How to sell your company's data to AI companies

Short version

AI labs and the teams that build training environments for them pay for the records of real working companies: email, tickets, CRM history, job files, drawings, accounting, code reviews. You license a copy, you keep ownership, and you can keep running your business on the same data. The process is an evaluation, an NDA, a small sample, a price, and then a contract. Nothing moves until you've agreed to it in writing.

Why AI companies want your records

Models are now good at answering questions. They are still bad at doing a day's work at a real company: finding the right file in a messy shared drive, reading three email threads and a spreadsheet, and getting the invoice, quote or schedule right.

The labs can't fix that with the data they already have. Their own product logs only show the tasks people already try with today's models, and synthetic data doesn't look like the inside of a real business. A company archive does. Every request that came in, the back-and-forth that followed and the outcome you reached is a worked example of real work, with the answer already recorded.

That's why a ten-year-old roofing contractor, an architecture practice or a wound-down software startup can hold something a lab will pay for.

What kinds of data sell

The most useful archives are the ones that show work being done end to end, across several tools. Typical examples:

• Communication: email, Slack or Teams, support tickets, call recordings and transcripts.
• Systems of record: CRM, ERP, accounting, job management, project management, inventory.
• Work product: quotes, estimates, contracts, drawings, CAD and BIM files, reports, spreadsheets and financial models.
• Engineering: repositories with pull requests and reviews, issues, design docs, incident history.

What makes an archive more valuable: more years of retained history, more people doing desk work in it every day, more tools that link together, and work that's hard to fake, such as engineering, manufacturing, finance, healthcare operations or legal. We go into this in detail in what your company's data is worth.

Who buys it

There are three kinds of buyer:

• AI labs buying directly for post-training, evaluations and reinforcement learning.
• RL environment builders, who turn real company data into environments that labs run their agents in. See real company data for RL environments for how that works.
• Data companies and marketplaces that sit between owners and labs. That's what we do at Grepped.

Most owners don't sell straight to a lab. Labs rarely talk to individual small and mid-sized companies, and their data teams want a vendor who has already checked the rights, packaged the data and run quality checks. An intermediary also lets you license the same data to more than one buyer.

How the process works

1. Get an evaluation

Tell us what you hold: the tools you use, how many years of records, roughly how many people work in them, and your industry. We come back with what it's likely to be worth and which buyers want it. There's no cost and no commitment at this stage.

2. Sign an NDA

We sign a standard mutual NDA before you share anything, so a sample is protected from the start.

3. Send a small sample

A small slice is enough: one mailbox for a month, a few hundred tickets, a handful of job folders. Buyers judge quality on samples, and a good sample is what turns an estimate into a firm price. You choose what goes in it and can leave out any system, folder or record type.

4. Get a price from a real buyer

Before you commit to anything, you see a real number. You decide whether to go ahead.

5. Sign the licence and deliver

The contract sets out scope, how the data can be used and how you get paid. You keep ownership of your data. Whether a deal is exclusive or non-exclusive depends on you and the listing, and both are possible.

What to check before you sell

Most of the work in a data deal is making sure you can sell what you hold. Check these early:

• Ownership. The company, not its clients, should own the records. Work done under a client contract can be restricted.
• Contracts. Some customer and vendor agreements limit how data can be used. Read them, or ask us and we'll go through them with you.
• Earlier deals. If you've already licensed the data to someone, especially on an exclusive basis, that can block other buyers.
• Sensitive categories. Health records, payment card and bank details, government ID numbers, children's data, biometrics and privileged legal files carry extra rules. Some buyers exclude them outright.
• Who signs. Make sure the person signing has authority to license the company's data, especially if the company has wound down.

Our security page covers how we access, process and store data once a deal is signed.

Common questions

Do I lose my data if I license it?

No. You license a copy. You keep ownership and keep using your systems as normal.

Can I sell data from a company that has shut down?

Often, yes. Wound-down startups are a common source, because their Slack, email, code and tickets show a whole company's life from start to finish. Whoever holds the company's assets now needs to sign.

Do I need to clean or organise anything first?

No. Messy, real archives are the point. We handle preparation and packaging.

How long does it take?

The evaluation is quick. After that it mostly depends on how fast a sample comes through and how long contract review takes on your side.