming experts in every underlying tool.
WHOOP designed Snowflake Semantic Tools so semantic definitions live alongside dbt models in the same repository, branch, and pull request. As the team demonstrated, model changes and their semantic definitions can be validated through local development and CI, then deployed together. The tool can also generate metadata from Snowflake and run more than 100 validation checks, reducing the risk that metrics, relationships, and definitions drift out of sync with the underlying data.
In a demonstration at dbt Summit 2026, the team asked which customers generated the most revenue. Instead of manually identifying the correct revenue field and joining tables, the analyst queried the semantic view using named metrics and dimensions. The view returned the same result as the original SQL query, with the relationships and definitions already governed — reducing the manual work needed to produce a consistent answer.
By making business context part of the development workflow, WHOOP can keep definitions, relationships, and metrics consistent for both people and AI without maintaining a separate layer by hand.
Adoption across 5 teams, and weeks of analysis in 30 minutes
WHOOP is already using the semantic layer across the business:
- Deployed more than 80 semantic views across 5 teams, with more than 50 users querying them.
- Processed roughly 1,800 queries through semantic views in 4 months, expanding governed access to business data.
- Reduced some analytical workflows from days or weeks to approximately 30 minutes, with semantic views handling joins, metric definitions, and data discovery that analysts previously had to resolve manually.
- Made deployed semantic views available to Snowflake Cortex Analyst, allowing AI agents to query governed business metrics without the data team recreating definitions for each new consumer.
“Once you build a semantic layer, it compounds,” says Madhura Pharande, Analytics Engineer at WHOOP. “Every new consumer, whether human or AI, gets a governed answer without additional work from your data team.”
As adoption grows, WHOOP can extend trusted analytics to more people and AI agents without repeatedly rebuilding business definitions. Stakeholders get answers faster, while analysts spend less time reconciling metrics and more time helping product, marketing, and growth teams make decisions that improve the member experience.
Making context part of the data product
WHOOP's experience shows that well-governed data models are an important foundation for AI, but agents also need consistent definitions, relationships, and business context to interpret that data correctly. The team's solution was to make semantic metadata part of the same development workflow as its dbt models, rather than maintaining it separately.
That approach is already supporting more than 80 semantic views across 5 teams, while reducing the manual work analysts spend on joins, metric definitions, and data discovery. WHOOP is continuing to expand its use of Snowflake's AI tools and evolve its semantic layer as standards develop.
For data teams facing similar challenges, WHOOP's lesson is practical: govern the meaning of your data alongside the data itself. When models and semantic definitions evolve together, people and AI can reuse the same trusted business context without requiring analysts to rebuild it for every new question or application.
