Guides · Published
How to sell your company’s data to AI companies
You license a cleaned copy of the data your team already has, like support tickets, docs, code and CRM notes, to an AI lab or a data buyer. You keep the original, customer identities come out first, and reported deals range from about $10,000 to millions. Here’s what sells, what it’s worth, the five ways to sell, and what to check before you say yes.
Why AI companies are buying company data now
AI labs have trained on most of the public internet. What they can’t find online is how real businesses operate: how a support team resolves a tricky ticket, how engineers change a codebase over years, how a sales team handles objections. That’s the data they need to train AI agents that do multi-step work.
Buyers are building what the field calls reinforcement learning gyms: simulated workplaces stocked with a real company’s Slack threads, tickets and documents, where an agent practices doing the job (Gizmodo). Vendors that supply the labs now approach companies directly. Mercor emailed the CEO of Warmly about buying or licensing its code base and task records eight days after Warmly agreed to be acquired (AI Weekly).
SimpleClosure, which helps startups shut down and sell what they leave behind, has seen the demand firsthand. Its CEO described it to Forbes:
“There’s a feeling of a gold rush from these companies trying to get their hands on real-world data.”
What company data sells
Labs pay for records of work, not contact lists. The most requested kinds:
- Support tickets from Zendesk, Intercom or Freshdesk: real problems and how your team solved them.
- Source code from GitHub, GitLab or Bitbucket, with the commit history that explains it.
- Docs and SOPs from Notion, Confluence or Google Drive: runbooks, onboarding guides, how-tos.
- CRM and sales notes from HubSpot, Salesforce or Pipedrive, with every contact removed.
- Chat and email from Slack, Teams or Gmail: how decisions actually get made.
- Ops data and logs paired with the notes that explain what happened.
Years of history, a specialized industry and well-documented outcomes all raise the price. A pile of unlabeled files lowers it.
What it’s worth: deals that have been reported
- The CEO of Cielo24, a software company that shut down, sold every Slack message, internal email and Jira ticket as training data for “hundreds of thousands of dollars” (Gizmodo).
- SimpleClosure, which helps startups shut down, closed about 100 data deals in a year at $10,000 to $100,000 per company. It says its AssetHub marketplace is on track to return more than $10 million to startups by the end of 2026 (SimpleClosure).
- Google won a bankruptcy auction for Spirit Airlines’ internal business data with a $10 million bid (Frankfurt Kurnit).
- At the top end, Reddit’s content deal with Google was reported at about $60 million a year (The Decoder).
For an operating company, licenses typically land between $20K and $5M depending on volume, history, specialization and what labs are buying that quarter. Our 60-second estimate gives you a range for yours.
Five ways to sell your company’s data
Each route suits a different company. In no particular order:
| Route | Best for | What you do | Watch out for |
|---|---|---|---|
| Direct deal with an AI lab | Very large or one-of-a-kind archives | Find the buyer, negotiate, clean and deliver the data yourself | Labs rarely take inbound pitches from mid-size companies |
| Wind-down marketplace (e.g. SimpleClosure AssetHub) | Startups closing, pivoting or retiring a product | List source code or workspace exports; buyers bid | Built around closures; workspace data was still in beta at launch |
| Data vendor that buys outright (e.g. Mercor) | Companies a vendor already wants to approach | Answer the outreach and negotiate one buyer’s offer | One bidder sets the price; you may not know who trains on it |
| Dataset marketplace (e.g. AWS Data Exchange, Snowflake Marketplace, Opendatabay) | Structured, repeatable datasets you can refresh | Package, document and price the dataset; handle support | Built for tables and feeds, not Slack, tickets or code history |
| Licensing partner (e.g. Smelted) | Operating companies with years of tickets, docs, code or CRM notes | Pick what’s in scope and approve the buyer and terms | The partner is paid from the deal; read the fee and exclusivity terms |
SimpleClosure’s AssetHub launched with source code live and workspace data (email, Slack, Notion, Google Drive) in beta, and has since opened to companies that aren’t closing (Business Wire, SimpleClosure).
Before you say yes: five rights to check
Holding data is not the same as having the right to sell it. A law firm’s review of these deals found that most companies can’t sell everything a buyer asks for (Frankfurt Kurnit). Check:
- Other companies’ secrets. Email and chat hold customers’ and partners’ confidential information, often covered by NDAs and contracts.
- Code and content you don’t own. Licensed data, open source and vendor code, and files attached by third parties can carry their own restrictions.
- How the data leaves each tool. Slack’s API terms, for example, bar third-party apps from using API data to train large language models, so the export method matters.
- Personal data. Customer and employee identities have to come out before any buyer sees a sample, under the privacy laws where those people live.
- People with a stake. Spirit’s sale drew objections from its flight attendants’ union and from a software vendor that says some of the data is its own (Frankfurt Kurnit).
How a sale works, step by step
- Inventory. List your tools and how many years of history each holds.
- Scope. Pick systems, folders and date ranges. Exclude sensitive customers, projects and competitors as buyers.
- Clean. Remove identities and secrets, then review whole records, because context alone can identify someone.
- Package. Document what’s in the dataset so buyers can judge quality from redacted samples.
- Choose the buyer. Approve who trains on it, for what use, and when raw data must be deleted.
- Sign and get paid. Delivery follows signature; with Smelted, payment comes in as little as 7 days.
Questions founders ask
Can a business legally sell its data to AI companies?
Often, but rarely all of it. You can license data you own and have the right to share. Customer personal data has to be removed first, and email or chat often contains other companies’ confidential information covered by NDAs. Some tools’ terms also limit how exported data can be used, so check how the data leaves each system.
How much do AI companies pay for company data?
Reported deals run from about $10,000 for a small startup’s archive to millions for large, unique datasets. SimpleClosure reported roughly 100 deals at $10,000 to $100,000 each; Google bid $10 million for Spirit Airlines’ internal data. Volume, history, specialization and how clean the data is drive the price.
What kind of company data do AI labs want?
Records of how real work gets done: support tickets and their resolutions, internal docs and SOPs, source code with its history, CRM and sales notes, and team chat and email. Labs use it to train AI agents to do multi-step work, not to learn who your customers are.
Do I have to be shutting down to sell my data?
No. The first wave of deals came from startups winding down, but operating companies license copies of their data too. You keep the original, choose what’s included and can exclude competitors as buyers.
Will selling data expose my customers?
It shouldn’t, if identities are removed before any buyer sees a sample. That means names, contact details, account numbers and secrets in code, plus a review of whole records, since context alone can identify someone. Anything that can’t be cleaned should stay out.