Sell your company data, fully anonymised

What does it mean to sell anonymised company data?

It means a buyer gets the substance of your work and none of the people in it. An AI lab learns how your team handles a complaint, scopes a project or fixes a bug. It never learns who wrote the message, which client it was about or how to reach anyone.

We buy that kind of data from companies of about 10 people and up: Slack and Teams messages, email, documents, tickets, CRM records, code and procedures. We clean it, then license it to AI labs that train and test models. You keep your originals and keep using them. You license a copy.

Our claim is simple: what we deliver is fully anonymised. That claim only means something if it is backed by concrete steps. This page lists them, and it also says plainly where the GDPR still applies to you.

What is anonymised data under the GDPR?

Anonymised data is information that no longer relates to an identified or identifiable person. The GDPR does not apply to it. That rule sits in Recital 26 of the GDPR, which says the data protection principles should not apply to data rendered anonymous so that the person is no longer identifiable. In US spelling the same idea is called anonymized data.

The test is strict. Recital 26 asks whether anyone could identify a person using all the means reasonably likely to be used, by you or by someone else. It names cost, time and available technology as factors. Removing a name is not enough if the rest of the record still gives the person away.

Anonymisation

Anonymisation is the process of turning personal data into anonymised data, for good. Nobody, including us and including you, can trace a record back to a person afterwards. The result is outside the GDPR.

Pseudonymisation

Pseudonymisation is defined in Article 4(5) of the GDPR. It means processing data so it can no longer be linked to a person without additional information, which is kept separately and protected. Swapping "Anna de Vries" for "Employee 0042" and keeping the list that maps one to the other is pseudonymisation.

Recital 26 is clear that pseudonymised data is still personal data, because it can be linked back with that extra information. That is why we keep no list, no lookup table and no mapping of any kind.

Anonymisation vs pseudonymisation: what is the difference?

The difference is the key. Pseudonymised data has one, somewhere. Anonymised data has none, anywhere. Everything else follows from that.

A practical example: in a pseudonymised file, "Employee 0042" appears in 300 messages, so a reader can follow one person across the whole archive. In our files every name becomes the same tag, [NAME]. The conversation still makes sense, but there is no thread to pull.

Pseudonymised data has a key somewhere. Fully anonymised data has no key anywhere.

What do we remove before data is sold?

We remove every direct identifier, and then the context that could point to one person. The first part is mechanical. The second takes judgement.

• Names of employees, customers, suppliers and contacts, including nicknames and @mentions
• Email addresses, phone numbers and postal addresses
• Customer and supplier company names
• Bank account and IBAN numbers, invoice and contract numbers that lead to an account
• Customer numbers, user IDs, licence plates and other identifying numbers
• Signatures, email footers and profile details
• Context that points to one person, such as "our only tax lawyer in Leeds" or the date of someone's sick leave

Each item is replaced by a tag for its type, so the text keeps its shape. The examples below show what that looks like on everyday records.

Context is the hard part. "Our CFO" names one person in any company. A job title in a team of three, a town plus a rare event, or a quoted salary can do the same. Where a detail adds little for a buyer, we drop it. Where it matters, we make it more general, for example "a senior manager" instead of a title held by one person.

What stays out of the dataset entirely?

Some data is never part of a sale, cleaned or not. The risk is too high or the data is not yours to sell.

• Health data, including sick notes and medical details in HR files
• National ID numbers, passport numbers and similar government identifiers
• Other special categories under the GDPR, such as religion, trade union membership, sexual orientation and biometric data
• Client data you hold as a processor, for example payroll you run for clients or data you host for customers
• Privileged or client-confidential material, unless it can be fully stripped

Professions with a duty of secrecy need extra care. Our page on selling data from a law firm shows what stays in and what stays out. Templates, workflows and internal know-how are usually still valuable without a single client detail.

How do you anonymise Slack, Teams or email data?

We combine automated detection with human review. Software finds the obvious patterns at scale: names, email addresses, phone numbers, IBANs, postcodes. People then read samples of the output and look for what software misses, such as context that points to one person.

Slack and Teams

Chat data carries identity in more places than the message text. Display names, @mentions, user IDs, channel names like #client-brightwater and shared file names all get the same treatment. Emoji and timestamps stay, because they show how work flows, but exact times can be coarsened where they would identify someone. Our pillar on selling Slack, Teams and Jira data explains what labs learn from each tool.

Email

Email hides identifiers in headers, signatures, legal footers and long quoted reply chains. We clean the From, To and Cc lines, strip signatures and footers, and treat every quoted reply as text to be cleaned again. A name that was removed from the top of a thread must not survive in a quoted message at the bottom.

Tickets, CRM and documents

Tickets and CRM notes are full of customer names, account numbers and contact details. Documents add headers, author fields and file metadata. We clean the visible text and the metadata, because a file's author field can name a person just as well as its first line.

1. We agree in writing which sources are in and which are out.
2. Automated detection tags identifiers across the full set.
3. Tagged items are replaced by type tags such as [NAME] or [IBAN].
4. Reviewers read samples and look for revealing context.
5. Whatever they find is fixed across the whole set, not just in the sample.

Can anonymised data be re-identified?

Badly anonymised data can. Well anonymised data should not be. The Article 29 Working Party, where the EU's data protection authorities worked together, set out three ways it goes wrong in their Opinion 05/2014 on anonymisation techniques: singling out one person, linking records about the same person, and inferring facts about a person from other details.

Our steps map onto those three risks. Generic tags stop linking, because every name looks the same. Removing revealing context stops singling out. Leaving out health data and other special categories limits what anyone could infer.

The key matters most. If a list linking tags to real names existed, anyone who got hold of it could undo the work. So we never create one. Nobody holds it: not us, not you, not the buyer. Once the originals are deleted, there is nothing left to match against.

No method is perfect, and we would rather say so. That is why a person reviews the output, why you check a sample, and why we license only to AI labs and research teams under contract. Nothing is published.

What do you check before anything is sold?

You check a cleaned sample and sign it off. Nothing is sold without that approval. The steps below are the end of how a sale works from handover to payment.

1. Secure handover: encrypted, read-only access to the sources you approve, under NDA from day one.
2. We clean it: names, customer details and other identifiers are removed during processing.
3. You approve: you read a cleaned sample. If you spot something, we fix it across the set and you check again.
4. Originals deleted: once processing is done, we delete our copy of your originals. Yours stay where they were.
5. Paid: the money is in your account within 7 days after the data passes review.

Buyers see redacted samples only. The full package is released after they pay. We never license to your competitors.

Anonymisation does not wipe out the value. Labs want the reasoning, the back and forth and the decisions, not the names. Company data can fetch €10K–€600K, and larger or specialised datasets can earn up to hundreds of thousands of euros. Our page on why business data is worth money explains what drives the price.

Do employees have to be told?

Yes. The finished dataset is anonymised, but making it is not. To clean your messages, someone has to process them while they still contain names. That step is processing of personal data under the GDPR, and the Opinion 05/2014 above says the same: anonymisation is itself further processing.

Anonymised data is outside the GDPR. Making it is not.

In practice that means three things on your side. You need a documented assessment: a compatibility test for the new purpose, usually a legitimate interest assessment and often a DPIA. You need transparency: an updated privacy notice that tells staff and contacts that de-identified copies may be licensed for AI training, with a simple way to object. And if you have a works council, involve it early.

Our guide on whether it is legal to sell company data under the GDPR walks through each step with a checklist. The scope and terms we sign with you also give you a record for your own files.