A data deal starts with a mutual NDA, signed before anyone sees a single record, so both sides can talk about what exists and what it might sell for. After that, a data room holds the scrubbed sample, the scrub report and the paperwork a buyer needs to judge the dataset, and the full dataset stays with you until a license is signed.
This is general information about how deals are usually run, including how we run ours. It is not legal advice, so have a lawyer read any NDA before you sign it.
Why the NDA comes first
The first call about a data sale already involves confidential information. You describe which systems you use, how many years of records you keep, roughly how many customers or projects those records cover and what your contracts say. A buyer, or a broker acting for buyers, describes what it is short of, which can reveal its training plans. Both sides want that conversation protected.
A mutual NDA covers both directions. You agree not to repeat what you learn about the buyer's needs, and the buyer agrees not to use what it learns about your data for anything except evaluating the deal. One-way NDAs that only protect the buyer are common in procurement, and you should ask for the mutual version.
In our process the mutual NDA goes out the same day as the first 30-minute call. The call qualifies the deal on rights first, then systems, volume, years of history and who at your company signs. If the rights answer is clearly no, we stop before anyone signs anything.
What a data-deal NDA should say
Most NDAs are templates, and a template written for software procurement can miss what matters in a data deal. Read for these points:
- What counts as confidential. It should include the sample, the scrub report, field lists and schemas, and anything derived from them, such as statistics a buyer computes on your sample.
- Permitted use. The receiving party may use the information only to evaluate a possible license. Training a model on the sample, even a test model, should need a signed license.
- Deletion. If the deal does not go ahead, the buyer deletes the sample and anything derived from it, and confirms in writing.
- No re-identification. The buyer agrees not to try to link scrubbed records back to people or companies. Our buyer licenses carry the same ban, and putting it in the NDA covers the evaluation period too.
- Term. Confidentiality should outlast the evaluation. Data does not go stale the way a pitch deck does.
- Who is bound. Contractors and advisers who see the sample should be bound by the same terms.
The NDA does not replace the license. It protects the evaluation. The license terms cover permitted uses, length, exclusivity and deletion once money changes hands.
What goes into the data room
A data room is a controlled place where a buyer reviews material about the dataset before making an offer. In a company sale the data room holds contracts and financials. In a data sale it holds evidence that the dataset is real, legal to license and useful.
A buyer reviewing a dataset for AI training typically wants to see:
| Item | What it answers |
|---|---|
| A representative scrubbed sample | Does the data look like what was promised? |
| The scrub report | What was removed or generalized, counted by type, and the re-identification test result |
| Field and source notes | Which system, which version, which date range, and what each field means |
| A rights statement | Can the seller license this, and on what basis? |
| Volume and history figures | How many records, over how many years, growing by how much each month |
| Proposed license terms | Uses, length, exclusivity and refresh pricing |
A good data room answers the buyer's diligence questions before they are asked. Every gap in it turns into an email thread and a week of delay.
The sample is the only data that leaves early
Sellers worry most about the sample, and the worry is reasonable. A sample is real data leaving the building before a contract exists.
Three things keep that risk small. The sample is scrubbed before anyone outside your company looks at it. It is small: our readiness report works from a sample of about 1,000 rows. It is covered by the NDA, including the deletion clause.
Our privacy rule is strict about this step. Through the SDK, raw data never leaves your machine; only scrubbed output and the scrub report are uploaded. The one exception is the readiness report sample, which you can upload in the portal under NDA. The server scrubs it, stores only the scrubbed output and the report, and deletes the raw file. The readiness report you get back says what is sellable, what has to go and whether any record could still be singled out.
Who sees what, and when
A deal moves through stages, and access widens only as commitment grows. Our stages run in this order: applied, NDA sent, NDA signed, rights review, sample received, readiness report, buyer interest, offer, licensed, paid and recurring.
Before the NDA, a buyer hears only the industry, the systems and rough volumes. After the NDA and the rights review, a buyer can see the scrubbed sample and the scrub report. The full dataset moves only after a license is signed, and you approve every buyer before that happens. You can rule out any company, including your competitors.
Each stage change is logged with a date in your deal room timeline, so you can always see where the deal stands and who owns the next step.
Common mistakes
- Sending a raw export "just to show them." Once raw data is out, no NDA can pull it back. Scrub first, always.
- Signing the buyer's paper unread. Some buyer NDAs allow "internal research use" of evaluation material, which can include training. Strike it.
- Skipping the deletion confirmation. Ask for written confirmation when a buyer passes.
- Letting the sample grow. A buyer asking for 50,000 rows to "evaluate properly" is asking for a free dataset. A representative sample plus clear field notes should be enough to make an offer.
- No record of who saw what. Keep a list of every party that received the sample and when. It makes deletion requests and audits simple.
How this fits the rest of the deal
The NDA and data room come early, and they set the pace for everything after them. A clean data room shortens how long a data deal takes, because the buyer can make an offer without a long back-and-forth. Preparing the sample well is covered in how to prepare a data export.
Once a license is signed, the same discipline carries into recurring deliveries. Each monthly batch is scrubbed on your machine and uploaded with its own scrub report, so the buyer's diligence continues month by month. Recurring data revenue explains how those refresh batches pay.
See what your records could earn
Before you sign anything, get a rough number. Our calculator gives a range from your industry, systems, years of records and team size, and we confirm it on a 30-minute call. If you want the full picture of how a sale works, start at selling data to AI companies.
Frequently asked questions
Do I need an NDA before the first call?
No. The first call covers industry, systems and rights in general terms, and we send the mutual NDA the same day. Nothing that identifies your customers or employees should be shared before it is signed.
Can a buyer train on my sample during evaluation?
Not under a well-drafted NDA. The permitted use should be limited to evaluating a possible license, and any training should require a signed license. Check the buyer's paper for "internal research" language that would allow it.
What happens to the sample if no deal is signed?
The NDA should require the buyer to delete the sample and anything derived from it, and to confirm the deletion in writing. Keep a list of every party that received it so you can ask each one.
Does the NDA cover the monthly batches after a license?
No. Once a license is signed, the license governs use, deletion and the ban on re-identification for the history and every refresh batch. The NDA covers the evaluation period before that.