Home › Services › Data De-identification

Data de-identification and rights clearance

For data owners and data brokers.

Prepared by the Buckingham Capital Consulting data services team · 26 October 2026

Data de-identification is the removal or replacement of everything in a dataset that identifies a person, so the data can be shared, licensed or sold.

Who we work with

We work across the whole market, for any company that holds data and any company that buys it.

  • Data owners: companies that want to license or sell data they hold, and need it cleared and de-identified first.
  • Data brokers and marketplaces: companies that buy data from many owners and need every source cleared to one standard.
  • Data buyers: teams that need datasets to a specification. We source and supply them, cleared to the same standard. AI labs can read about this on our AI training data page.

What data we handle

We handle structured data, such as spreadsheets, databases, transaction tables and logs, and unstructured data, such as emails, chat messages, documents, call transcripts and tickets. Unstructured data means free text written by people. It carries the most hidden identifiers, so it receives the most attention.

We work in the formats the owner already uses, and deliver in the formats the buyer needs, such as CSV, JSON Lines (one record per line) or Parquet (a compressed table format).

What you receive

Every release carries the records that both the owner and the buyer need to approve it.

DeliverableWhat it containsWho relies on it
Clearance recordOwnership, authority, contracts, confidentiality, copyright and database right, earlier licences and the permitted usesOwner and buyer legal teams
Privacy assessmentThe lawful basis for the de-identification work, the exclusions and the analysis under each law that appliesPrivacy teams on both sides
De-identification reportMethods used, where processing ran, how the key was handled, test results and remaining riskSecurity, privacy and data teams
Exclusion recordEach category removed, why and in what volumeBuyer privacy and data teams
Data sheetWhat the dataset contains, its source, date range, structure, volumes and known limitsBuyers and their governance teams
Licence termsScope of use, term, exclusivity, onward rights, retention, deletion and the ban on re-identificationBoth parties
Release approvalOwner sign-off for each data packageThe owner's board and the buyer

See also: Data broker registration · How to sell data to AI companies · Anonymisation vs pseudonymisation · HIPAA de-identification · AI data licensing deals · AI data licensing agreement

Get in touch

Tell us where you are and where you want to be.

[email protected]
+44 7563 115980 (phone and WhatsApp)

Further detail

Background

We run it on the data owner's own premises, so raw records never leave its systems, and we clear the rights and the privacy position at the same time. Each release comes with a clearance record, a de-identification report and test results that a buyer's legal and privacy teams can rely on.

On this page

We clear the rights and the privacy position, de-identify datasets on the owner's own premises and test the result before anything is released. We also source and supply cleared datasets from companies that want to sell their data.

Rights clearance

Rights clearance confirms that the owner is entitled to license the data for the buyer's intended use. The buyer's legal team needs it in writing before a deal can close.

  • Ownership and authority: who owns the data, and who can sign for the company.
  • Customer and supplier contracts: terms that restrict how information received from others may be used.
  • Confidentiality: obligations owed to clients, partners and staff.
  • Copyright and database right: who owns the text and documents, and whether the dataset is a protected database. In the UK, database right protects a database where there was a substantial investment in obtaining, checking or presenting its contents, and lasts 15 years from the end of the year it was completed or first published.
  • Earlier licences and exclusivity: anything already granted that a new licence must respect.

Where a restriction applies, the material is removed, consent is obtained or the permitted use is narrowed. The clearance record lists each check, its result and the uses the owner can grant.

Privacy clearance

De-identification is itself a use of personal data. UK guidance confirms that anonymising personal data counts as processing, so the owner needs a lawful basis (a legal ground the law allows for using personal data), a defined purpose and notice to people. We document all three before work starts.

We then set what must be excluded altogether: special category data (information revealing racial or ethnic origin, political opinions, religious or philosophical beliefs or trade union membership, genetic data, biometric data used to identify a person, health, and sex life or sexual orientation), criminal offence data, identity numbers, account credentials and precise location.

In the US, under the California Consumer Privacy Act (CCPA), information is deidentified only if the business takes reasonable measures to stop it being linked to a consumer, publicly commits not to re-identify it, and binds every recipient by contract to do the same. Health information covered by HIPAA (the US federal health privacy law) is de-identified by one of two methods: removing 18 listed types of identifier with no actual knowledge that the rest could identify anyone, or a documented expert finding that the risk is very small.

On-premises de-identification

On-premises (on-prem) means the work runs on the owner's own servers or inside its own cloud account. Our process is set up there, and only approved, de-identified output is released.

  1. Inventory. We map every source, field and file type, and flag where identifiers are likely to appear.
  2. Direct identifiers. Names, email addresses, phone numbers, account and reference numbers are replaced with consistent placeholders. The same person or company gets the same placeholder in every source, so links between records survive.
  3. Indirect identifiers. Rare values and combinations are generalised. An exact date becomes a month or a year, a postcode becomes a region, and a small group is merged into a larger one.
  4. Free text. Messages and documents are reviewed for identifying details written in ordinary sentences.
  5. Key control. The table linking placeholders to real identities stays with the owner and never leaves with the data. Where the owner has no need for it, it is destroyed.
  6. Staging. Output is held in a separate area inside the owner's environment until it passes testing and the owner approves its release.

How results are tested

UK guidance sets the motivated intruder test: could a reasonably competent person, using the internet, public records and ordinary investigative methods, identify someone if they set out to?

  • Sample review: trained reviewers read samples drawn from every source, with extra weight on free text.
  • Uniqueness checks: we look for records that stand out because of a rare combination of details, and generalise them further.
  • Linkage tests: we try to match records against public sources such as company registers, news reports and social media profiles.
  • Whose hands: we assess the risk in the hands of the specific buyer, taking account of other information it holds or could obtain.
  • Re-testing: results are reviewed when the data, the buyer or the public information available changes.

Data is released only when the risk of identification is sufficiently remote. Results and the remaining risk are written up in the de-identification report.

Anonymised, pseudonymised and de-identified

These words are often used loosely. In law they have precise and different effects.

TermWhat it meansLegal effect
Anonymous (UK and EU)No one can be identified by any means reasonably likely to be used (GDPR Recital 26)Data protection law does not apply to it
Pseudonymised (UK and EU)Identifiers are replaced, but people could be identified using additional information kept separately (GDPR Article 4(5))It remains personal data, and the law applies in full
Pseudonymised data held by a recipient (EU)In September 2025 the EU Court of Justice held that pseudonymised data is not necessarily personal data for a recipient who cannot re-identify anyoneStatus is judged in the hands of each holder, and the original owner's duties continue
De-identified (UK offence)Re-identifying it without the consent of the organisation that de-identified it is an offence (Data Protection Act 2018, section 171)A licence's ban on re-identification is backed by the criminal law
Deidentified (California)Cannot reasonably be linked to a consumer, backed by a public commitment and contracts with recipients (CCPA section 1798.140(m))It falls outside the CCPA's definition of personal information
De-identified (US health)Either 18 types of identifier removed with no actual knowledge of identifiability, or an expert finding that the risk is very small (HIPAA Privacy Rule, 45 CFR 164.514(b))It is no longer protected by the HIPAA Privacy Rule

How an engagement runs

Each engagement follows the same six steps, scaled to the size of the dataset.

StepWhat happensOutput
1. ScopeThe dataset, the intended buyer or use, and the owner's systems and obligationsWritten scope
2. ClearRights and privacy reviewed, restrictions resolved and permitted uses agreedClearance record and privacy assessment
3. Set upDe-identification set up inside the owner's environment and run on a pilot set of recordsPilot output for review
4. TestSample review, uniqueness and linkage tests, with adjustments until the data passesDe-identification report
5. ReleaseThe owner approves each package, and data moves to the buyer in the agreed formatsRelease approval, data sheet and exclusion record
6. MaintainNew records processed on a schedule, with tests repeated when anything material changesVersioned releases with change notes

Selling data and sourcing datasets

We also source and supply datasets. Companies that want to sell or license their data engage us to prepare it and bring it to buyers. Data brokers and AI teams engage us to find data that matches a specification, from companies that have chosen to license it.

For a data owner, this means one team handles clearance, de-identification and the sale. For a buyer, it means every dataset arrives with the same records, whichever company it came from.

How the commercial terms work

  • Licence, not transfer: the owner keeps ownership of its data and grants defined rights to use it.
  • Exclusive or non-exclusive: an exclusive licence reserves data for one buyer, a field of use or a period. A non-exclusive licence lets the owner license the same data more than once.
  • Term and territory: a fixed period, and the countries in which the data may be used.
  • Permitted use: analysis, training, fine-tuning, evaluation or resale, each named separately.
  • Onward rights: whether a broker may sublicense the data, and on what conditions.
  • Retention and deletion: what happens to copies at the end of the term.
  • Re-identification: a contractual ban on any attempt to identify people, passed down to every recipient.
  • Pricing: set by volume, exclusivity, term and refresh frequency.

Questions

What is data de-identification?

The removal or replacement of everything in a dataset that identifies a person, so the data can be shared, licensed or sold while staying useful.

Is de-identified data still personal data?

It depends on how well it is done. Data that no one can reasonably use to identify a person is anonymous and outside UK and EU data protection law. Data that could be re-linked using separately held information remains personal data.

What is the difference between anonymisation and pseudonymisation?

Anonymisation removes the link between the data and the person. Pseudonymisation replaces identifiers with codes, but the link can be restored with information held separately. UK guidance describes anonymisation as reducing the personal data you hold, and pseudonymisation as reducing the risks of the personal data you hold.

Does raw data leave our premises?

No. De-identification runs on your own servers or in your own cloud account. Only output that has passed testing, and that you have approved, is released.

How do you test that data cannot be re-identified?

We apply the motivated intruder test, review samples, look for records that stand out, try to link records to public information and assess the risk in the buyer's hands.

Can de-identified data be licensed or sold lawfully?

Yes, where it is truly anonymous and the owner has the rights to license it. Contracts and copyright still apply to anonymous data, so we clear the rights as well as the privacy position.

Do you act for the data owner or the buyer?

Either, agreed for each engagement. The clearance record and the de-identification report are prepared so that both sides can rely on them.

Can you review de-identification we have already done?

Yes. We test an existing dataset against the same standards, report the remaining risk and set out what needs to change before release.

Can you source and supply datasets as well as clear them?

Yes. We source and supply company datasets to a buyer's specification, each from a company that has chosen to license its own data, cleared to the same standard.

How do you handle exclusivity and earlier licences?

Existing licences, exclusivity and earlier sales are checked during rights clearance, so that a new licence never conflicts with one already granted.

Sources

From the blog

All articles