Data Warehouses (Snowflake, BigQuery…) as AI training data
A modern data warehouse is mostly code and conventions: dbt models turning raw tables into metrics, analysts' SQL, BI dashboards and tests that catch broken pipelines. That layer shows how a business defines revenue, churn or margin, and how analysts answer questions. Much of it carries little personal data, which can make it a practical place to start a data partnership.
Can Data Warehouses (Snowflake, BigQuery…) be used as AI training data?
Data warehouses on Snowflake, BigQuery or Redshift may be a practical place to start licensing AI training data, since much of the value is code with little personal data. dbt projects, semantic layer metric definitions and BI dashboards with their SQL are the strongest assets. Query history is stripped of user names and literal values, and third-party datasets are excluded.
No files needed upfront. We sign an NDA before reviewing any sample.
What can be included
You choose the scope. These are the parts of Data Warehouses (Snowflake, BigQuery…) that are typically most useful.
- ✓dbt projects: models, sources, tests, macros, documentation and exposures
- ✓Query history from Snowflake ACCOUNT_USAGE, BigQuery INFORMATION_SCHEMA.JOBS or Redshift system tables
- ✓Semantic layer and metric definitions in LookML, MetricFlow or Cube
- ✓BI dashboards with their underlying SQL from Looker, Tableau, Power BI or Metabase
- ✓Orchestration DAGs from Airflow or Dagster, with run and failure history
- ✓Data quality test results and the incident notes written when pipelines broke
Why AI labs value it
From question to answer
A dashboard tile, its SQL and the metric definition behind it link a business question to a working query. Analytics agents need many such pairs from real companies.
Lineage that shows intent
dbt lineage shows how raw source tables are cleaned, joined and aggregated into business entities. It records modelling choices that are invisible in the final tables.
Debugging with outcomes
Failed tests, broken DAG runs and the commits that fixed them form a problem-to-fix trail with a clear pass or fail result.
How it is scoped, exported and anonymized
Start with code and metadata
The dbt repository, LookML and DAG code can usually be assessed first. They describe your warehouse without moving any table data.
Clean query history
Query logs include user names and literal filter values, such as customer IDs or emails. These are stripped or parameterised before further use.
Exclude third-party datasets
Tables loaded from purchased datasets, data marketplaces or partner feeds usually come with their own licence terms. They are excluded unless you have the right to include them.
How a partnership begins
Share Basic Information
Share only your contact, company, industry, and partnership interests—no raw data required.
Initial Value Assessment
We discuss the data type, completeness, scarcity, and the tasks it could support.
Define the NDA and Boundaries
Before reviewing any sample, we define confidentiality, permitted use, retention, and security requirements.
Receive an Offer and Begin the Partnership
We provide a purchase proposal and contract, then support de-identification and secure data transfer.
Frequently asked questions
- Is a dbt project worth anything without the data?
- It can be. A mature dbt project with tests and documentation shows how a business models its data, which is useful on its own.
- Does it matter which warehouse we use?
- Not much. Snowflake, BigQuery, Redshift and Databricks projects can all qualify. Dialect differences can add value rather than reduce it.
- Can warehouse tables be shared directly?
- If table data is in scope, the transfer method is agreed in the contract after the NDA. Options such as a secure share or a Parquet export to storage are discussed at that point.
- Our warehouse contains CRM and product event data. Is that a problem?
- It means those sources need the same privacy review they would get on their own. Many companies exclude raw event tables and include only modelled, aggregated layers.
Related
Let's talk about your Data Warehouses (Snowflake, BigQuery…).
Share only the essential details. No files are required at this stage.
Contact the Momentra team