Licensing vs. selling data: what is the difference?

When a founder first hears that AI labs pay for company data, the natural phrase is "selling our data." That phrase is understandable, but it describes a different transaction from the one most companies should want. A licence grants permission to use something for a defined purpose, under defined limits, while the company keeps control of what it allowed and why.

The distinction is not just vocabulary. It shapes what the buyer can do with the material, what the seller can promise its employees and customers, and how comfortable a board or liquidator will be signing. This post walks through the difference in plain terms.

This is general information, not legal advice, and any company considering a data licence should involve its own counsel before signing.

A sale, in the everyday sense, transfers ownership. Once you sell a car, the buyer can drive it, lend it, resell it or take it apart, and you have no say. When people talk about selling data, they often picture the same thing: a file changes hands and the new holder can do whatever it likes.

That model is a poor fit for operational data. A company's Slack history, email, tickets and CRM records are full of other people: employees, customers, vendors, candidates. A buyer with unrestricted rights could, in principle, repackage that material for purposes nobody at the company would have agreed to, such as marketing lists or people search.

Even with anonymization, an unrestricted transfer leaves the seller with no contractual way to object to later uses. That is a weak position for anyone who has to answer for the decision afterward.

A licence is a permission, and permissions can be narrow. The licensor says, in effect, "you may use this material for these purposes, in this way, and not otherwise." Everything outside the grant stays prohibited.

In company data licensing for AI, the grant is typically limited to training and evaluation. A lab can use the anonymized data to train models, build reinforcement learning environments, post-train and run evaluations. It cannot use the data for anything else. The licences we arrange go further and spell out what the buyer may never do:

• Market to, profile or contact anyone who appears in the data
• Attempt to re-identify any person or company behind a token
• Resell or pass the data on for marketing or people-search purposes
• Use the material outside the training and evaluation purposes in the grant

Those restrictions travel with the data. They are what make a licence defensible to employees, customers and creditors in a way that a bare sale is not.

A licence also lets the seller decide what is in scope before anything moves. Under the process we run, the company sets its scope and exclusions first, and those terms are annexed to the contract before any data is touched. If a particular client, a Slack channel or a shared drive folder should stay out, it stays out.

Some categories are excluded by default. HR, payroll, medical and legally privileged folders are left out unless there is a specific reason and the right to include them. Direct messages are excluded unless the company chooses to include them and has the right to do so.

Then there is the question of what form the data takes when it leaves. The buyer never receives raw records. Every person, company, address, email, phone number and account is replaced with a consistent token, credentials and card numbers are stripped out, and faces, logos and signatures are blacked out in images and documents. For a deeper look at how that works, see Anonymization 101: tokenization vs. redaction.

In a sale, the seller's involvement usually ends at closing. In a licence, the seller keeps a set of rights and assurances that continue after payment.

The most practical one is deletion. In our process, the raw archive used to produce the anonymized dataset is deleted within 30 days of the seller signing off on the sample, and that obligation is written into the contract with a deletion certificate at the end. The seller is not left wondering whether an unredacted copy of its workspace is sitting on a server somewhere.

The seller also keeps the use restrictions themselves. If the buyer ever used the data for something outside the grant, that would be a breach of the licence terms, not a gray area. The restriction is only as good as its wording, which is why it is worth reading the use clause as carefully as the price.

It might seem that buyers would always want the broadest possible rights. In practice, serious AI labs care a great deal about where their data came from and whether it was obtained cleanly. A licence with clear scope, clear authority to sign and clear use limits gives them provenance they can document.

An unrestricted "sale" of the same material can actually be harder for a careful buyer to accept. It raises questions about whether the seller had the right to transfer everything, whether employees and customers were considered, and whether the material might resurface elsewhere. A narrow licence answers those questions up front.

There is also a pricing reality. Value comes from connection depth, years of history, headcount and the number of systems the same work runs through, not from whether the buyer gets to own the data outright. Typical licences we see range from $20K to $400K+, and nobody can give a firm number before the data has been counted. Owning the data forever does not make it more useful for training; being well connected does.

Imagine a 60-person logistics software company that is winding down. Its directors want to recover something for creditors, but they also worked alongside the people in that Slack workspace for years. A sale would ask them to let go of the data entirely. A licence asks a narrower question: are you comfortable allowing anonymized copies of these records to be used to train and evaluate AI systems, with marketing, profiling and re-identification ruled out, and the raw archive deleted afterward?

Most people find the second question much easier to answer. It also leaves a clearer paper trail, which matters when a board, a liquidator or an auditor asks later what was done and on what terms.

If you are weighing what your company's records could be worth, and on what terms they would be used, book a 15-minute call and we can walk through scope, exclusions and how a licence would work for you.