This article was previously published as part of the AI Aperture Series, at https://www.linkedin.com/feed/update/urn:li:activity:7506372653369286657/

There is no reason to grant a license so broad that it is the same as putting yourself out of business.  Data is not exhausted by use.  The same corpus can be licensed to a clinical AI company for diagnostic support, to an insurer for underwriting, and to a device maker for maintenance scheduling.  Three grants, three payments, one dataset.

Field of use is the clause that is supposed to make that partition possible.  I have written before that it is inherently imprecise, and that two licensees in the same industry force carve-outs and complications so the licensor does not give away the asset by granting a single license.  That imprecision costs more now that data is a commercial asset rather than a byproduct.

AI adds a failure the carve-outs cannot reach.  Once a corpus trains a foundation model, the model is deployed across every field at once.  A restriction drawn around the licensee's industry has nothing left to attach to.

What survives is the decision.  Scope the grant to the decisions the licensee may make with the output rather than to the market the licensee sells into.  That is the Decision Rights model.  It also negotiates better, and that is the part I did not expect.  The dataset does not have to be granted as a block.  Divide it into subsets and give each subset its own scope of decisions.

I have drafted agreements this way.  It ties the licensed use to the data at the level of the data rather than at the level of the licensee.  Both sides can then see what is in scope without arguing about where an industry ends.  The licensor grants what it is willing to grant on each subset and holds the rest.  The licensee takes what it needs for the decisions it actually makes and does not pay for scope it cannot use.  Each side addresses its own risk on its own terms and the deal moves.  Everything not granted is retained.  A licensor who scopes by decision is still selling the corpus after the first deal closes.

One dataset can carry several revenue streams.  Field of use is what stops it.