SnowRock helps companies explore licensing their data to AI developers. It also helps buyers look for companies with data that meets a defined requirement. The work starts with a description of what one side has and what the other needs, followed by introductions and coordination where there is a possible match.
These questions cover the decisions that usually come before a company shares a sample or a buyer commits to a search. They explain the choices rather than assume every company, dataset or deal should follow the same route.
What kinds of company data can be considered?
Companies can describe documents, research, images, technical files, financial reports, support conversations, transaction histories and other material created through their work. SnowRock’s scope spans industries. A software company, professional services firm, retailer or engineering business may each have a different collection to discuss.
The important question is whether the material supports a particular buyer’s task. A folder of final reports may fit one request; another buyer may need the questions, working files, revisions and final decisions behind those reports. Explain the contents and their context before spending time on a large export. Also identify material owned by others or subject to restrictions.
Do we need a large company or millions of files?
There is no single size threshold that makes data suitable for every buyer. The required amount depends on the task, how varied the examples are and what the buyer already has. A narrowly defined collection may be worth reviewing when it captures work the buyer cannot find elsewhere. A large archive may still miss the required details.
Give an estimate of unique cases or documents, the dates covered and how often new material is created. Distinguish a business event from the files it produces. One customer case may include several messages and attachments; counting them separately can make the collection sound larger without adding more independent examples.
How much could the company earn?
A meaningful price requires a defined collection, permitted uses and a buyer interested in those terms. Relevant factors include how well the data meets the request, its coverage, the work needed to prepare it, whether updates are included and whether the buyer seeks exclusivity. SnowRock does not offer a universal rate per employee, file or year of history.
Compare offers using the same scope. A payment for one approved snapshot is different from a payment that includes continuing updates or extensive preparation. Ask which costs the company must cover and when payment becomes due. Until a buyer has made an offer, an estimated opportunity should not be treated as expected revenue.
Who pays SnowRock, and who pays the company?
Under SnowRock’s buyer-paid approach, the buyer pays SnowRock for the agreed introduction or sourcing work. The buyer’s payment to the company for the data license is separate. The company does not pay an upfront SnowRock introduction fee. Any different service or additional work should be described and priced before it begins.
Keeping the payments separate makes the agreement easier to assess. The company can see what it receives for the license. The buyer can see what it pays for brokerage services. Both can identify who pays for exports, review, preparation and delivery, rather than assuming those costs are included in either fee.
Would licensing mean giving up ownership or control?
A license can grant particular uses while the company retains rights in the material. The actual agreement controls the scope. Read the terms for permitted uses, exclusivity, redistribution and duration, and check what happens to copies or work produced under the license when it ends.
The first ownership question is whether the company can grant those rights at all. Possessing a document does not automatically mean owning its copyright. The U.S. Copyright Office distinguishes rights in a work from ownership of a copy. Customer and contributor agreements may also affect what can be shared. Have the person responsible for those agreements involved early.
What should we share before there is a deal?
Start with a written description: the data types, approximate volume, history, source systems and known restrictions. A field list or a made-up example can sometimes explain the structure without exposing the underlying content. Label made-up examples clearly so a buyer does not mistake them for a sample of real data.
If the buyer needs a real sample, agree who can see it, the review purpose, how it will be transferred, how long it can be retained and what happens if the discussion ends. A confidentiality agreement addresses confidentiality; the sample terms should also state whether training or other reuse is permitted. Share only the approved sample through the agreed channel.
Can we remove names and then license the files?
Removing names may be one preparation step, but it does not settle every permission or privacy question. A combination of dates, locations and other details may still identify someone. In its UK guidance, the Information Commissioner’s Office distinguishes anonymisation from replacing direct identifiers with substitutes.
First identify what the buyer actually needs. Exclude unnecessary information, check the relevant obligations and agree a preparation method with the people responsible for the data. For sensitive or regulated material, appropriate specialist review may determine whether the proposed use can proceed at all. SnowRock’s first conversation needs a description, not an upload of confidential files.
What should an AI buyer include in a request?
Describe the task first. ‘Financial data’ could mean invoices for document extraction, working papers for testing accounting analysis, or transaction histories for a different model. Those requests lead to different sources, fields, permissions and review methods.
Then specify the data format, history, approximate volume, required links between records, intended use and timeline. Separate requirements from preferences. If a missing field would make a source unusable, say so. Include the person who will review a sample and the process for making a purchase decision. These details help SnowRock research a defined need rather than produce a list of vaguely relevant companies.
Does the request need to distinguish training from evaluation?
Yes. Training changes a model using examples; evaluation tests how a system performs. The intended use affects both the rights being requested and how the data should be selected and handled. Evaluation may require examples that have not been used for training.
NIST’s AI Technology Evaluation program uses a controlled test environment to help prevent training and test data from mixing. Its program description illustrates why some evaluation collections are deliberately kept out of public circulation. A buyer should explain any similar requirement before a company prepares or distributes a sample. Public availability can be useful for one purpose and unsuitable for another.
When does a buyer need an introduction or a sourcing assignment?
An introduction is appropriate when a potentially relevant company has been identified and the next step is a direct conversation. A sourcing assignment is a commissioned search against a written brief. It can include research, outreach where agreed, and reporting on what was found.
Before commissioning a search, agree the deliverables, scope, timing and fee. Research can identify a promising source without establishing that the company is willing or able to license data. The report should distinguish those states so the buyer knows what is ready for a conversation and what still needs work.
| Approach | What to agree before starting |
|---|---|
| Introduction | The parties being connected, the purpose of the introduction and the buyer’s fee arrangement. |
| Sourcing assignment | The search requirements, research and outreach scope, expected deliverables, timing and payment terms. |
| Data license | The company’s approved data, the buyer’s permitted uses, preparation, acceptance and license payment. |
Who prepares the data, and is that included?
Preparation must have a named owner. Depending on the agreement, the company, buyer or a separate provider may handle exports, format changes, removal of sensitive details and checks on the delivered files. SnowRock’s role in coordinating a deal does not by itself mean those technical services are included.
Estimate the work after reviewing a representative sample. A collection that looks simple in a file count may require substantial effort to link attachments, interpret fields or remove restricted material. Agree who pays for that effort, how changes to scope are approved and what happens if preparation reveals a problem that makes the original plan impractical.
When is delivery complete, and when does payment happen?
Delivery is complete according to the agreed acceptance process. Define the checks, the buyer’s review period, how problems are reported and whether the company can correct them. Checks can cover readable formats, included dates, required fields and the completeness of agreed links between files.
Payment may follow a deposit, delivery milestone, acceptance or another negotiated schedule. Put the trigger and due date in writing. Avoid making the company’s payment depend on an undefined future improvement in a buyer’s model. If model performance is part of the bargain, both sides need to agree the measurement method before committing to it.
How long does a deal take, and what if there is no match?
Timing depends on buyer demand, access to the right decision-makers, the clarity of permissions and the work needed to prepare the collection. An initial description can be reviewed before the data is ready to transfer. A delivery date should follow an agreed scope and a realistic preparation plan.
A discussion may end because the data does not match the task, the terms do not work, or the preparation cost is too high. A paid sourcing assignment should still produce its agreed research deliverables even when no license results. The distinction between paying for a search and paying for a completed transaction belongs in the assignment terms.
What is the most useful next step?
For a company, prepare a short description of the material, the systems holding it, the period it covers and who can approve a discussion. For a buyer, prepare the task, required examples, intended use and target timeline. Include the main uncertainty you need help resolving.
Send that description through SnowRock’s inquiry form. It gives the first conversation a practical starting point: whether there is a possible match, which questions need answers and what work should happen next.