The data behind AI’s next leap should pay the companies that created it.
DataDouble is the trusted middle. We pay companies for their data, protect the people in it, and give labs data they can stand behind.
The most useful data was never on the internet.
The first generation of large models learned from the public internet. That supply is running out, and the next gains depend on something the web never captured: how skilled people actually do their jobs.
That knowledge lives inside companies, in support queues, call recordings, codebases, contracts and internal docs. Most of it sits idle, and the companies that created it see none of its value.
The ways this data changes hands today are often crude, with raw archives sold and little regard for the people in them. We think there’s a better trade: pay owners fairly, protect every individual, and give labs clean, documented data with clear rights.
Why DataDouble
Every dataset we license is a data double: a de-identified copy that keeps the knowledge and loses the people.
You keep the original. Labs get the double. You get paid.
Four principles we don’t trade away.
Owners get paid.
Data has value, and the companies that created it should capture that value.
Privacy is non-negotiable.
We would rather lose a deal than release a record that could identify someone.
Provenance or nothing.
Every dataset carries its chain of title. If we can’t document the rights, we don’t sell it.
Plain terms.
Sellers and labs see the same clear terms, written in plain English.