What is a data sample, and should you share one?

Updated October 25, 2026 · How we research

Why do buyers ask for a sample?

Buyers cannot price records they have not seen. A sample shows them how complete the records are, how they are structured, and whether they fit the kinds of tasks their AI customers need. This is standard in data markets generally; one large data marketplace's provider guide notes that most customers want to evaluate a sample before committing.

A sample also helps you. It turns a vague conversation into a concrete one, and it shows you how the buyer handles your information.

Expect the buyer to look at things like how many records have useful detail, how consistent the format is, how many years the records cover, and how much would be lost after personal information is removed.

What are the two kinds of samples?

There are usually two different samples in a deal, and it helps to keep them apart.

• Evaluation sample: a small set of records, or just a description of their structure, shared early so the buyer can assess value
• Review sample: a slice of the de-identified output the buyer has prepared, which you check and approve before the full dataset is delivered

What should an evaluation sample contain?

As little as needed. Some buyer programs say they can scope a deal from structural information alone, such as a list of tools, field names and record counts, without seeing any content. Start there.

If real records are needed, pick a small set that represents the whole, such as a few dozen tickets or threads from different years and teams. Remove or replace names, contact details, account numbers and anything sensitive before it leaves. Keep a copy of exactly what you shared and when.

What protections should be in place first?

Sign a confidentiality agreement before sharing any content. A typical NDA limits use of the information to the stated purpose, limits who can see it, and requires the recipient to return or destroy it on request and confirm that in writing. Check that the NDA covers the sample, any notes or analysis made from it, and what happens if no deal is signed.

Do not hand over system passwords or admin access just to produce a sample. If the buyer needs access later, it should be read-only, limited to the agreed systems, and set out in the contract.

• Signed NDA covering samples and anything derived from them
• Clear statement that the sample may not be used for training
• Return or deletion on request, with written confirmation
• A named contact on both sides

What mistakes should you avoid with samples?

A few common mistakes are worth avoiding. They mostly come from moving too fast in an early conversation.

• Sending raw exports before an NDA is signed
• Picking only your best records, which leads to a price the full dataset cannot support
• Forgetting attachments, signatures and quoted replies, which often hold names and contact details
• Including records from a customer whose contract bars outside use
• Losing track of who received which sample

How do you review the buyer's de-identified sample?

Several buyer programs describe a step where you review and sign off a de-identified sample before the full dataset moves, and some say payment follows acceptance rather than submission. Use that step seriously.

Read the sample as an outsider would. Could you tell which customer, employee or deal a record is about from job titles, places, dates or unusual events? Are confidential customer details still visible? If so, ask for changes and a fresh sample before approving.

General information, not legal, tax or financial advice.