Selling versus licensing business data: what changes and what does not
A practical guide to ownership, permitted uses, exclusivity and the contract terms that determine control.
Start with what the agreement transfers
A licensing arrangement grants specified uses; an asset sale may transfer the asset or associated rights. The document, not the marketing verb, determines the transaction. micro1 and Troveo say businesses retain underlying ownership. Centific also describes licensing, but says a wind-down transaction may use an asset-purchase agreement.
Do not assume that an offer labeled “sell your data” transfers ownership, or that one labeled “license” preserves every form of control. Request the actual grant of rights, definitions of the dataset and derivatives, and any assignment language. This guide is a review framework, not a legal interpretation of your proposed agreement.
Retained ownership is only the first question
Retaining ownership answers who owns the underlying asset. It does not by itself answer whether the license is exclusive, perpetual, revocable, transferable or available to competitors. Troveo’s ownership statement therefore should not be read as a complete description of its downstream terms.
Ask what you can still do after signing: operate the business, use the records internally, license the same material to another buyer, develop a competing product or sell the company. If the answer depends on carve-outs, have those carve-outs recorded explicitly rather than inferred from a reassuring headline.
- Define the included systems, dates, teams and record types.
- Separate ownership of source records from rights in prepared or synthetic datasets.
- Ask whether internal reuse and future licensing are expressly preserved.
- Check whether restrictions survive termination or a change of company ownership.
Permitted uses need more detail than “AI”
Handshake describes company data being used for model training and evaluation. Appen describes licensed workflows and optional annotation from people who did the work. Those descriptions help explain the opportunity, but your agreement still needs to identify which uses you authorize.
Consider the difference between testing a model, training it, constructing an agent simulation and creating a resaleable derived dataset. Do not assume that approving one authorizes the others. Request definitions that a non-specialist owner can understand, and identify who can access each version of the material.
- Ask about training, fine-tuning, evaluation and benchmarking separately.
- Ask about synthetic rewrites, derived products and model or output rights.
- Identify onward licensing, buyer categories and any competitor exclusions.
- Specify approval rights for uses outside the initial scope.
Exclusivity is a commercial constraint, not a quality badge
An exclusivity proposal should be examined by scope, purpose, buyer group, territory and duration. A narrow exclusive right for one cleared workflow is different from exclusivity over every record the company has or will produce. These are negotiating questions, not descriptions of a standard contract.
If a partner needs exclusivity to prepare and market the dataset, ask what it commits in return. Possible discussion points include a deadline for a first transaction, minimum performance, reporting and an exit if no buyer emerges. Do not treat any of those protections as present unless written into the agreement.
- Separate existing records from future refreshes.
- Ask whether exclusivity attaches to the source data, prepared version or both.
- Check whether derived products remain restricted after the original term.
- Compare the payment with the future opportunities you would give up.
Termination, deletion and completed use are different
Request distinct rules for raw exports, prepared datasets, backups, onward recipients and already-created products. micro1 says retention and deletion follow agreed terms, but its public page does not provide a complete operational-dataset deletion schedule. A general website privacy notice is not a substitute for your negotiated dataset terms.
Do not promise stakeholders that deleting a source file will reverse model training. Instead ask how completed training, derivatives and buyer rights are treated after termination, and what evidence of deletion can be provided for copies that are subject to deletion. If the answer is unclear or unacceptable, pause before granting rights.
Make the rights review part of the price comparison
Two offers for the same records can purchase very different permissions. Compare price only after mapping term, exclusivity, downstream uses, repeat-sale accounting, warranties and liability. A larger headline amount is not automatically the better transaction if it buys much broader rights.
Before signing, involve an authorized company decision-maker and qualified counsel for the relevant jurisdictions. Possession of an export does not establish licensing authority. Do not proceed if client restrictions, contractor IP or personal-data obligations cannot be cleared for the proposed use.
- Summarize each grant of rights in plain language beside its payment terms.
- Ask who is contractually responsible for extraction, transformation and downstream handling.
- Check indemnities, liability caps, dispute procedures and obligations that survive termination.
- Keep a dated copy of the approved scope and executed agreement.
Public program descriptions are not seller-specific contracts or legal advice. See our evidence methodology.