Guide · Compare offers
How to compare offers for your company's data
Updated 2026-10-16
This guide is general information, not legal advice. Have your own counsel check your contracts and employee notices.
Exclusivity term
What it is. Whether the buyer is the only one allowed to license your records, and for how long.
Why it matters. Exclusive deals can pay more, but you cannot sell the same records to anyone else for the whole term. Do not accept a perpetual exclusive license unless someone gives you a specific reason and a price that reflects it.
What brokers describe publicly. In the reinforcement-learning market, exclusive deals are described as paying about 4 to 5 times non-exclusive ones. A two-year exclusivity is the length brokers and partners most often describe, and shorter terms are said to lower the price. A carve-out limited to one field of use is described as a middle path.
Acceptance window
What it is. The period after delivery in which the buyer checks the data and either accepts it or rejects it.
Why it matters. Payment usually follows acceptance, so a long window delays your money, and a loose right to reject lets a buyer walk away after it has seen the data. Ask what the buyer may reject for, whether you can fix and resubmit, and what happens to any copies it already holds if it rejects after sampling.
What brokers describe publicly. Brokers describe acceptance or quality checks taking from about 5 to 30 days. One partner of micro1 describes roughly 30 days of quality checking, followed by payment 15 to 30 days later.
Net terms
What it is. How many days after acceptance the buyer pays, and in how many instalments.
Why it matters. The same headline number is worth less if it arrives in 60 days or in milestones. Ask whether payment is one lump sum, tied to delivery or to acceptance, and what interest or remedy applies if the buyer is late.
What brokers describe publicly. Published payment timing across buyers runs from about 7 days to Net 60. The side-by-side figures, with the date we recorded each, are on the compare page.
Scope of use
What it is. What the buyer may do with the data: train models only, or also use it inside products, evaluations or services it sells.
Why it matters. Narrow scope protects you; broad scope is worth more to the buyer, so it should be worth more to you. Check whether trained models may keep being used after the license ends.
What brokers describe publicly. Brokers list scope of use among the main terms to negotiate. Public program pages are often silent on it, and one commentator notes that a major program's public terms name no end buyer, so ask.
Sub-licensing
What it is. Whether the buyer may pass the data, or rights to it, to other companies.
Why it matters. If the buyer can resell, your records may reach parties you never heard of. Ask for a list of permitted recipients, or a rule that any onward license carries the same restrictions.
What brokers describe publicly. Marketplaces describe reselling a de-identified dataset to many buyers on a revenue share, which is a different deal from a one-off sale to a single lab. Know which one you are in.
De-identification obligations
What it is. Who removes names and other identifiers, who pays for it, how you know it was done, and when raw copies are deleted.
Why it matters. This is where employee and client privacy is protected or lost. Ask whether the buyer, a third party or you do the work, whether you can approve the vendor, whether you get a certificate of deletion, and the time limit for deleting raw data.
What brokers describe publicly. Practice differs by buyer. One program promises its own engine and deletion of raw data within 30 days; a partner of another describes raw data deleted within 120 days; some register entries say the buyer covers anonymization costs.
Liability caps and representations
What it is. What you promise about the records (that you have the right to license them, that they are what you described) and how much you could owe if a promise is wrong.
Why it matters. You are usually the one giving the promises, and a claim that the records included material you did not own can be larger than the price. Ask for a cap tied to the amount paid, limit warranties to what you actually know, and have counsel read this clause first.
What brokers describe publicly. Brokers list liability among the main terms to negotiate, and a law-firm note on selling company data recommends tight deal terms and excluding high-risk categories rather than relying on scrubbing.
Sources
- AI Data Broker Master Guide, Ryan Lock of Polyshares, 7 October 2026. Lock co-founded Polyshares, a buyer Minewell holds a referral link for, so read it as a commercial source.
- An FAQ on reinforcement learning environments, Epoch AI, 12 January 2026.
- Selling Company Data for AI Training, Frankfurt Kurnit Klein & Selz, 1 October 2026.
Ready to ask for offers? Get quotes or compare buyers side by side.