What is your company's data worth to AI labs?

Short version

There's no list price for company data. What a buyer pays depends on how much real work your records show, how many years they cover, how many tools they span and how hard that work is to find anywhere else. Our average deal is around $1.1 million to $1.3 million, and it's rising as we work with more high-end buyers. Plenty of archives are worth less than that, and some are worth a lot more. The only way to get a real number is an evaluation and then a sample.

What buyers are actually paying for

A lab isn't buying gigabytes. It's buying a better model. Your archive is valuable when it lets a researcher show a model failing at a real task, train on it and see the score go up.

So the question a buyer asks isn't "how much data is there?" It's "how many realistic, checkable tasks can we get out of this, and can we get them anywhere else?" Everything below comes back to that.

What pushes the price up

Years of retained history

Ten years of records beat two, but only if you kept them. A ten-year-old company that cleared out old Slack messages has two years of Slack. When we ask about history, we ask when each tool was started and what's still there.

People doing desk work

Buyers count the people who work in email and software every day, not field crews or drivers. More people at desks means more requests, more handoffs and more decisions on record.

How many tools, and whether they connect

An inbox on its own is worth something. An inbox, a CRM, an accounting system, a job tracker and a shared drive that all refer to the same customers and jobs is worth much more, because an agent has to work across them the way your team does. Five or more tools is a good sign. Eight or more is better.

Hard, specialised work

Back-office and operations work across messy files and spreadsheets is in demand everywhere. Engineering and design files, manufacturing records, finance, healthcare operations, legal work, chip design and bio lab workflows are harder to find and price higher. Generic marketing content prices lower.

Outcomes on record

Records that show what happened in the end are worth more: the quote that was won or lost, the ticket that was resolved, the invoice that was paid, the design that passed review. The outcome is what lets a buyer check whether an agent got the task right.

Clean rights

Data you clearly own, that isn't tied up by client contracts or an earlier exclusive deal, can be sold to more buyers. That raises what it's worth to you over time.

What pulls the price down

• Public or near-public material. If it's already on the open web, labs probably have it.
• Thin history. A few months of records, or one tool only.
• Mostly non-English text, unless a buyer is specifically after that language.
• Restricted categories that buyers have to exclude, such as health records, payment details or privileged legal files, unless the right permissions are already in place.
• Records the company doesn't own, for example files that belong to clients.

One sale or several

A non-exclusive licence lets you sell the same data to several buyers, so your total can be well above any single deal. An exclusive licence goes to one buyer, usually at a higher price. Both are possible, and which one makes sense depends on you and on what the buyer wants. Either way, you keep ownership of your data, and the licence terms are agreed in writing before anything moves.

How to get a real number

1. Get an evaluation. Tell us your tools, years of records, team size and industry. We'll give you an estimate and tell you which buyers want it.
2. Send a small sample under NDA. A sample is what turns an estimate into a firm price.
3. See the offer. You get a real number from a real buyer before you commit to anything.

If you want the background on the whole process first, read how to sell your company's data to AI companies.

Common questions

Is a small company's data worth anything?

Often, yes. A ten-person firm with eight years of linked email, jobs and accounting can be worth more than a much bigger company with one messy shared drive. Desk headcount, history and tools matter more than revenue.

Why won't anyone give me a price up front?

Because the price depends on what's actually in the records, and nobody can see that without a sample. An evaluation gets you a range. A sample gets you a real number.

Does it cost anything to find out?

No. Evaluations and samples are free, and you don't commit to anything until you've seen the offer.