AI parent enquiry triage.

What parent enquiry triage covers

Parent enquiry triage reads each message arriving in the office inbox or parent app, gives it a category and urgency, names the staff member who owns it, and suggests a reply from approved templates. Typical categories are absences, payments and donations, trips and permissions, uniform and stationery, enrolment, pastoral concerns, complaints and formal information requests. A staff member sends every reply. It is one of the workflows in our guide to AI for education administration.

Why the under-13 guidance shapes the whole design

The Ministry of Education’s generative AI guidance is blunt: “Never send any AI tool the personal information of anyone under the age of 13.” Almost every parent message names a child. At a primary or intermediate school, a naive build breaks that guidance on the first email.

A defensible design removes learner details before the model sees anything:

1. Match against the roll inside school systems. A deterministic step compares the message with the student management system roll and replaces learner names, National Student Numbers, birth dates and room numbers with tokens such as [CHILD_1].
2. Send only redacted text. The model classifies “[CHILD_1] is unwell and won’t be in today” without knowing who the child is.
3. Re-identify after the model. The school’s own system maps tokens back, so the owner sees the real name.
4. Record the decision. The board signs off a privacy impact assessment, which the Privacy Commissioner expects before AI tools are used.

Redaction is imperfect. Nicknames, siblings and health details slip through, so sample for misses every week. Processing also leaves the country: OpenAI’s data controls page lists Australian storage but no Australian processing, and the school stays responsible for what its provider does with the text.

The triage pipeline at a glance

Inputs

Office inbox and parent app messages, the current roll from the student management system (KAMAR, Hero or similar), approved reply templates, and a routing table of staff owners.

Systems

Microsoft 365 or Google Workspace mail, the student management system, a redaction service running on school-controlled infrastructure, a business-tier model with no training on inputs, and a review queue.

Steps

Redact, classify category and urgency, detect formal requests and safety concerns, route to an owner, suggest a template reply, re-identify for the owner, log.

Human review

Office staff confirm every category in the first four weeks and send every reply; anything flagged as a concern is read by a person within an hour.

Outputs

A sorted queue with owner and due time, draft replies, and a register of formal requests with received dates.

Measures

Correct category rate, time from arrival to owner, redaction misses per 100 messages, and formal requests answered within time.

Risks

A redaction miss, a safety concern routed as routine, disclosure to an adult who is not entitled, and missed deadlines.

Illustrative worked example: a 600-student intermediate school

Illustrative example: a 600-student intermediate school (all learners under 13) receives 220 free-text parent messages a week in term. Assumptions: the office spends 3 minutes reading and routing each one today; with suggestions it spends 1 minute checking each; a weekly redaction sample of 25 messages takes 25 minutes. Synthetic figures.

Routing time falls from 660 to 220 minutes, and the redaction sample adds 25, so about 6.9 hours a week return to the office. If most absences already arrive through a structured app form, the saving roughly halves, which is worth knowing before you build.

Requests that start a legal clock

School boards are organisations under the Official Information Act 1982, through Part 2 of Schedule 1 of the Ombudsmen Act 1975, so an OIA request needs a decision within 20 working days. A request for personal information under the Privacy Act 2020 has the same 20 working day limit in section 44. Triage should log the received date the moment either is detected.

The harder rule is about who gets what. The Privacy Commissioner’s education guidance states that “the Privacy Act does not provide parents with an automatic right to information about their children”, outside specific duties such as reporting on progress. Separated parents and protection orders make this real, so drafts must never contain learner information. Every school also needs a privacy officer, who owns these requests.

What stays with staff

• Sending every reply that mentions a learner or a decision.
• Attendance coding, using the Ministry’s 15 attendance codes; the board must take all reasonable steps to ensure attendance, and section 45 of the Education and Training Act 2020 lets the principal, not software, exempt a justified absence of up to five school days.
• Reading pastoral and safety concerns, and deciding what to share with which adult.

Where triage fails

Cost drivers and first steps

The cost sits in the redaction layer, the student management system connection and the privacy assessment, not model fees. Start with four weeks in shadow mode, where AI suggests and staff route as normal, then compare. Our guide to protecting customer information when using AI and our note on human review queues cover the controls, and AI customer support systems is the build service. Related workflows: the staff policy assistant and administrative document workflows.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.