How to license your company's data to AI companies
Some AI companies pay for company records. The only repeat public price is for startups that shut down.
Yes, some AI companies pay for company records, but the public proof is thin. The one repeat price is $10,000 to $100,000 per shut-down startup, per Gizmodo.
People search for how to sell company data. In practice you license it. A working company has no public price yet. Most deals run through a marketplace, after you confirm you hold the rights.
$10,000 to $100,000
What SimpleClosure's nearly 100 deals paid per shut-down startup
Public figure, see sourceThat range is for a company closing its doors. No raw data leaves your building until a contract says so.
The buyer numbers are self-reported, and this guide leans on public reporting, court filings and buyer pages. Check anything you plan to rely on.
What do AI companies want from a business?
Troveo, micro1 and Mercor list records of real work. A support ticket shows a problem, the steps taken and the result. A deal in your CRM (the system that tracks customers) shows the same arc.
Labs and agent builders use records like these to build practice workplaces for AI agents, sometimes called RL gyms. The agent works through real Slack threads, tickets and email (Gizmodo, Futurism).
Most marketplace pages keep buyers generic, as in "frontier AI lab." For the limits of this market, see what can go wrong.
What qualifies, and what stays out?
The companies list the same core material:
- micro1: written procedures, knowledge bases, CRM data and project histories.
- Troveo: chat, email, documents, tickets, CRM and finance systems.
Size and history matter too. micro1 asks for 30 or more employees, Handshake AI for 20 or more full-time employees and 3 or more years in operation, and Troveo states no minimum. The table shows where each says so:
| Company | Stated minimum | Where it says so |
|---|---|---|
| micro1 | 30 or more employees, mature operations, primarily English documentation, US prioritized | Data partnerships page and its company referral page |
| Handshake AI | 20 or more full-time employees and 3 or more years in operation | Referral page |
| Troveo | No minimum stated | Troveo |
Acceptance at micro1 depends on its review and current demand. Confirm exceptions directly with micro1.
What stays out is anything you lack the right to license. Your email and chat hold other companies' information, often under NDAs (confidentiality agreements).
Counsel also flags trade secrets, waived privilege and employee talk protected by US labor law (Frankfurt Kurnit).
Your tools have terms too. Salesforce's 2025 Slack API terms bar bulk export and AI use of data pulled through its API (the connection other tools use) (Hunton). Ask whether an admin's own export is treated differently.
How does a deal work?
Steps vary by buyer. This is the common shape.
Assessment
List your tools, years of records and rough counts.
Scope
Agree in writing which systems and dates are in or out.
Preparation
Someone extracts the records and removes names.
Contract and transfer
Sign first. Then the data moves.
Payment
Assessment usually starts from counts and dates, not raw records. Troveo says it needs no OAuth into your systems and no platform to install (Troveo). Many buyers also run a short estimate form, then a call.
Preparation differs a lot. Troveo says it scrubs names, contact details and client identities, and needs no sign-in access to your systems. Others connect read-only to the tools you approve.
micro1's CFO said in an undated recorded forum talk that its engineers help your IT people with the transfer, about 10 hours of work over a week or two from one or two of your IT people.
micro1's David Revland, who works on human data, said micro1 verifies the data before it pays.
Revland also said micro1 swaps each name for a consistent stand-in, so roles and cross-tool trails survive. Its page says personal information "is removed when relevant." Ask each buyer which approach it uses and what it means for your people.
Payment comes in four models:
- One-off transfer. SimpleClosure, whose terms give the buyer all right and title.
- Direct partnership. micro1's corpus track covers a full archive of company records, paid under the agreed partnership. micro1 says companies keep ownership of their underlying data.
- Revenue share. Troveo pays a cut of repeat deals. The split is unpublished.
- Deal revenue share. Protege pays a revenue share when your content enters qualified deals (Protege).
Exclusivity changes the price. Epoch AI's interviews put exclusive deals at roughly 4x to 5x the non-exclusive price. Troveo says its exclusivity is time-limited.
What do the public figures show?
There is no public price list, and Troveo's own guide says so. What exists is a short list of deals.
-
Spirit Airlines archive. Google's $10M bid won the 14 Aug 2026 auction, and court approval is pending.
The archive held about 100 million emails, 500 million Teams chats and 30 million lines of code. Mercor was the alternate at $7.5M (Time). micro1 later offered $12.5M (Fortune).
-
SimpleClosure. Nearly 100 deals in a year at $10,000 to $100,000 per shut-down startup (Gizmodo).
-
Mercor. Offers to shuttered startups reported at up to $300,000 each (BigGo Finance).
-
micro1 and Troveo. micro1 lists $100K+, $500K+ and $1M+ tiers. Troveo says full-company deals typically start at six figures.
Every public deal above is a distressed company selling a whole archive. None shows what a working company would get. The value guide covers what moves the number, and the deals tracker follows the public figures.
Which companies take company data?
Troveo is a marketplace that pays a revenue share. micro1 and Mercor deal with companies directly. SimpleClosure serves shutdowns.
The buyer directory covers each player neutrally, for all of them and more.
Some companies in this market also pay people who introduce a business.
Ask any comparison, this one included, whether its author gets paid when you sign up. Ours may be, and the disclosure page says how.
Do you have the right to license your company's records?
The largest risk sits in the archive. Your customers and employees wrote those messages, and your contracts may not let you license them.
Removing names is harder than it sounds. The UK data regulator says pseudonymised data is still personal data for anyone who holds the extra information (ICO). A label like "Person 14" keeps the links between records.
Law differs by place. The EU's EDPB Opinion 28/2024 requires anonymity claims to survive a test of "all the means reasonably likely to be used." The FTC calls quiet retroactive changes to terms for AI training potentially unfair or deceptive (FTC).
Whether you need consent from employees and customers depends on your contracts and your state. That question comes before any buyer call.
How do you do this safely?
Three steps before any call, in this order:
- Run the rights check. Seven questions on whose words sit in your systems and what your contracts say.
- Read the risk map. It maps the legal and security failure points for data owners.
- Hold your raw data. Keep exports in your own systems until a signed agreement says who receives them and when they are deleted.
Check each buyer's security record too. Mercor says a supply chain attack in March 2026 exposed sensitive information for a very limited subset of its experts (Mercor).
What contract questions go to a lawyer?
We are not lawyers, so read this as a list to bring to one.
- Scope. Which systems and dates, and what is excluded.
- Payment. One-off, revenue share or royalty, with the trigger in writing.
- Exclusivity. How long, for which buyers, at what premium.
- Removing names. Who does it, to what standard, and whether links survive.
- Raw data. When it moves, over what access, and when it is deleted.
- Liability. Who pays if a record leaks or a rights claim lands.
- Rights promises. What you warrant about ownership and consent.
- Buyer limits. What the buyer may do with the data, and whether it can resell.
- Term and exit. How long the license runs and how it ends.
- Referral fees. Whether anyone between you and the buyer takes a cut.
What are the red flags?
- A payout clock. A promise to pay within days usually starts once your data passes review, not at the first call. Ask what starts the clock.
- Headline ranges. A headline range can sit far above what the same site's calculator returns. Ask what the range is built from.
- Urgency. Posts saying owners who wait will get less cite no source. More than 350 companies have listed on SimpleClosure's hub since April, so supply is rising too (Fortune).
- No terms or privacy policy. If a buyer's terms or privacy page returns "not found," ask for the contract. Do not give any buyer read access before you have read it.
Where do you start this week?
Run the ten-minute check in how much your business data is worth to AI. It gives you the counts every buyer will ask for. Then the free fit check is the next step.
If your business runs a community, a course library or a podcast, read the guide to licensing your business's content to AI first. Members and guests hold rights the business may not.
Then put the contract questions above in front of a lawyer before any buyer call.
micro1 may pay us a referral fee if a company we introduce goes ahead with it. How we are paid
Apply to micro1's data partnership (paid link)micro1's own data-partnerships pagePaid link. How we are paid
Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one. Ask for a referral
