A service provider sends newsletters for around two hundred customers. The product team has an idea: the open rates of all customers could produce a benchmark per industry, and the addresses could train an in-house spam filter. Technically, everything is in place. None of it has been agreed. A.2.2.3 requires you to use data you process for a customer only for the purposes set out in their documented instructions, and to let the customer verify that boundary.
Purpose limitation is the core of processing on someone else’s behalf. As soon as a processor pursues goals of its own with the data, it leaves its role. It becomes responsible for that processing itself, needs a lawful basis of its own and loses the trust of the customer who handed over the data.
What does the standard require?
The control is the processor’s counterpart to A.1.2.2. The controller sets the purposes, and the processor keeps to them.
- Purposes from the instructions. Data from an assignment is used solely for the purposes the customer has laid down in documented form.
- Objective and time frame in the contract. At a minimum, the contract describes the result the service is meant to achieve and the period in which it does so. Further content is possible.
- Room for choosing the method. The standard accepts that there can be technical reasons to choose how the processing is done without the customer instructing every detail. Its example: computing or network capacity is allocated according to certain characteristics of the data subject so that resources are used efficiently. This latitude stays tied to the customer’s general requirements and purpose.
- Verifiable for the customer. The customer should be able to check that purpose specification and purpose limitation are respected. This includes your subcontractors: they, too, process nothing for other purposes.
In practice
Collect the instructions in one place. For each customer, record where their instructions come from: contract, service description, annexes, written individual instructions, settings in the product. Anyone planning a new feature internally finds the limit of what is allowed there.
Name objective and term in concrete terms. Describe in the service description what the service is for, precisely enough that later processing can be measured against it. Give the assignment an end. Without a term, there is no point at which data has to be returned or deleted.
Check internal projects in advance. Decide that every new analysis of customer data, whether for statistics, quality assurance or training models, is checked by the data protection officer before it starts. The outcome is one of three: covered by the assignment, possible with the customer’s consent, or not permitted.
Separate your own data from customer data. Billing data, contract data and details of contact persons are usually processed by you as a controller. Your customers’ content data is processed on their behalf. Separate storage locations and clear labels keep the two worlds from blurring.
Bring subcontractors in. Pass purpose limitation on to every subcontractor by contract and have them confirm that they use the data neither for their own analyses nor to train their own services. With cloud services that include AI features in particular, the terms of use deserve a close look.
Make checks possible. Give the customer the means to verify compliance: audit reports, logs of access, an overview of subcontractors and the right to run their own audits.
Typical audit evidence
For A.2.2.3, auditors typically expect:
- Contracts and service descriptions stating the objective and term of the service
- Documented individual instructions from customers with date and confirmation
- Approval procedure for new internal uses of customer data with the results of the checks (→ Change Log)
- Contracts with subcontractors that pass purpose limitation on (→ Supplier Register)
- Audit reports or customer audits showing that purpose limitation is respected
KPI
% of processing purposes formally documented and reviewed
What counts are the purposes for which you process customer data. Formally documented means: the purpose can be traced to an instruction or a contract clause. Reviewed means: it was compared with actual use within the last year. The target is 100%.
Supporting KPIs:
- Number of internal projects involving customer data that were checked before they started
- Share of subcontractor contracts with explicit purpose limitation
- Number of customer audits per year without a finding on purpose limitation
Link to the GDPR
Under Article 28(3)(a) GDPR, the processor processes personal data only on documented instructions from the controller. Article 29 confirms this for everyone with access to the data under the processor’s authority, and Article 32(4) calls for suitable arrangements. A processor that determines purposes and means itself is treated as a controller for that processing under Article 28(10), with all the duties and liability that follow. The principle of purpose limitation in Article 5(1)(b) is addressed to the controller. A.2.2.3 makes sure you do not undermine it from your position (→ GDPR General Data Protection Regulation).
Related controls
A.2.2.3 enforces the customer’s purposes inside your organization:
- A.1.2.2 Identify and document purpose: the controller’s side, where the purposes are set.
- A.2.2.2 Customer agreement: the contract that states the objective and term of the service.
- A.2.2.4 Marketing and advertising use: a specific form of purpose limitation for advertising.
- A.2.2.5 Infringing instruction: what to do when an instruction itself breaks the law.
- A.1.4.3 Limit processing: the controller’s limiting of processing to the purpose.
Sources
- ISO/IEC 27701:2025, Annex A, Table A.2, A.2.2.3, and Annex B, B.2.2.3: Organization’s purposes
- Regulation (EU) 2016/679 (GDPR), Articles 5, 28, 29 and 32
- ISO/IEC 27701: Privacy Information Management System: structure of the standard and roles