Skip to content

The steps in a data deal

7 min read
  • Start with a description of the company’s records and the buyer’s intended use. Keep full exports out of the first conversation.
  • Use a small, approved sample to check format, quality and preparation needs before agreeing a larger delivery.
  • Name who prepares and checks the files, who pays each cost and when payment is due. The buyer pays the owner for the license and agrees SnowRock’s fee separately.
  • Agree how to handle corrections, updates and permitted uses after delivery.

A data license lets a buyer use a company’s information for an agreed purpose, such as developing or testing AI. If your company wants to license records, begin with a description of what you have and who can approve its use. If you are buying, explain the task and the examples you need. Both sides can assess the fit before anyone prepares a full export.

SnowRock finds potential matches, arranges introductions and coordinates discussions. The company and buyer agree the license, and the company delivers approved data directly to the buyer. The buyer agrees SnowRock’s introduction or sourcing fee separately. These steps identify who makes each decision, what must be ready and which costs to settle before proceeding.

1. Describe what the company has

The company names its record types, source systems, available years and staff who use them. A software business might have support conversations linked to bug reports. A distributor might have orders linked to shipping changes and returns. A professional services firm might have project briefs, reviewed drafts and final work.

Ask someone who works with those systems to check the description. Mark file counts as estimates until checked. Explain gaps, including missing attachments or older files in another system. SnowRock uses this summary to assess whether a buyer conversation is useful; the company does not need to export everything first.

2. Define what the buyer needs

The buyer describes what its AI should do or what it wants to test. For customer service, it might need delivery-problem conversations linked to the employee’s action and final outcome. That gives the company a specific set of records to look for.

The buyer’s brief states whether the data is for training, which changes how a model responds, or testing, which measures its performance. Add required languages, dates, formats and exclusions. Name a reviewer who can assess the sample. SnowRock compares the brief with the company’s description and brings open questions to the first discussion.

3. Approve an introduction and the information to share

The owner approves which company details SnowRock may share and with which buyer. It may allow a description of its support workflow while withholding customer names, examples and commercial details until confidentiality terms are agreed. SnowRock confirms that boundary and the people to introduce.

The company names the person authorized to approve a license and the staff who can check restrictions. A shared drive may mix employee work, customer material and third-party content. The U.S. Copyright Office explains that creators, employers and parties receiving transferred rights may hold copyright. Having access to a file does not by itself give someone the right to license it. Resolve relevant ownership, contract and privacy questions before sharing a sample.

4. Review a small, approved sample

The buyer and company agree the sample’s purpose, contents, permitted recipients and treatment after review. They also agree whether preparing it is paid work. The buyer’s reviewer checks whether the files open, related records connect and the contents support the intended task.

Include ordinary cases and cases that could change the purchase decision. For a returns workflow, examples might include a completed refund, a rejected request and a case with no recorded outcome. These are illustrative choices. The buyer should explain which cases it needs and why they are enough for an initial decision.

Google’s data-quality guidance identifies missing values, duplicates and incorrect labels as common problems. Record any relevant problems found in the sample. If a correction could resolve them, agree a small retest before expanding the order.

5. Agree the license and commercial terms

Once the sample shows a plausible fit, the company and buyer negotiate the data package and permitted uses. Specify whether the price includes one delivery, later updates, preparation or support. Count the company’s extraction and review costs when assessing an offer.

The buyer pays the company for the license and agrees SnowRock’s buyer-paid introduction or sourcing fee separately. A sourcing fee covers the specified search work; the license payment covers agreed use of the data. State what happens if preparation costs rise, fewer examples qualify or the buyer requests a wider use.

DecisionWhat to settle
UseThe allowed AI tasks, users and any onward sharing.
ScopeThe included data, excluded material and delivery period.
DurationHow long access or use continues and what happens when it ends.
ExclusivityWhether the company may license the same material elsewhere.
AcceptanceThe checks, review window and remedy for a failed delivery.
PaymentThe amount, currency, invoice requirements and due dates.

6. Assign preparation and transfer work

Assign each preparation task to a named person or provider: exporting files, connecting related records, removing excluded content and writing a field dictionary that explains each column. Agree who checks the package and who pays for corrections. SnowRock coordinates discussions; the company, buyer or an agreed provider prepares the files under the agreed scope.

The FTC’s business guidance recommends limiting sensitive information and access, protecting transfers and assessing service providers. The company and buyer choose a transfer method, name permitted recipients and set access limits. The company transfers approved data directly to the buyer. Keep a versioned record of the delivery so later questions refer to the same files. Use the agreed transfer method for sensitive exports rather than an ordinary outreach email thread.

7. Check the delivery against the agreement

The buyer’s reviewer checks the package within the agreed review period. Confirm the promised files, formats and record connections, then assess the contents against the agreed criteria. Acceptance depends on those checks. A disappointing AI experiment does not automatically mean the delivery failed the contract’s requirements.

For a problem, identify the affected files and unmet requirement. The company or preparation provider then corrects, replaces or explains the issue through the agreed process. SnowRock helps coordinate that discussion. A new language, task or labeling scheme may need a revised scope and price rather than a defect correction.

8. Complete payment on the agreed schedule

The contract sets payment milestones, such as an advance, an accepted sample or completed delivery. For each milestone, name the event that makes an invoice payable, the person who approves it and the due date.

The company issues its invoice and confirms payment details through its normal finance process. The buyer pays the license amount; SnowRock invoices its own agreed fee separately. For late payment, follow the agreed notice and resolution steps. The company decides which payment risk it will accept; SnowRock cannot guarantee the buyer’s payment.

9. Keep track of obligations after delivery

Keep a named contact on each side for corrections, permitted-use questions and promised updates. Store the version, delivery date and terms together. Hugging Face’s dataset-card guidance describes documenting a dataset’s contents, context and license. A short private delivery note can organize the same information for the buyer and company.

If an error appears later, identify the affected files and explain the correction. Check the license before approving a new use. Agree retention, deletion and treatment of outputs or trained models during negotiation: deleting a source file does not necessarily reverse its effect on a trained model.

Start with the description, then decide the next step

In a fictional example, a retailer describes support tickets linked to returns. A buyer asks for difficult return decisions. An approved sample reveals that older tickets cannot be linked reliably to outcomes, so the parties narrow the proposed delivery to a later period before committing to full preparation.

To begin, tell SnowRock what records your company has or what data your team needs. Include the person who understands the files. The first discussion should establish fit, permissions and whether a sample is worth preparing.

Sources and further reading

Discuss the next step in your deal

Tell us whether you are describing a source, reviewing a sample or discussing terms, and what you need to resolve next.

Talk to SnowRock