Give education staff more time for the work that needs them

A practical AI paper for Melbourne education organisations.

You may already have staff exploring AI while still working out what good use should look like across your institution. We help you turn those experiences into a clearer direction and choose a worthwhile next step. There is no need to have an improvement project defined first.

The useful starting point. Improve recurring staff administration. An accurate internal operations pack from approved material is a useful first experiment. Test whether it reduces searching, rewriting and correction, leaving capacity for work requiring staff attention and judgement.

Start with the staffroom's recurring work

A useful question for an education leader is what staff repeatedly have to reconstruct. The answer might be room-use instructions, event preparation steps, a meeting pack or an induction guide. Instructions may be spread between a policy library, an email thread and a document someone keeps updating privately.

An administrator may not know which checklist is current or who must approve a change. Asking AI to rewrite everything risks making those gaps less visible.

Follow one task with its producers and users. Establish an acceptable result, authoritative information and human-owned decisions before choosing AI or ordinary document management.

This paper focuses on staff administration. Student ranking, grading, admission, disciplinary decisions, welfare assessments and individualised learning judgements are outside the proposed pilot. Student reports and direct communication with families are also excluded. The people and relationships at the centre of education remain a reason to design the process carefully.

Know which guidance applies

The Victorian Department of Education's generative AI policy sets requirements for Victorian government schools, including restrictions on uploading personal and sensitive information. It directs staff away from AI use that replaces teachers' and leaders' professional judgement or authentic communication. Its requirements must be read with other departmental technology and privacy policies. These school rules should not be described as the policy of every university, TAFE or independent provider. Victorian school AI policy.

The accompanying appropriate-use guidance recommends disclosure, records and monitoring, and asks schools to consider administrative and financial costs. That supports checking the whole workflow rather than celebrating faster text generation. Victorian appropriate-use guidance.

The Australian Framework for Generative AI in Schools guides responsible and ethical use for people connected with school education. Education Ministers endorsed its 2024 review in June 2025. It is a national guidance framework, not approval for a particular product or a replacement for institutional requirements. Australian schools framework.

Before implementing a generative AI tool, the Victorian policy also requires schools to ensure accessible and inclusive use. Include the relevant reasonable adjustments in the proposed workflow and trial, rather than leaving accessibility to the final export.

A tertiary provider should identify its own approved-tool, information, accessibility and records requirements before adopting this example. We offer a practical planning method, not a determination of compliance.

A fictional workflow: the staff induction pack

A fictional Melbourne secondary school prepares an induction pack for new staff each term. The administrator gathers public or internally approved non-personal instructions about facilities, document locations and routine processes. A leader checks the pack. Old copies contain inconsistent room names and links to superseded procedures.

The proposed assistant drafts sections from a deliberately small source collection and attaches a link and version to every instruction. It produces a separate list of missing or conflicting information. It cannot read student records, analyse staff performance, answer welfare questions or distribute the pack.

The administrator checks the links and facts; process owners confirm the instructions; a leader approves the result. A new staff member then uses the draft to find an ordinary procedure and reports what remained unclear. This tests usefulness as well as correctness.

Use role names rather than a personal staff directory. This fictional example does not imply previous school engagements.

What would make this worth continuing?

The following is a decision aid for the fictional trial, not an observed result. Agree the limits with the people responsible before testing.

• Accepted result: An approved induction pack that a new staff member can use to find the intended procedure.
• Who judges it: The content owner and approving leader, with a new staff member testing usefulness.
• Hold and return to the manual route when: Personal information enters the trial, an instruction lacks a source or an omitted step could misdirect a consequential action.
• Evidence for another step: Compare total preparation and clarification effort for similar packs; test source changes and accessible use.

In a fictional check, ask a new colleague to find the current room-booking procedure using the pack. Record whether the link works, the instruction is clear and the format is usable with their access needs. The trial has more to learn if the colleague still needs to find the author.

Use the one-workflow worksheet to record the agreed question, evidence and next decision.

Five states that keep the decision practical

Understand. Follow the pack's preparation and a new staff member's attempt to use it. Identify duplicated instructions, authoritative owners and approval steps. The benefit hypothesis is less reconstruction and fewer follow-up questions. The evidence gate is an agreed scope, approved sources, named owner and baseline. If the real issue is a missing policy decision, resolve that first.

Try. Prepare an offline draft from approved non-personal or synthetic material. Include a broken link, contradictory room instructions and a missing approval step. The benefit hypothesis is less assembly time. The gate is a draft with traceable instructions, visible gaps and manageable review effort. Student or staff personal information entering the trial triggers a pause.

Repeat. Have another administrator prepare a new version using the same documented process. Include a changed source. The benefit hypothesis is consistency between terms and colleagues. The gate is reproducible quality, correct handling of changes and a usable manual route. A checklist that works only for its author needs further refinement.

Operate. Introduce the approved process within a defined scope, with a content owner, backup and review calendar. Keep dissemination under human control. The benefit hypothesis is less chasing for a dependable pack. The gate is an exercised fallback, appropriate access and an accepted maintenance workload. An extra feature in the software does not expand permission.

Improve. Review actual preparation effort and the recipient's ability to find and act on instructions. Remove redundant content and correct the source library before adding more generation. The benefit hypothesis is continuing usefulness rather than increasing volume. The gate is a documented decision to retain, revise or stop, with evidence and a next review trigger.

Understand → Try → Repeat → Operate → Improve is our original discussion framework. It is not a validated assessment, school quality rating or official standard. Different workflows may be in different states.

A proposed 30-day pilot

In days 1–7, select one pack, observe the manual workflow and confirm source and tool approval. In days 8–14, compare manual and assisted drafts using the same material. In days 15–23, repeat with another preparer, check an updated procedure and run a short usability exercise. In days 24–30, review the results and decide whether routine use is justified.

Schedule the work around staff capacity and the institution's approval process. A month without an appropriate event can still support a simulation, but it cannot establish performance during a busy term. These are proposed steps, not a commercial timetable, promised result or live-use authorisation.

Count work returned to staff

The primary measure is total staff effort to produce an accepted pack. Include preparation, review, fixing links and answering clarification questions. Compare documents of similar scope and record whether policy changes altered the workload.

Quality guardrails include incorrect instructions, omitted approval steps, outdated sources and whether a recipient can complete the intended task. Check accessible formatting and clear language with actual users. Record review time separately: faster drafting that requires longer checking may not help.

Cost includes licences, initial setup, training and maintaining the collection. Agree acceptance thresholds before the test. Stop if personal information appears, if an instruction could misdirect a consequential action or if the process cannot identify its source. Preserve unsuccessful drafts so errors inform the decision.

Additional capacity is a benefit hypothesis. Ask staff what they could spend time on instead; do not infer improved student outcomes or financial savings from a quicker induction document.

A process people can maintain

Limit permissions to the approved source collection and a draft workspace. Keep versions, responsible owners and approval status visible. Documents retrieved by the assistant must not instruct it to send messages or broaden access. Distribution and source amendments remain human actions in the pilot.

If drafting fails, return to the existing template and document list. Hold unverified sections rather than silently substituting a plausible answer. If an approved pack contains an error, use the school's correction route and identify which copies need replacing. Test that a backup administrator can prepare the pack manually.

Our team can map the work, design a bounded trial and implement the agreed process with staff. We can help organise approved information, connect suitable tools and produce instructions, checks and a handover. Institutional leaders retain policy, educational and publication decisions. Wider rollout and continuing support require separate agreement.

Sources inspected 3 October 2026. The linked policy and guidance support the specific statements above; refresh applicable institutional requirements before implementation.

If you are already trying AI and are unsure what deserves further investment, tell us what you have tried and what you want to improve. We can help find the starting point together.