A business data license is an agreement between a company providing data and a buyer receiving permission to use it. The company may earn money from material it already creates. The buyer gains a defined input for a product, research project or AI system. The agreement explains what each side can do and what it must deliver.
A useful license is specific enough for the people doing the work. They should be able to identify the files, prepare the delivery, decide whether it meets the requirements and calculate the payment. This guide explains those commercial decisions. Legal counsel should turn the agreed terms into language suited to the parties, data and applicable law.
Understand what is being licensed
WIPO distinguishes a license, which grants permission to use intellectual property, from an assignment, which transfers ownership. A company can therefore grant defined usage rights while retaining rights of its own. The agreement needs to explain which rights are granted and which are retained.
Business data is not one uniform legal category. The U.S. Copyright Office explains that copyright does not protect facts themselves, although it can protect original expression. A package may also contain confidential information, contractual restrictions and material contributed by other parties. A file’s presence in a company database does not settle all those questions.
Describe the actual material and the authority to provide it. Avoid treating a broad phrase such as “we own the data” as a substitute for understanding customer terms, third-party contributions and the proposed use.
Personal information needs its own review. A commercial agreement between two companies does not, by itself, resolve the privacy requirements that apply to the people described in the files. Identify the information involved and the proposed use before promising it in a license. The parties can then determine what permissions, exclusions or other steps are required.
Write a scope both teams can recognize
The scope identifies the delivery. It can be a schedule attached to the agreement, containing file types, dates, required fields and exclusions. The useful level of detail depends on the package, but the buyer and company should interpret it the same way.
This hypothetical customer-support license illustrates the decisions. None of the quantities or conditions below is a market standard.
| Decision | Illustrative term |
|---|---|
| Unit of delivery | One completed billing-support case, including its message sequence and recorded resolution |
| Time range | Cases closed during a stated twelve-month period |
| Required context | Policy version, case category and a stable replacement identifier |
| Excluded information | Payment-card data, passwords and unnecessary personal contact details |
| Format | A structured export plus a field dictionary and file list |
| Updates | A fixed historical delivery; later updates require a separate agreement |
Name the permitted uses and users
“AI use” leaves important questions unanswered. The buyer may want to train a model, test it, supply reference information to an application or create new annotated examples. Those are different activities. Discuss each activity the buyer expects to perform and include the agreed permissions in the license.
Also identify who may receive the material. A buyer might use an external hosting provider, a labeling contractor or an affiliated company. The contract can define which recipients are permitted, what restrictions follow the data and who is responsible for their conduct.
Consider the rights in outputs and derivatives. The parties should discuss prepared datasets, annotations and models trained using the licensed material. Agreement on access to the original files alone may leave their later use unclear. A company should understand what permission continues after training has happened.
Separate exclusivity from the rest of the deal
A non-exclusive license can allow a company to enter other licenses, subject to its agreements. An exclusive arrangement reserves specified rights for a buyer. WIPO recognizes these as different licensing choices. Neither choice should be assumed from the purchase price or from a request for confidentiality.
If exclusivity matters, define its boundaries. Does it cover all of the company’s data or one collection? Does it apply to one industry, one use or all uses? How long does it last? A narrow reservation for a defined period has different business consequences from a permanent restriction.
The company should assess what it gives up. The buyer should explain what commercial problem exclusivity solves. This makes the price discussion concrete without pretending there is a standard premium for every dataset.
Agree sample access before the full license
A sample lets both sides test assumptions before committing to the larger delivery. The sample agreement should name the recipients, review purpose, permitted actions and retention period. A buyer evaluating a source does not necessarily need permission to train a production model on that sample.
A nondisclosure agreement governs specified confidential disclosures and uses. Its title alone does not establish the training rights either side expects. WIPO’s trade-secret guidance emphasizes controlling confidential information during negotiations as well as under the final license.
Use sample findings to update the scope. If software issue histories lack the test results the buyer needs, decide whether those results can be supplied. If extracting them would cost more than the deal supports, change the package before agreeing a full-delivery deadline.
Assign preparation, delivery and acceptance
Someone must export the information, remove exclusions, link related files and answer questions. Name that party. If an outside provider handles preparation, define its work and price. SnowRock coordinating a discussion does not, by itself, mean extraction or technical processing is included.
Acceptance is the process for deciding whether the delivered package meets the agreed specification. Set the checks, the reviewer, the review period and the response to defects. File counts and readability may be straightforward to test. Expert judgments about content require a clear rubric and a person qualified to use it.
Distinguish a missing required file from a model experiment that produces disappointing results. Whether payment depends on file conformity, a performance test or another outcome must be agreed. An undefined requirement to “improve AI” gives neither side a dependable delivery target.
- Who prepares the data, and what must they remove or transform?
- Where is it delivered, and who confirms receipt?
- What checks establish acceptance, and by what date?
- Can defects be corrected, and how are disputes or partial deliveries handled?
Make the payment calculation visible
The buyer pays the company for the agreed license. Preparation and delivery may create additional costs. SnowRock’s introduction or sourcing fee is a separate buyer-paid amount with its own scope and payment trigger. State these amounts separately so gross receipts are not confused with profit.
Consider this fully hypothetical example. A buyer licenses 1,000 accepted cases for $20,000. The buyer also pays a preparation provider $4,000 and SnowRock an agreed $6,000 sourcing fee. The buyer’s total is $30,000, or $30 per accepted case. These figures illustrate arithmetic; they are not SnowRock prices or evidence of market value.
In that arrangement, the data company receives $20,000 before its own costs and taxes. SnowRock receives $6,000 before its own costs and taxes. The preparation provider receives $4,000. If the contract permits the same fixed charges for only 800 accepted cases, the buyer’s cost becomes $37.50 per case. This shows why acceptance and pricing must be read together.
Agree when each invoice becomes payable. Possible milestones include starting the search, providing a sample or accepting a delivery. Decide who finances preparation before payment and what happens if the project stops. A signed contract, an invoice and cash received are different events.
Define obligations after delivery
Decide whether the company owes corrections, updates or support. A fixed historical collection may need no ongoing refresh. A changing product catalog may need a schedule and a process for correcting errors. The price should reflect the work actually promised.
The agreement should also address security responsibilities, reporting of relevant incidents, records needed to check compliance and consequences of a material breach. Any promises about rights or quality should describe what the company can support. Parties should negotiate liability and remedies with their advisers rather than assume a generic template allocates risk fairly.
Finally, explain expiration and termination. State whether the buyer must delete or return source files, whether limited copies may remain for required purposes and which usage rights survive. Do not assume that deleting a source file reverses prior model training. That question needs to be resolved as part of the intended-use discussion.
Put the brokerage role in writing
SnowRock helps companies find buyers and helps buyers source business data. It coordinates introductions and the deal process. The company remains responsible for its decision to license; the buyer remains responsible for assessing fit and agreeing its intended use.
Before work begins, the buyer and SnowRock should agree whether the engagement is an introduction or a defined sourcing assignment, what it includes and when the fee is earned. The company’s license payment is negotiated separately. There is no upfront SnowRock charge for a company introduction.
A practical first discussion needs a description of the data and the buyer’s task. The next documents can then follow the actual decisions: a scoped sample arrangement, agreed commercial terms and the full license. Resolve material gaps while the deal is still small enough to change.