Training data supplier types
"Training data provider" covers businesses with opposite economics. Some sell labour by the hour; others sell rights to data that already exists. A seller who confuses the two will pitch the wrong counterparty.
The two axes that matter
We classify suppliers on two questions. First: is the data created to order, or does it already exist? Second: who holds the rights at delivery — the buyer, the supplier, or the original owner under licence?
Those two answers help identify likely pricing structures, contract questions and whether an outside data owner has a role. Verify the particular programme and agreement.
Five supply models
- Annotation / labeling service
What is deliveredLabels on buyer-supplied raw data
Rights at deliveryContract determines output rights
Role for a dataset ownerUsually a service purchase, not an archive licence
- Expert-generated data network
What is deliveredNew prompts, answers, rubrics and judgments
Rights at deliveryOften assigned under a services contract
Role for a dataset ownerExperts contribute work rather than sell an archive
- Dataset licensor / broker
What is deliveredRights to existing third-party datasets
Rights at deliveryOwner grants defined rights
Role for a dataset ownerA possible channel for existing assets
- Archive or rights holder
What is deliveredIts own content corpus
Rights at deliveryOwner retains title and grants a licence
Role for a dataset ownerThe owner is the supplier
- Evaluation / benchmark builder
What is deliveredHeld-out test sets and grading tasks
Rights at deliveryContract-specific; training use may be excluded
Role for a dataset ownerA narrow route for specialist material
What this means if you own data
- Dataset licensors and archive owners are the relevant supply models for existing-asset licensing.
- An expert-work programme recruits people to create data. A company may also run a separate acquisition programme; verify the specific offering rather than judging the entire company by one service.
- An annotation business is not automatically a buyer of your archive. Check whether it has a separate licensed-data channel.
Have a dataset worth discussing?
Tell us what you own, which rights you can license and the commercial scope you have in mind.