# Can a SaaS Company License Its Database and Warehouse?

> AI buyers ask software companies for warehouse tables, dbt models and pipeline code with run history. Who qualifies and what customer contracts allow.

Canonical: https://origindatapartners.com/guides/license-saas-warehouse-data-ai
By: Origin Data Partners
Status: Published
Published: 2026-11-15
Updated: 2026-11-15
Sources checked: 2026-11-15

## Quick answer

Yes for the company's own operating data, and only with customer agreement for data it holds on customers' behalf. AI buyers ask software companies for production and warehouse tables together with the modelling and pipeline code that explains them, plus finance, CRM, billing and support records. Content your customers put into the product is theirs: under most contracts and data processing agreements you may use it only to provide the service. Personal information has to be removed or de-identified first.

## Key points

- Split the warehouse into your own operating data and customer content before anything else.
- For customer content you are usually a processor or service provider and may act only on the customer's instructions.
- Hashed or tokenized identifiers are still personal data under European law.
- Query and job history ages out of the warehouse. Export it before it does.

## Can a SaaS company license its database and warehouse to AI?

Yes, in part. A SaaS warehouse mixes two kinds of data that have different answers. The first is the record of your own business: the ledger, billing, the sales pipeline, the support desk and the code that models them. The second is what your customers put into your product. The first is yours to license once personal information is dealt with. The second is not, unless the customer agrees.

The table sorts a typical stack. This page is general information, not legal advice. Your own customer contracts decide.

| Data | Whose it is | Can it be licensed? |
| --- | --- | --- |
| General ledger, billing, subscriptions and invoices | Yours | Yes, after personal information and customer names are removed |
| CRM pipeline and support desk records | Yours, with other people's personal information in them | Yes, after de-identification |
| dbt or SQLMesh project, pipeline code, run history, query logs | Yours | Yes. Scan for secrets and for customer names in code and comments |
| Product usage and event tables | Depends on how the contract defines usage data | Often, in aggregated or de-identified form. Read the clause |
| Customer content: what customers upload, enter or create in the product | The customer's. You process it for them | Only with the customer's permission |
| Bought-in enrichment data and payment processor data | The vendor's, under its terms | Usually no |


## What do AI buyers ask a SaaS company for?

A working data stack, with the code that explains it. According to one buyer's public listings, read 15 November 2026, a buyer looks for "the company's own production databases and a cloud warehouse", "dbt or SQLMesh with 100+ models, or 50+ scheduled pipelines with run history", "a general ledger, CRM, billing and support desk in daily use" and "50 GB or more of tabular data". The licence covers database and warehouse tables, modelling and pipeline code with run history, and finance, CRM, billing and support records.

The modelling layer is what turns tables into something a model can learn from. In dbt, "models are primarily written as a select statement and saved as a .sql file" ([dbt documentation](https://docs.getdbt.com/docs/build/models)). In SQLMesh, "models are made up of metadata and queries that create tables and views" ([SQLMesh documentation](https://sqlmesh.readthedocs.io/en/stable/concepts/models/overview/)). A hundred models are a hundred written-down decisions about how raw events become revenue, churn or a support backlog. Run history adds what happened when those decisions met real data: which runs failed, how long they took, what was fixed.

The ledger, CRM, billing system and support desk matter for the same reason. They are separate systems that describe the same customers, and the joins between them are the hard part of real analytical work.


## Which software companies qualify?

The published fit is a company of 50 to 1,000 employees with a data team and three or more years of operating data, in any country (one buyer's public listings, read 15 November 2026). The kinds of company named are B2B SaaS, consumer subscriptions, marketplaces and e-commerce.

Two things are listed as not a fit: agencies holding their clients' databases, and empty or barely used CRMs. The first is a rights problem, because the data belongs to the clients. The second is a depth problem.

A company with source code but no warehouse, or with transaction tables and no modelling layer, fits other record types better. See [who we help: software companies](https://origindatapartners.com/who-we-help/software-companies) and the [guide to ERP and accounting records](https://origindatapartners.com/guides/license-erp-accounting-records-ai).


## What do customer contracts and DPAs allow?

For customer content, very little beyond running the service. Under European law a processor's contract must say that it "processes the personal data only on documented instructions from the controller". A processor that goes further is exposed: "if a processor infringes this Regulation by determining the purposes and means of processing, the processor shall be considered to be a controller in respect of that processing" ([GDPR, Article 28](https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng)). Licensing customers' personal data to a third party is a new purpose the customer did not instruct.

California uses the same idea. A "service provider" is one that processes personal information for a business under a contract that prohibits "selling or sharing the personal information" and "retaining, using, or disclosing the personal information for any purpose other than for the business purposes specified in the contract" ([Cal. Civ. Code 1798.140](https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=CIV&sectionNum=1798.140)). Most data processing agreements (DPAs) repeat those limits.

Contracts also protect customer content that is not personal data at all. The confidentiality clause in a master subscription agreement usually covers everything the customer puts into the product.

| Clause to read | What to look for |
| --- | --- |
| Definition of customer data | Usually everything the customer or its users submit. That is the part you do not control |
| Usage data or service data | Many agreements let the provider use data about how the service is used. Check whether it must be aggregated or de-identified |
| Aggregated or anonymized data | Some agreements let the provider use such data for any business purpose. Others limit it to improving the service |
| Confidentiality | Covers customer content even where privacy law does not apply |
| DPA purposes and sub-processors | The list of purposes is closed. A data buyer is not a sub-processor doing your customer's work |
| Negotiated enterprise terms | Large customers often struck the usage-data clause. Keep an exceptions list by customer |


## What about personal information in the tables and logs?

Your own operating data is full of other people's details: billing contacts, sales leads, the users who wrote support tickets. Two rules apply even though the data is yours. In Europe, personal data must be "collected for specified, explicit and legitimate purposes and not further processed in a manner that is incompatible with those purposes" ([GDPR, Article 5](https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng)). In California, to "sell" covers making a consumer's personal information available "to a third party for monetary or other valuable consideration" ([Cal. Civ. Code 1798.140](https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=CIV&sectionNum=1798.140)), and the state's privacy agency says the law protects residents who are "employees or job applicants, and contacts for business customers, vendors, or independent contractors" ([CPPA](https://cppa.ca.gov/faq.html)). A licence of identifiable records for money can meet that definition.

The way through is de-identification that holds up. California defines "deidentified" as information that "cannot reasonably be used to infer information about, or otherwise be linked to, a particular consumer", and it requires the business to take reasonable measures and commit not to reidentify. European law is stricter about half-measures: data "which could be attributed to a natural person by the use of additional information should be considered to be information on an identifiable natural person" ([GDPR, Recital 26](https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng)). A hashed email or a customer ID that you can still look up is pseudonymous, not anonymous.

Three places are easy to miss: free-text columns such as ticket bodies and CRM notes, query logs that contain literal values typed into a WHERE clause, and seed files or test fixtures in the modelling repository. The [guide to GDPR and CCPA in data licensing](https://origindatapartners.com/guides/data-licensing-gdpr-ccpa) and the [de-identification guide](https://origindatapartners.com/guides/de-identify-company-data-before-licensing) go further.


## What is a SaaS data stack worth to an AI buyer?

The one published range we can cite is $55,000 to $350,000 for the company, which the listing says "depends on data volume and stack depth" (one buyer's public listings, read 15 November 2026). Read it as a ceiling for a mature stack, not a typical result. It is what one buyer advertises, not what any company has been paid.

A licence to an AI buyer is a different thing from listing a dataset for many subscribers. The [comparison of data marketplaces](https://origindatapartners.com/guides/data-marketplaces-compared) covers that route.

| Raises the value | Lowers the value |
| --- | --- |
| Ingestion configs and query logs | Tables with no code that explains them |
| A semantic layer or LookML | A modelling project under the published thresholds |
| Dashboards built on the warehouse | A CRM or support desk that is barely used |
| Three or more years of history in linked systems | Systems that cannot be joined on a common key |
| Run history for models and pipelines | Customer content that cannot be separated from your own data |

The first three rows of the left column are from the listing's own list of what adds to what the company earns. The rest is our reading of its fit criteria.


## How do you count what is in the stack?

A buyer asks which warehouse you use, how many dbt models or pipelines, roughly how much data in GB, and how long the data team has been in place. Each has a place to look it up. Nothing has to leave the warehouse to answer.

Check retention first. Snowflake's query history view covers queries "within the last 365 days (1 year)" ([Snowflake documentation](https://docs.snowflake.com/en/sql-reference/account-usage/query_history)). BigQuery's jobs view "displays running jobs along with job history for the past 180 days" ([Google Cloud documentation](https://cloud.google.com/bigquery/docs/information-schema-jobs)). If query logs are part of the offer, copy them into a table of your own on a schedule, or the oldest months are gone before a deal is signed.

| What to count | Where to look | Note |
| --- | --- | --- |
| dbt models | The list command: "the dbt ls command lists resources in your dbt project" ([dbt ls](https://docs.getdbt.com/reference/commands/list)) | Filter by resource type to count models only |
| Everything in the dbt project | manifest.json, "a full representation of your dbt project's resources (models, tests, macros, and more)" ([dbt manifest](https://docs.getdbt.com/reference/artifacts/manifest-json)) | One file per project state |
| Model run history | run_results.json, which holds "information about a completed invocation of dbt, including timing and status info for each node" ([dbt run results](https://docs.getdbt.com/reference/artifacts/run-results-json)) | One file per invocation. History exists only if the files or the orchestrator's records were kept |
| Scheduled pipelines and their runs | The orchestrator. In Airflow, "a Dag Run is an object representing an instantiation of the Dag in time" ([Airflow](https://airflow.apache.org/docs/apache-airflow/stable/core-concepts/dag-run.html)) | Count pipelines, then runs per pipeline |
| Data volume in Snowflake | The view that "displays table-level storage utilization information" ([TABLE_STORAGE_METRICS](https://docs.snowflake.com/en/sql-reference/account-usage/table_storage_metrics)) | Sum active bytes by schema |
| Data volume in BigQuery | The view that "provides a current snapshot of storage usage for tables and materialized views" ([TABLE_STORAGE](https://cloud.google.com/bigquery/docs/information-schema-table-storage)) | Sum logical bytes by dataset |


## Example: inventory of a 120-person subscription company

A fictional B2B subscription company has run Snowflake and dbt for four years. Its head of data builds a one-page inventory in an afternoon. The figures are invented to show the method.

| Item | Count | Rights position | Decision |
| --- | --- | --- | --- |
| dbt models | 340 | Company code | In, after a scan for secrets and customer names |
| dbt run results kept | 26 months | Company records | In |
| Airflow pipelines | 62, with run history | Company records | In |
| Finance and billing schemas | 45 GB | Company data with customer names | In, with account names replaced by stable codes |
| CRM and support schemas | 70 GB | Company data with contacts' personal information | In, after de-identification of contacts and ticket text |
| Product event tables | 600 GB | Usage data; clause allows de-identified use for most customers | In for standard-terms customers; 14 enterprise accounts on the exceptions list are filtered out |
| Customer content schema | 1.1 TB | Customer data under the DPA | Out |
| Enrichment vendor tables | 8 GB | Vendor licence | Out |

Fictional example. Most of the bytes are in the schema that stays out. What the company describes is the modelling project, its run history and about 700 GB of its own operating and usage data.


## What does a licence usually allow and forbid, and how long does it take?

A licence for a data stack is normally non-exclusive and limited to model training and evaluation by a named party. It usually forbids resale of the data, any attempt to reidentify people or customers, and use of the data to contact anyone in it. It should say where the data is stored, who may access it and what is deleted at the end of the term. The [data licensing agreement guide](https://origindatapartners.com/guides/data-licensing-agreement) lists the clauses to expect.

No public source gives a reliable timetable for a data licence, so this page does not quote one. The order of work is fixed: split your own data from customer content, read the contracts and build the exceptions list, de-identify a sample, agree terms, then deliver. The contract review is the slow step, and legal and the data team can finish it before any buyer is involved.


## What if the product is being sold or shut down?

Do the inventory before the accounts close. A warehouse is rented. When the subscription ends, the tables, the query history and the orchestrator's run records go with it, and the modelling repository is only half the asset without them.

Customer contracts do not end with the product. Deletion and return clauses in the DPA usually apply at termination, so customer content has to be returned or deleted on the contract's terms. Your own operating data is what remains. The guides to [selling company data after closing](https://origindatapartners.com/guides/sell-company-data-after-closing) and [shutting down a startup](https://origindatapartners.com/guides/how-to-shut-down-a-startup) cover the order of steps.


## Common mistakes software companies make

Most of them come from treating the warehouse as one thing.

- Counting customer content toward the 50 GB when it cannot be licensed.
- Reading the standard terms and forgetting the enterprise customers who negotiated their own.
- Treating hashed emails and internal IDs as anonymous.
- Leaving personal information in ticket text, CRM notes and query logs.
- Shipping a modelling repository with credentials, API keys or customer names in it.
- Letting query and job history age out before it is copied.
- Including tables bought from an enrichment vendor.
- Cancelling the warehouse during a shutdown before the inventory is done.


## A SaaS data stack as a data asset

A warehouse with a few hundred maintained models, years of run history and the finance, sales and support systems behind it is a record of how a real company measures itself. That is what AI buyers ask software and subscription companies for. The part you can license is your own operating data and code, not your customers' content.

Origin works with buyers and helps a company describe that part, find out what it could be worth and decide whether to go further. Nothing is exported at the start. If your CRM is the largest system, the [Salesforce export guide](https://origindatapartners.com/guides/export-salesforce-data) shows what an export contains. Origin Data Partners is an independent data partner that helps companies get their data bought. A buyer may pay Origin when a deal closes.

Does your company hold records like these? [See what your data is worth](https://origindatapartners.com/check-fit).


## Questions and answers

### Can a SaaS company license its database for AI training?

It can license its own operating data and code: ledger, billing, CRM, support records, warehouse models and pipeline history, once personal information is removed. Content that customers put into the product belongs to them and needs their permission.

### Do SaaS customer contracts allow licensing customer data to AI buyers?

Usually not. A processor under GDPR Article 28 acts only on the customer's documented instructions, and a California service provider may not use the data outside the contract's business purposes. Some agreements allow use of aggregated or de-identified usage data. Read the clause for each customer.

### Why do AI buyers want dbt models and pipeline run history?

Because the code explains the tables. Each model is a written decision about how raw data becomes a business measure, and run history shows what failed and how it was fixed. Tables alone do not show that reasoning.

### How much warehouse data is enough for an AI data licence?

One buyer's public listing, read 15 November 2026, asks for 50 GB or more of tabular data, 100 or more dbt or SQLMesh models or 50 or more scheduled pipelines with run history, and three or more years of operating data.

### Is hashed or pseudonymized customer data safe to license?

Not on that basis alone. Under GDPR Recital 26, data that can be attributed to a person with additional information is still personal data. California's deidentified standard needs data that cannot reasonably be linked to a consumer, plus a public commitment not to reidentify.

### What happens to warehouse data when a SaaS company shuts down?

It is deleted when the warehouse account closes, along with query history and pipeline run records. Customer content has to be returned or deleted under the DPA. Inventory and export your own operating data before the accounts are cancelled.

## Sources

- [Regulation (EU) 2016/679 (GDPR): Recital 26 and Articles 4, 5 and 28](https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng)
- [California Civil Code 1798.140: CCPA definitions (deidentified, sell, service provider)](https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=CIV&sectionNum=1798.140)
- [California Privacy Protection Agency: frequently asked questions](https://cppa.ca.gov/faq.html)
- [dbt documentation: About dbt models](https://docs.getdbt.com/docs/build/models)
- [dbt documentation: About dbt ls (list) command](https://docs.getdbt.com/reference/commands/list)
- [dbt documentation: Manifest JSON file](https://docs.getdbt.com/reference/artifacts/manifest-json)
- [dbt documentation: Run results JSON file](https://docs.getdbt.com/reference/artifacts/run-results-json)
- [SQLMesh documentation: Models overview](https://sqlmesh.readthedocs.io/en/stable/concepts/models/overview/)
- [Apache Airflow documentation: Dag Runs](https://airflow.apache.org/docs/apache-airflow/stable/core-concepts/dag-run.html)
- [Snowflake documentation: QUERY_HISTORY view (Account Usage)](https://docs.snowflake.com/en/sql-reference/account-usage/query_history)
- [Snowflake documentation: TABLE_STORAGE_METRICS view (Account Usage)](https://docs.snowflake.com/en/sql-reference/account-usage/table_storage_metrics)
- [Google Cloud documentation: BigQuery INFORMATION_SCHEMA.JOBS view](https://cloud.google.com/bigquery/docs/information-schema-jobs)
- [Google Cloud documentation: BigQuery INFORMATION_SCHEMA.TABLE_STORAGE view](https://cloud.google.com/bigquery/docs/information-schema-table-storage)

## Related reading

- [/who-we-help/software-companies](https://origindatapartners.com/who-we-help/software-companies)
- [Data marketplaces compared: Snowflake, Databricks, AWS](https://origindatapartners.com/guides/data-marketplaces-compared)
- [License ERP and Accounting Records for AI Training](https://origindatapartners.com/guides/license-erp-accounting-records-ai)
- [How to Export Salesforce Data: Every Method, Step by Step](https://origindatapartners.com/guides/export-salesforce-data)
- [GDPR, CCPA and licensing company data for AI](https://origindatapartners.com/guides/data-licensing-gdpr-ccpa)
- [Closing a software business: what could have value beyond the code?](https://origindatapartners.com/guides/sell-company-data-after-closing)
- [Does an NDA or Customer Contract Block Licensing Your Data?](https://origindatapartners.com/guides/nda-customer-contract-data-licensing)
- [Data Licensing as Non-dilutive Funding for Startups](https://origindatapartners.com/guides/non-dilutive-funding-data-licensing)
