Zum Hauptinhalt springen
ISO/IEC 27701:2025 · Annex A, Table A.1 · Control for PII controllers

A.1.4.5 PII minimization objectives

Updated on 5 min Reviewed by: Cenedril Editorial
ISO/IEC 27701:2025GDPR

An online shop builds a data warehouse for sales analysis. The development team copies the complete customer database with names, email addresses and dates of birth into the analytics environment, and into three test systems while they are at it. For the analyses, a postcode area, an age band and an internal customer number would have been enough. Four complete copies now sit where the fewest safeguards apply. A.1.4.5 requires the organization to define and document, per purpose, how far the type and amount of personal data are limited and which mechanisms, such as de-identification, achieve that objective.

Data minimization as a general principle appears in almost every privacy policy. The control turns it into objectives that can be checked: for every processing activity it should be clear which details are needed, at what level of precision, and how much link to the person remains.

What does the standard require?

The control asks for objectives and for the means of reaching them, both in writing.

  • Explain the limitation per purpose. The organization shows how the type of details and their amount stay within what the defined purpose requires. De-identification is one possible means among several.
  • Describe identifiable processing. If a purpose needs data that stays linked to the person, the organization must be able to describe that processing.
  • Measure the remaining link to the person. If de-identified data is enough for the purpose, the organization sets how strongly the data still has to be tied to the person and which mechanisms and techniques reach the objective.
  • Record the mechanisms. Which mechanisms fit depends on the processing and the systems. Those actually in use, technical system settings included, are documented, and de-identification takes effect promptly.
  • Techniques. The standard mentions removing attributes that point to individuals, generalization such as rounding, and randomization such as added noise. For details it refers to ISO/IEC 20889, and for cloud services to the identification qualifiers in ISO/IEC 19944.

In practice

Write one objective per activity. A verifiable objective names the attributes needed, their precision and their lifetime: year of birth where no full date of birth is needed, postcode where the full address can be spared, monthly totals where single transactions add no value.

Justify identifiable activities. Invoicing, shipping and customer service need the link to the person. Reporting, analysis, testing and development usually manage without it, and that is exactly where copies with real names pile up unnoticed. Note for each activity whether it processes identifiable data and why.

Target test data and analytics. Synthetic or masked data suits test environments (→ A.8.33 Test Information, A.8.11 Data Masking). Analytics environments receive their data through a loading pipeline that removes or pseudonymizes names and contact details before they arrive.

Check for re-identification. A pseudonym offers little protection if postcode, date of birth and gender sit next to it. Check whether the remaining attributes, combined with other available data, lead back to a person, and coarsen them until that is ruled out.

Document the mechanisms. Describe the pseudonymization service, the masking rules, database views and the configuration of the loading pipelines. Keep the description under version control, so that a change to the technology does not silently miss the objective.

Typical audit evidence

For A.1.4.5, auditors typically expect:

  • Minimization objectives per processing activity stating the attributes needed, their precision and the chosen level of identifiability
  • Justification for every activity that processes identifiable data
  • Description of the techniques in use and their configuration (→ Data Deletion, Masking & Leakage Prevention Policy)
  • Test data concept for development and test environments
  • Assessment of the re-identification risk for de-identified data sets, as part of A.1.2.6 Privacy impact assessment where the risk is high

KPI

% of processing activities with defined and enforced PII minimization rules

Defined means: the objective is set in writing for the activity. Enforced means: a technical or organizational mechanism ensures it is met, and this can be shown. The target is 100%. A rule that exists only on paper does not count for this figure.

Supporting KPIs:

  • Share of test environments without real personal data
  • Share of analytics data sets with a documented re-identification risk assessment
  • Number of copies of production customer data outside production

A.1.4.5 makes data minimization under Article 5(1)(c) GDPR concrete, together with storage limitation under Article 5(1)(e), which allows storage in identifiable form only as long as the purpose requires. Article 25(1) explicitly names pseudonymization as an example of a design measure that puts principles such as data minimization into effect. Article 4(5) defines pseudonymization. According to Recital 26, pseudonymized data remains personal data, while anonymous information falls outside the Regulation. For research and statistics, Article 89(1) requires safeguards, which may include pseudonymization (→ GDPR General Data Protection Regulation).

A.1.4.5 turns the principle of minimization into measurable objectives:

Sources

Frequently asked questions

Is pseudonymization the same as anonymization?

No, and mixing them up is costly. With pseudonymization, a key remains that allows the person to be identified again, so the data stays personal data. Data is only anonymous once nobody can link it to a person with reasonable effort, taking other available data into account.

Do I have to de-identify for every processing activity?

No. Many purposes need the link to the person, such as invoicing, delivery or answering a request. In that case you describe the processing and justify the link. De-identification is the obvious tool wherever analysis, testing or statistics work without names.

Which techniques are available?

Removing identifying attributes, generalization such as rounding or age bands, randomization such as adding noise, and pseudonymization with a separately held key. ISO/IEC 20889 gives an overview of the techniques. Which one fits depends on the purpose and on the systems involved.

No Cookies!

This wiki collects nothing, bakes nothing and leaves nothing behind. There's nothing to consent to. Privacy doesn't get better than this.