Insights / Data, Privacy & Security · · Updated · 9 min read
Child data minimisation in edtech: what to collect, what to refuse and who decides
A practical review of child data in edtech: define the learning purpose, assess responsibilities, limit collection and check what reaches AI providers.
Child data minimisation starts with a practical question: what information does this particular learning activity need? Define the task before adding a profile field, event or AI integration. Then document who receives each item, how long it is needed and how access ends.
This guide proposes a review method for product teams building educational services. Examples are hypothetical; they do not describe verified EduRelia features, contracts or data practices. For the group’s education context, see EduRelia. Apply the laws relevant to the actual users, organisation and processing; this is general design guidance, not an assessment of a deployment’s legal compliance.
Why children’s data needs a careful review
A learning record may contain identifiers, assessment results and personal disclosures. Consider what a child can understand about the interface, what adults can see, and what might happen when records are copied into reports or other systems. A short notice cannot compensate for unnecessary collection.
The UK ICO Children’s Code guidance on data minimisation ties collection to the individual service element the child uses. It also connects purpose, precision and retention, and calls for separate choices around additional elements. Check whether the code applies to the service under review. Use that distinction as a design question: which fields support the requested activity, and which belong to optional extras?
Start from the learning purpose
Replace a broad purpose such as “improve learning” with an observable task. The following are hypothetical examples, not product specifications:
- Choose a practice exercise: assess whether recent answers and a topic identifier are sufficient.
- Help a teacher identify a difficulty: assess whether class membership and skill-level progress are sufficient.
- Show a family a weekly summary: assess whether aggregates can replace a detailed activity history.
For each proposed field, write the task, minimum useful precision, authorised viewers and deletion trigger. If the task can work without the field, test the simpler version. An age or identity check may have separate requirements; do not assume a year group always replaces the evidence needed for that check.
Assess who decides and who processes
Where the GDPR applies, the EDPB’s controller and processor guidelines assess roles from the parties’ actual activities. A controller determines purposes and means; a processor handles data on a controller’s behalf. Contract labels help document the arrangement but do not settle it independently of the facts. A provider’s separate use can require a separate role assessment.
Map enrolment, teaching reports, service administration, product analysis and any model development as separate activities. Record who decides each purpose, which instructions apply, and whether another party determines essential aspects. Do not assign the school every responsibility simply because it signs the licence.
Review institutional licensing alongside the actual data flow and agreement. Keep institutional and direct-family arrangements distinct. A parent holding an account does not, by itself, establish every legal role or authorise every use of a child’s information.
Explain the service and assess consent requirements
Separate three questions: what will the service do, who should receive an explanation, and what authority permits the processing? A school agreement, a parent’s awareness and a child-friendly notice answer different questions. They should not be presented as interchangeable layers of consent.
Identify the applicable lawful basis and any required authorisation before collection. In the United States, assess the scope and current requirements of COPPA using the FTC’s rule materials; do not apply a GDPR role label as a substitute. Other education and privacy rules may also apply. Obtain a deployment-specific assessment where the answer is uncertain.
- Explain actual recipients, purposes and retention in language appropriate to the audience.
- Separate optional uses from the core learning activity; review marketing and product consent.
- Document the applicable basis, instructions and any consent evidence, with versions and responsible contacts.
- Make requests understandable. Distinguish withdrawing consent from requesting deletion; neither should be described as an unconditional promise that every record disappears immediately.
Fields to challenge before collecting
Use a written list of fields to question rather than a blanket claim that every education service must refuse them. Start with the learning task and assess any separate safety, accessibility, account or legal need. These are proposed review prompts, not an EduRelia policy:
- Precise location: ask why the learning activity needs it and whether a less precise alternative works.
- Contacts, photos and microphone: avoid default access; evaluate a feature’s specific need and an alternative without capture.
- Biometric templates: require a separate assessment of necessity, risk and applicable restrictions.
- Pupil address, phone and exact birth date: distinguish service needs from information already held by a school or family.
- Open profile text: consider structured choices that reduce unintended disclosures.
- Advertising identifiers and third-party trackers: inspect what integrations actually transmit and whether that collection belongs in the service.
- Social exposure and inferred traits: assess recipients and consequences, not just the visible profile fields.
Record why a challenged field is rejected, reduced or retained. A written decision should identify an owner and a review trigger, rather than quietly turning an exception into the default.
Minimising inside the data you do collect
Precision. Choose the least detailed representation that serves the documented task. Test whether a skill summary can replace intermediate answers, or a coarse duration can replace a detailed interaction trace.
Identifiers. Separate direct identifiers from working records where useful, and restrict the link between them. An ID in place of a name does not automatically make records anonymous.
Retention. Specify a reason and time or event for retaining each category. Include reports, logs, cached answers and retained copies in the retention and deletion review. Explain any required exceptions and backup handling rather than promising instant erasure.
Access and testing. Define teacher, administrator, support and engineering access by task, then test it. Use least privilege, school and class boundaries, and synthetic test records. Do not describe an untested access diagram as a proven control.
How minimisation shapes AI learning features
If a feature sends a question to an AI provider, draw the full path: browser, application, retrieval, prompt, provider, response storage and monitoring. A helpful answer may need the exercise and relevant teaching material; it does not follow that it needs an entire pupil profile.
- Input: reduce unnecessary fields before transmission and assess what free text may reveal. Redaction can miss information; test its limitations rather than promising complete removal.
- Provider: verify the product, settings, contract, retention and any training use actually applicable to the account. A general privacy statement is not proof that a deployment has a particular configuration.
- Retrieval and output: enforce authorised scope, validate the result and test tenant boundaries. Structured output alone does not establish factual correctness.
- Logs: decide what diagnostics need content, who can read it and when it expires. Avoid copying full conversations merely because the logging tool permits it.
For a hypothetical sensitive disclosure during an exercise, agree an appropriate human handling route before release. Limit unnecessary copies and explain who may receive the message. See human escalation and safety hooks for related design questions; neither link proves a specific product has implemented those controls.
What schools and parents should be able to see
Prepare a practical evidence pack for the arrangement being reviewed:
- A field inventory that explains the purpose and precision of each category.
- Recipients, service providers and relevant data-flow diagrams.
- Retention rules, deletion handling and justified exceptions.
- A role and access map, including support access and school boundaries.
- A clear route for questions and applicable access, correction or deletion requests.
Check that the notice matches observed behaviour. Follow a test record through the service and compare the result with the documentation. A feature description or contract is not enough to prove the data takes the claimed route.
Common mistakes to check for
- Collecting now, deciding later: require a defined purpose before adding an event or field.
- Unreviewed analytics: inspect payloads, identifiers and recipients rather than assuming a first-party label makes collection appropriate.
- Reusing school enrolment for growth: assess a proposed secondary use separately; a pupil account is not evidence of marketing permission.
- Open text everywhere: use structured responses when they serve the task without unnecessary disclosure.
- Copying production into test: prefer synthetic records and review every environment that can receive real data.
- Forgetting derived records: include summaries, embeddings, exports and cached explanations in the inventory.
A minimisation review for new features
Before release, record answers to these questions and name the person responsible for unresolved items:
- Which specific learning activity does the feature support?
- What does it collect, at what precision, and could the task work with less?
- Which parties decide the purposes and means, and which rules apply?
- Who receives or can retrieve the records, including providers and support staff?
- How are access, retention, requests and deletion tested?
- What happens when a child supplies sensitive information or permission changes?
Keep the decision with the feature’s source-of-truth documentation. Review it when recipients, purposes, integrations or access rules change. An unanswered critical question should trigger investigation, not a cosmetic checklist tick.
A hypothetical pupil record, from enrolment to deletion
Consider a hypothetical classroom practice tool. This is a planning example, not a description of EduRelia’s implementation.
Enrolment. The team proposes a teacher-recognisable identifier, class membership and year group. It documents why each is needed and separately assesses any age or identity checks.
First use. A test pupil answers an exercise. Inspect the request, analytics events and logs to see whether they contain only the intended data. Check what a teacher and another pupil can retrieve.
Ongoing use. Test the proposed summary and retention rule. Confirm which raw answers remain, why they remain and whether copied reports follow the same decision.
Moving class or leaving. Revoke the old access and test the new scope. Assess any transfer separately; neither indefinite retention nor automatic transfer should be assumed.
End of agreement. Follow the applicable return or deletion instructions and obligations. Test the primary records, derived copies and scheduled backup handling, documenting what remains and when it is removed.
Why this matters for a company builder
For a company builder, an education product is a useful reminder that shared technology does not settle data responsibilities. Document each business’s purposes, users and permitted recipients independently. See shared infrastructure, separate data. A policy proposed for one product is not evidence that every company follows it.
Summary
Begin with a defined learning task, challenge unnecessary fields and inspect every recipient and retained copy. Assess the school and provider’s real responsibilities rather than assigning them by default. Test access, provider settings and request handling against the documentation. Keep a record of observed results and unresolved questions. This makes minimisation a reviewable product decision instead of an unsupported promise.
Questions and answers
What is data minimisation in edtech?
Collect and use only the student data needed for a defined learning activity. Review precision, recipients, access and retention as well as the initial form fields.
Is the school always the data controller?
No. Where the GDPR applies, assess who actually determines the purposes and means for each processing activity. The school and provider may have different roles for different uses; a contract label alone does not settle the assessment.
How should a team check what reaches an AI provider?
Map inputs, retrieval, prompts, provider settings, outputs and logs. Verify the configuration and terms for the actual service, then test representative synthetic records. Do not assume redaction, no-training settings or deletion work without evidence.