Start with the distinction that changes the obligations

Pseudonymisation replaces or separates identifying details so that attribution requires additional information. It can be a valuable safeguard, but personal data that can be attributed through that additional information remains within data protection law. Anonymous information is different: identification must be sufficiently remote when assessed against the relevant context and the means reasonably likely to be used. A reassuring label is not the test.

For a practical project brief, describe the intended output and audience. Are analysts working within your organisation, selected partners receiving a controlled file, or anyone able to download a public dataset? Those are different release environments. Write down which information you retain, which the recipient can access and what outside information may be available. That description should come before choosing a technical method.

Removing direct identifiers is only the beginning

Names and email addresses are obvious identifiers, but combinations can matter just as much. Rare characteristics, precise times, locations or a distinctive sequence of transactions can enable identification. Information about an identifiable person can remain personal even if the organisation does not know their civil name. The ICO’s guidance addresses indirect identification and singling out, not just the visibility of a name column.

Our suggested first test is to look for uniqueness rather than obvious labels. Which rows describe only one person, team or household? Could a colleague recognise an event? Would a public announcement or social-media post make a record recognisable? Do not use these questions to perform unauthorised re-identification. Use them within an approved assessment to decide what needs generalising, suppressing or protecting.

Choose techniques for the intended use

Aggregation, grouping values, reducing precision and suppressing unusual records can reduce exposure. Pseudonymous identifiers can support analysis without routinely exposing identities. None is automatically sufficient in every situation. A small group average may reveal something sensitive about one person, while repeated releases can allow comparisons that undo the protection achieved in an individual file.

We recommend documenting the analysis that the business genuinely needs before retaining fine detail. A report about regional demand may not need exact addresses; a staffing trend may not need individual daily histories. Test whether useful conclusions survive less precise data. If a purpose requires identifiable detail, be honest about that and apply the relevant controls instead of describing the output as anonymous for convenience.

Test the release environment, not only the file

The ICO presents the motivated-intruder approach as a useful way to examine identifiability. The assessment considers practical possibilities and reasonably likely means, rather than requiring proof that identification has already happened. Relevant considerations include available information, capability, motivation, time and cost. This is a structured risk assessment, not a challenge to invent an implausible attack or dismiss every possible link.

For implementation, involve someone who understands the data’s subject matter as well as the technical method. A specialist analyst may miss an obvious local clue that an operational colleague recognises immediately. Keep the test authorised and bounded. Record the assumptions, limitations and outcome, and escalate uncertain high-consequence releases rather than treating a lack of immediate success as proof of anonymity.

Separate keys and control re-identification

Pseudonymisation works best when the additional identifying information is genuinely protected. Keeping the lookup table beside the coded dataset with identical permissions defeats much of the practical benefit. Hashing, tokenisation and encryption have different properties; an unqualified claim that a hash is irreversible does not establish that a record cannot be linked back to a person.

Practical controls should specify who may re-identify, for what purpose, using which approval and with what log. Separate routine analysis from exceptional identity recovery. Review exports, backups and support access as well as the main database. The legal position can depend on the holder and context, so avoid announcing that every coded dataset is anonymous to everyone simply because one team lacks the key.

The transformation itself still needs governance

Creating an anonymous output from personal data involves processing the original information. The relevant duties apply to that activity; a proposed anonymous end product does not retrospectively remove them. The business should assess purpose, lawfulness, fairness, transparency, minimisation and security at the transformation stage. Effective anonymisation can reduce subsequent data protection exposure, but it is not a universal exemption from other legal or contractual obligations.

Our recommended workflow has a release owner, a documented technique, a peer check and an approved destination. Keep raw inputs out of the release folder. Test the final export, not merely a sample produced earlier in development. Confirm that hidden sheets, comments, filenames, metadata and attachments do not reintroduce identifiers. Administrative mistakes can undermine sophisticated statistical work.

Reassess when the world around the data changes

Identifiability can change as more information becomes available or technology develops. The ICO describes an ongoing need to consider whether the likelihood of identification has changed. This is particularly important for repeated releases and information kept publicly accessible. Contracts restricting recipients can be useful controls, but an agreement alone does not transform identifiable information into anonymous information.

Set a review trigger for new datasets, changed recipients, broader access and new intended uses. Where uncertainty remains, manage the material as personal data while obtaining appropriate specialist assessment. A defensible conclusion explains why the chosen controls fit the use. It does not depend on a certificate from a software tool or the assumption that nobody would be interested enough to connect the dots.

Sources and further reading

Primary guidance behind this briefing. Requirements depend on the facts; check the current guidance before making a decision.

This is general guidance, not a legal opinion on your circumstances or a guarantee of compliance. An individual assessment may be needed, particularly where sensitive information, disputed rights or regulatory action is involved.