More capacity for public service: less time rebuilding information

A practical AI workflow paper for Melbourne government teams.

Executive takeaway

Your organisation may already have AI experiments underway without a shared view of what good looks like. We help connect those efforts to public-service priorities and choose where to go next. The example below makes that conversation concrete; you do not need a project selected before speaking with us.

A service team can spend valuable time rebuilding information that already exists: checking which notice changed, finding the authoritative date and turning several documents into a usable staff update. AI may help with that preparation. The useful outcome is more capacity for service work, with reliable information and clear responsibility.

Begin with one factual workflow, authorised sources and a person who owns the result. Compare total preparation and checking effort with the existing approach. Keep administrative decisions, eligibility assessments and consequential advice outside the pilot described here. A faster draft is useful only if the recipient can rely on it after the agreed checks.

Our proposed approach combines advice and implementation. We establish where the work is getting stuck, build a bounded change and help the organisation decide whether it deserves routine use. This paper describes a practical experiment, not a claim of government approval or a promise of savings.

Find the service bottleneck

Useful discovery questions concern the work itself. Does a coordinator repeatedly compare published updates? Do frontline staff search several places for the current instruction? Does a reviewer rewrite a summary because conditions or dates were omitted? Does every handover depend on one colleague remembering where the authoritative material lives?

These are possible bottlenecks to investigate, not findings about Melbourne agencies. Follow a recent example with the people who prepared and received it. Separate waiting for approval from active drafting. Ask what would make the recipient's next action easier. A shared source list or better update template might solve much of the problem without AI.

Choose a workflow that can be observed within a bounded trial. Avoid starting with decisions about an individual's rights, benefits or access to services. Even preparatory work needs scrutiny when it could influence a consequential decision.

A fictional running example

Fictional example, not a client or government case study: A Melbourne-based government service team prepares a weekly staff reference note from published service notices. A coordinator finds changed pages, collects dates and drafts an update. A communications owner checks it before distribution. Staff use the note to locate the authoritative notice; it does not replace that source.

The proposed tool produces a draft with links, source dates and unresolved questions. It cannot publish, contact the public, decide eligibility or change service rules. Information absent from a source remains an explicit gap. Internal approval of the tool, workflow and information is required before any trial. The example reports no measured benefit.

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: A checked staff reference note with current source links, dates, conditions and visible unresolved questions.
• Who judges it: The authorised communications owner, with the staff who rely on the note.
• Hold and return to the manual route when: A condition is invented, an authoritative source cannot be verified or a draft crosses into eligibility or service decisions.
• Evidence for another step: Compare total preparation, source checking and recipient clarification for notes of similar scope.

Record the actual organisation, relevant policy and authorised policy owner in the experiment brief. A Victorian agency, Commonwealth body and local council need their own applicability decision; a paper about government does not establish one for them.

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

Build dependability through five stages

Understand. Enter with the recurring reference-note job and its owner. Map source discovery, drafting, review, approval and distribution. Establish which notices are authoritative and what staff need from the note. The benefit hypothesis is less searching and fewer clarification requests. Exit only with a process map, baseline, source/permission boundaries, a named approver and a testable question. If no one can resolve conflicting notices, fix that responsibility before introducing AI.

Try. Enter after the sponsor agrees the permitted public sources, tool, checks and stop conditions. Test synthetic or approved public-information examples, including an expired notice, a changed condition and an unreachable page. The hypothesis is reduced draft-preparation effort without losing dates, conditions or links. Exit with reviewed observations, corrections and checking time, plus an owner decision. Stop if a condition is invented or a source cannot be verified; prepare the note manually.

Repeat. Enter when the trial justifies another test. Use consistent source dates, instructions and review criteria. Ask a second staff member to follow the process. The hypothesis is fewer omissions across repeated updates. Exit with reproducible instructions, exception records, a tested manual route and comparison against the baseline. A change that works only with tidy examples or its original author is still a trial.

Operate. Enter only after routine use in this scope is authorised. Name an owner and backup, control access and make failed runs visible. In the fictional team, a changed source list or new tool feature goes through review before use expands. The hypothesis is more consistent staff updates with less chasing. Exit with operating guidance, approved scope, visible exceptions, exercised recovery and an owner review. Pause if approval, ownership or source integrity changes.

Improve. Enter when routine observations support a review. Compare preparation, checking, correction and staff usefulness with the original purpose. The hypothesis is retaining useful changes while avoiding unhelpful expansion. Exit each review with a documented retain, revise, reduce or retire decision and its next trigger. A clearer template with less AI may be the better outcome. Broader delegation requires a fresh approval and evidence decision.

This is Alongside People's original workflow framework, Understand → Try → Repeat → Operate → Improve. It is not a certification, validated assessment or compulsory ladder. Human approval can remain at every consequential step.

Identify the jurisdiction before applying a rule

The Victorian generative AI administrative guideline applies to its specified public service bodies, public entities and personnel. For organisations within its scope, it requires agency-approved tools to be used ahead of publicly available tools, limits information entered into unapproved tools to public information and retains personnel accountability. Information entered into an approved tool must stay within the protective marking the organisation permits for that tool. Organisation and sector policies may impose higher thresholds.

OVIC's enterprise generative AI guidance addresses privacy and information-security preparation and review for its Victorian public-sector audience. Purchasing an enterprise product does not complete that work.

The Commonwealth responsible AI policy, version 2.0 applies to non-corporate Commonwealth entities with specified exceptions. It has mandatory requirements and is effective from 15 December 2025. It is not automatically a policy for Victorian agencies or local councils. A council must establish its own applicable requirements; this paper makes no blanket claim about council coverage.

Ask the organisation's authorised privacy, security, records, legal and procurement specialists to determine applicability and acceptance. Our framework helps prepare that conversation; it does not provide legal assurance, permission to use information or supplier-panel eligibility.

A proposed 30-day pilot

This is a planning example, not a service delivery promise. Begin the clock after the sponsor has the authority, people and approved environment needed. Approval may take longer than the experiment; no deadline overrides it.

Days 1–7: Map the reference-note workflow, select examples and record baseline observations. Agree source currency, review responsibilities and the exact scope. Produce an experiment brief and a decision on whether to proceed.

Days 8–14: Prepare a draft workflow using synthetic or authorised public sources. Run difficult examples alongside normal ones. Review every output, record defects and adjust the instructions. Hold any result with unresolved source questions.

Days 15–23: Repeat in shadow mode, so existing work remains authoritative. Ask another staff member to use the instructions and exercise the manual fallback. Collect the effort of preparer, reviewer and recipient.

Days 24–30: Review the comparison and remaining limits. Deliver the evidence, proposed operating responsibilities and an owner decision to continue testing, authorise a bounded next step, revise or stop. Routine use is a separate decision.

Measure capacity without hiding work

The primary measure could be total staff effort per accepted reference note. Record preparation, source verification, review, correction and distribution separately. Pair it with end-to-end turnaround, unanswered recipient questions and source accuracy. Track omitted conditions, stale information and failed-source handling as guardrails.

Preserve the original examples and criteria. Explain differences in workload or source complexity. Sparse observations reveal what happened in the sample; they cannot prove organisation-wide productivity. Released time may support other work, but it is not automatically a cash saving.

Record data flows, retention, access, audit needs and supplier involvement before implementation. Use approved storage for evidence, and avoid putting personal or sensitive material into a tool simply to test its limits. Everyone should know who can pause the workflow and how staff receive a corrected note.

How our team helps

Alongside People provides practical AI advice and hands-on implementation. Paul Volpato, our founder and CEO, brings the work's purpose, people and technical choices into one conversation. Our team can help map the process, design the pilot, implement an agreed draft-and-review workflow and prepare checks, training and handover.

Your organisation retains authority over information, approvals and service outcomes. We agree scope and operating responsibilities rather than imply indefinite support. The useful first conversation starts with a recurring job, its owner and what would make the result more dependable.

Official references

Accessed 3 October 2026. Recheck before consequential application.

• Victorian administrative guideline, updated 12 February 2025: minimum requirements within its defined scope.
• OVIC enterprise generative AI guidance, updated 26 June 2026: regulator guidance for the specified Victorian audience.
• Commonwealth responsible AI policy v2.0, effective 15 December 2025: mandatory policy within its defined Commonwealth scope.

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.