What is a data sample, and what should I send?

A data sample is a small, representative slice of your records, usually de-identified, that lets a buyer check whether your archive matches your inventory before it commits to a firm price. Send only what the buyer has asked for, with the scope agreed in writing and an NDA in place.

What is a sample for?

It turns an indicative range into a firm price. Polyshares says “We read a sample of the record. That is how we price it.” The steps around it are in how data licensing offers work.

Samples also run the other way. Some programs show you a prepared sample before anything is licensed. Appen says “You review a representative sample pack before anything is licensed”, and Corpus says the seller receives a redaction report with samples and signs off before transfer (Corpus). The first kind lets the buyer check your data. The second lets you check the buyer’s de-identification. Ask for both.

What do buyers look for in a sample?

Work they can follow from start to finish. A sample that holds one whole job, from the first request through the approvals to the invoice, shows a buyer more than a large batch of unrelated messages.

Within that, buyers check four things:

• Links across systems. Forbes reported Sunset’s CEO as saying that traceability across systems drives value, so a ticket tied to a specific code commit is worth more than a standalone document (Forbes). Scrimdata’s version is “Interconnection matters more than volume”.
• People deciding things. Threads where someone asks, decides and explains carry more than automated alerts and notifications.
• Dates that match the inventory. If the inventory says email from 2014, the sample should be able to show 2014.
• De-identification that holds. Polyshares says that if an anonymized sample still contains identifying material, “the record goes back for another pass before pricing continues”.

How much should a sample include, and what should stay out?

Agree the scope in writing before you pull anything: which systems, which date window, how many records, which workflows. A few complete workflows over a defined period usually tell a buyer more than a random cut across everything. Scale AI says partnerships “can begin with a very limited scope”, and a sample can be narrower still.

Leave out anything you would not license. Troveo’s closing-company guide lists what comes out before delivery: “Customer personal information, employee personal information, credentials and secrets, third-party confidential material, and anything a customer contract restricts”. Miro Advisory adds that “Client information should not be included simply because it can be anonymized”.

To prepare a sample on your own computer first, the free sample tool counts what your Slack, email and spreadsheet exports hold and replaces names and contact details with consistent labels, so you can review every replacement before anything is shared.

Who should de-identify the sample?

That depends on the program, and it decides who sees your raw records. As of October 2026, Handshake AI says de-identification happens on its side, so you “transfer the data as it exists today”; Nyne describes the same arrangement (Nyne). Scale AI says it anonymizes “inside your environment, before any data is extracted”, and Mercor says “No data leaves your systems until you approve it”.

Before sending anything, ask three things: whether the sample leaves your systems raw or de-identified, who sees it in raw form, and when it will be deleted. The FAQ of Telegraph Lab (affiliated with this site) makes the limit explicit: de-identification “reduces risk; it cannot guarantee that re-identification is impossible”.

How can names come out without breaking the record?

By replacing each identifier with the same label every time it appears. The following is an illustration of the technique, not a description of any one provider’s system.

Amy, a property manager, emails a contractor for a quote, approves the cost in the accounting system and chases the invoice in chat. In the de-identified copy, Amy becomes Person 01 in all three places, and the contractor becomes Company 01. A reader can still see that the same person requested the quote, approved it and followed it up, without learning who she is. If every name were replaced with the same blank tag instead, the chain would break: the record would show that someone approved something, not that the person who asked for the quote also approved it.

Several providers describe this approach on their own pages. Telegraph Lab says it can “replace identifiers with consistent placeholders”, Sell My Business Data says “Replace identifiers with consistent anonymous labels”, and micro1 has published a model meant to “transform real identities into consistent synthetic identities”.

Consistency keeps the pattern of who did what, and in a small company that pattern alone can point to a person. A sample is the place to test it: ask whether someone who knows your company could tell who Person 01 is.

Do I need an NDA before sending a sample?

Yes, and several programs build one in early. Appen says “nothing is discussed further before a mutual NDA is in place”, Handshake AI says everything is “scoped upfront, protected under NDA”, and Scrimdata scopes under NDA using only your tool stack and metadata (Scrimdata).

An NDA covers confidentiality; it does not license anything. Check that it, or a short sample agreement, says the sample is for evaluation only, may not be used for training, and will be deleted if no deal follows.

How is a sample different from the full export?

	Sample	Full export
When	Before a firm price	After the license agreement is signed
Size	A few workflows over a short window	Everything in the licensed scope
Purpose	Evaluation	The licensed delivery
Governed by	The NDA or sample terms	The license agreement
Payment	None	Often triggers payment on delivery or acceptance

Don’t send a full export as a sample, and don’t start the full export until the agreement fixes the scope. Some programs never ask you to assemble a sample yourself: Mercor connects to your tools and processes data once you approve the scope (Mercor). Ask each buyer how its process works before preparing anything; the guide shows where the sample sits in the whole sequence.