# AI administrative document workflows.

URL: https://levystreet.com/industries/education/administrative-document-workflows/
Updated: 2026-10-19

AI administrative document workflows for NZ schools: enrolment packs extracted, checked and filed under the Public Records Act, with staff confirming records.

## Short answer

AI administrative document workflows read the forms and evidence a school office receives, such as enrolment packs and proof of address, extract the fields for a staff member to confirm, and file each document under the right records class. In New Zealand schools they must follow the Public Records Act 2005 and the current care records protection order, so the system never deletes or alters a record itself.

- Start with enrolment packs: fixed fields, peak volumes and clear checks against the enrolment scheme.
- A temporary protection order means schools must not dispose of, alter or destroy care records.
- Primary schools no longer had to keep immunisation registers from 8 January 2026; old records are kept 7 years.
- AI extracts and flags; staff verify eligibility evidence and the principal decides zone and ballot questions.

## Which school documents suit AI first

Enrolment packs are the best first document workflow for most schools. They arrive in predictable peaks, have fixed fields, and each one is checked against the same list: eligibility evidence, proof of address, previous school, contacts and any custody or parenting orders. Trip and activity consent forms come next. Health plans and learning support reports come last, if at all. This page sits within our guide to [AI for education administration](/industries/education/).

## How an enrolment pack is processed


- **Inputs**: Online or scanned enrolment forms, eligibility evidence such as passports, birth certificates or visas, proof of residential address, previous school details, any court orders, the school's enrolment checklist and its home zone description.
- **Systems**: A form tool or scanner, a vision or OCR extraction step, the student management system (KAMAR, Hero or similar), ENROL (the Ministry's enrolment register), and the school's records store with retention classes.
- **Steps**: Split each pack into documents, extract fields with page references, check completeness against the checklist, compare the address with the zone description, draft a record in the student management system, file each document with its records class, and log every action.
- **Human review**: The enrolment officer checks evidence, confirms every pre-filled field before saving and completes ENROL; the principal decides zone disputes and out-of-zone ballots; a senior staff member records any court orders.
- **Outputs**: Draft student records, a missing-items email for the family to approve, a zone check flag, and filed documents with class and date.
- **Measures**: Fields corrected per pack, packs completed without a second chase, days from receipt to completed enrolment, and filing errors found in audits.
- **Risks**: Misread names or birth dates, a wrong zone call, court orders visible to too many staff, and any automated deletion or overwrite.


## Illustrative worked example: Year 9 intake at a large secondary school

Illustrative example: a 1,100-student secondary school processes 260 Year 9 packs before Term 1 and 80 mid-year enrolments, about 6 documents each. Assumptions: manual processing takes 25 minutes a pack (retyping, zone check, filing, chasing); with pre-filled drafts it takes 10 minutes; accuracy sampling takes one hour a week for 12 peak weeks. All figures are synthetic.

| Measure | Manual | With AI drafts |
|---|---|---|
| Staff time for 340 packs | about 142 hours | about 57 hours |
| Accuracy sampling | none | 12 hours |
| Net staff time | about 142 hours | about 69 hours |
| Extraction fees | none | about US$15 a year |

The fee line assumes 2,500 input tokens per page image and 1,500 output tokens per pack, priced at OpenAI's gpt-6.1-sol [list price](https://developers.openai.com/api/docs/pricing) of US$2 per million input and US$10 per million output tokens on 19 October 2026. The 73 hours returned are modest across a year, but they land in the weeks before Term 1, when the office is busiest.

## Records rules that change the build

**Retention is set by the Public Records Act.** The Ministry's [records page](https://www.education.govt.nz/education-professionals/schools-year-0-13/administration-and-management/managing-school-records) says state and state-integrated schools must manage records under the [Public Records Act 2005](https://www.legislation.govt.nz/act/public/2005/40/en/latest/), and need the Chief Archivist's approval before disposing of anything. The register of admission, progress and withdrawal is kept permanently. Every filed document therefore needs a records class, not just a folder.

**Nothing may be altered or destroyed for now.** The same page describes a temporary protection order from Archives New Zealand: care records, which include all records held about students and staff, "must not be disposed of, altered or destroyed". In practice the pipeline writes new derived records and never overwrites a scan, merges duplicates or runs clean-up jobs.

**Eligibility evidence is checked once.** The Ministry's [eligibility guidance](https://www.education.govt.nz/education-professionals/schools-year-0-13/enrolment/eligibility-enrol-new-zealand-schools) lets schools ask for documented evidence of a student's right to enrol as a domestic student, and says there is no ongoing requirement to copy and retain that proof once the student is recorded in ENROL. Copies already held are still covered by the protection order.

**Zones and ballots are statutory.** Boards run [enrolment schemes](https://www.education.govt.nz/education-professionals/schools-year-0-13/enrolment/enrolment-schemes) under sections 71–75 and Schedule 20 of the Education and Training Act 2020, and out-of-zone applicants are accepted in an order the Act defines. Schools cannot charge a fee or ask for a donation for any part of the out-of-zone pre-enrolment process, so a "missing items" email must never imply one.

**Checklists go stale.** Primary schools stopped being required to keep [immunisation registers](https://www.education.govt.nz/education-professionals/schools-year-0-13/administration-and-management/immunisation-register-requirements-school-and-kura) on 8 January 2026. A pipeline still demanding certificates is chasing families for something the law no longer requires, while existing records stay for 7 years.

## What staff verify and decide

The system proposes; staff decide. A person compares evidence with the extracted fields, decides eligibility questions, confirms zone status, enters ENROL, and records who may and may not collect a child. Court orders go to a restricted folder that only named staff can open.

## Where extraction goes wrong

| Failure | Example | Control |
|---|---|---|
| Name or date misread | "Tūī" saved as "Tui", day and month swapped | Field-level confidence, side-by-side check, no auto-save |
| Address not placed | Rural delivery address or new street | Flag "cannot place" rather than guess |
| Overexposure | Parenting order visible to all office staff | Restricted class, access log |
| Destructive automation | Clean-up job deletes "duplicate" scans | Write-only storage for source documents |
| Vendor reuse | Forms used to train a model | No-training terms; the Privacy Commissioner [warns](https://www.privacy.org.nz/resources-and-learning/a-z-topics/protecting-children-and-young-peoples-privacy/childrens-privacy-guidance-for-the-education-sector/chapter-16-digital-technologies/) against relying on vendor assurances alone |

## What it costs and where to begin

Most of the cost is connecting to the student management system, mapping documents to records classes, and testing on a term's worth of real packs. List every document type, its owner and its records class with our [AI data inventory template](/resources/ai-data-inventory-template/), then compare extraction approaches using our note on [OCR versus vision models](/engineering/ocr-versus-vision-models/). Our [AI document processing](/services/ai-document-processing/) service builds this end to end. Related workflows: the [staff policy assistant](/industries/education/staff-policy-assistant/) and [parent enquiry triage](/industries/education/parent-enquiry-triage/).

## Questions people ask

### Which school documents should we automate first?

Pick documents with fixed fields, real volume and a clear owner. Enrolment packs are usually first, then trip and activity consent forms. Leave learning support reports, health plans and anything about care or discipline until the basic pipeline has proved itself, because errors in those records carry far higher stakes and they are often handwritten or unstructured.

### Can AI decide whether an address is in our home zone?

It can compare the address with your published zone description and flag likely matches, mismatches and addresses it cannot place, such as rural delivery addresses or new subdivisions. The decision belongs to staff, and out-of-zone places follow the order and ballot process set by the Education and Training Act 2020. Software should never rank or select applicants.

### Can we let the system tidy up duplicate scans and old copies?

Not at the moment. The Ministry of Education says a temporary protection order from Archives New Zealand requires care records, which include all records about students and staff, to be retained and not disposed of, altered or destroyed. Build the pipeline to add new derived records only, and send any disposal question to the Ministry's records team.

### Do we still need to collect immunisation certificates at enrolment?

Primary schools stopped being required to keep immunisation registers on 8 January 2026, because the Aotearoa Immunisation Register now holds that information. Schools can still choose to keep one. Existing records must be retained for 7 years under the school records retention and disposal schedule, so remove the field from your checklist without deleting what you hold.

### How accurate does extraction need to be?

Accurate enough that staff only confirm, not retype. Measure field accuracy on a sample of real packs before go-live, with names, dates of birth and addresses checked character by character, macrons included. If staff correct more than one field in five packs, fix the extraction or narrow the document types before scaling.


## Sources

- [Managing school records](https://www.education.govt.nz/education-professionals/schools-year-0-13/administration-and-management/managing-school-records)
- [Immunisation register requirements at school and kura](https://www.education.govt.nz/education-professionals/schools-year-0-13/administration-and-management/immunisation-register-requirements-school-and-kura)
- [Eligibility to enrol in New Zealand schools](https://www.education.govt.nz/education-professionals/schools-year-0-13/enrolment/eligibility-enrol-new-zealand-schools)
- [Enrolment schemes](https://www.education.govt.nz/education-professionals/schools-year-0-13/enrolment/enrolment-schemes)
- [Public Records Act 2005](https://www.legislation.govt.nz/act/public/2005/40/en/latest/)
- [Children's Privacy Guidance for the Education Sector, Chapter 16: Digital technologies](https://www.privacy.org.nz/resources-and-learning/a-z-topics/protecting-children-and-young-peoples-privacy/childrens-privacy-guidance-for-the-education-sector/chapter-16-digital-technologies/)
- [API pricing](https://developers.openai.com/api/docs/pricing)

Published by Levy Street (https://levystreet.com), Wellington, New Zealand.