Home / Blog / How to license your company's data to AI companies, step by step

How to license your company's data to AI companies, step by step

To license your company's data to an AI company, list what you have and how many years it covers, confirm the company has the right to license it, decide what stays out, share a small cleaned sample under NDA, agree a licence that sets field of use, exclusivity and term, then remove secrets and personal data before delivering a copy. You keep ownership; the buyer gets the right to train and evaluate models on the copy.

AI labs are building agents that write code, answer support tickets, update CRMs and work across a company's tools. To train and test those agents they need examples of how real companies actually work, and almost none of that is on the public web. It sits in private repositories, Slack workspaces, help desks and shared drives. This guide walks through how a company licenses that material, from first inventory to delivery. If your company has already shut down, read this alongside selling a shut-down startup's code and data.

Step 1: take an inventory

Buyers start with simple questions: what systems, how many years, how many people. A rough list is enough at this stage.

SystemWhat to note
Code (GitHub, GitLab, Bitbucket)Number of repositories, years of history, contributors, whether any of it was ever public
Chat (Slack, Teams)Years covered, channels, and what your plan lets you export
EmailShared inboxes and team mailboxes, years covered
Support (Zendesk, Intercom, Jira)Ticket volume, whether tickets link to code changes or chat
CRM (Salesforce, HubSpot)Deal and activity history
Docs and wikis (Notion, Confluence, Google Drive)Specs, runbooks, SOPs, version history
FinanceAccounting exports, invoices, budgets, board packs

Records that link across systems are the most useful: a ticket that points to a Slack thread, a pull request and an invoice shows how work really moves through a company. See business operations data for AI agents and private codebases for AI training for what buyers look for in each.

Step 2: confirm you have the right to license it

Owning the systems is not the same as having the right to license everything inside them. Before talking terms, check:

Step 3: decide what stays out

It is easier to agree scope up front than to pull records back later. Common exclusions:

Many companies also leave out direct messages and private channels, or limit the licence to specific teams and date ranges.

Step 4: share a small sample under NDA

Buyers decide whether the data fits their work from a sample, not a description. Keep it small and representative, and clean it to the same standard as the final delivery, so the buyer sees exactly what they would get. Never send a raw export before an NDA is signed.

Step 5: agree the licence

Most deals are licences of a cleaned copy, not a sale of the IP. These are the terms that matter:

TermWhat to decide
Field of useTraining, evaluation, fine-tuning, or all of them
ExclusivityWhether you can license the same copy to anyone else, and for how long
Term and territoryHow long the buyer can use it, and where
Re-identificationAn explicit ban on trying to identify people or companies in the data
ConfidentialityWhether the buyer can name you as a source
RemovalHow a record is taken out if a problem is found later
End of licenceWhether the data must be deleted, and what happens to models already trained
WarrantiesWhat you promise about rights and content, and the limits on that promise

Our training data rights checklist shows the questions buyers' legal teams will ask, which helps you prepare answers before they come up.

Step 6: clean the data and deliver it

Step 7: keep a record

Keep a copy of the signed licence, a list of exactly what was delivered, and a log of the cleanup steps. If a question comes up later from an employee, a customer or a regulator, you can answer it with evidence. Our post on data provenance shows a simple way to do this per record.

Mistakes to avoid

Talk it through with us

Datadhara takes your code and records to AI labs and data partners, brings you the best offer, and handles the cleanup of secrets and personal data. You keep ownership, and you don't pay us: the buyer or platform does. If you're weighing this up, book 15 minutes and we'll tell you plainly whether there's a fit.

This guide is general information, not legal or tax advice.

Questions

Can a company sell its internal data to AI companies?

Often, yes. Most deals are licences: the company keeps ownership and grants an AI lab the right to train and evaluate models on a cleaned copy, after checking that customer contracts and privacy notices allow it.

What company data do AI labs want?

Private codebases with real git history, and records of how the business ran: Slack or Teams, email, support tickets, CRM, docs, SOPs and financials, ideally linked across systems over several years.

Do I have to share personal data?

No. Names and contact details are replaced with pseudonyms, sensitive personal data is removed, and secrets are scanned out of the full history before anything is delivered.

How long does licensing company data take?

Usually a few weeks from first call to signed licence, depending on how quickly a sample can be prepared and approvals obtained.

Find out what your data could be worth

Tell us roughly what you have: repos, years of history, the systems you used. We'll give you a straight answer in 15 minutes.

Book 15 minutes →or tell us about your codebase