Create capacity for client advice: reduce the chase before review

A practical AI workflow paper for Melbourne accounting practices.

Executive takeaway

Your practice may already be trying AI and still be unsure what good use looks like across the team. We help you assess those experiments, decide what matters and choose a useful next move. You do not need to arrive with a process already picked out.

An accounting team may have capable practitioners and good software yet lose time collecting the information needed for a proper review. Missing documents, inconsistent descriptions and repeated follow-ups can leave experienced people rebuilding the same picture before they can exercise judgement.

AI may help prepare that picture: organise permitted information, link it to sources and draft questions for review. The intended outcome is more capacity for client service and a shorter path to a complete review pack. The practitioner still owns the analysis, advice and professional acceptance.

Start with information preparation, measure the entire job and keep tax positions, professional judgements and lodgements under authorised human control. Our team combines advice and implementation to make one bounded change reviewable. We do not promise a multiple of productivity, guaranteed turnaround or a reduction in headcount.

Look at work waiting to become reviewable

Useful discovery begins with a recent job that waited. Which information was missing? Was the client asked twice because the original request was unclear? Did a reviewer discover that two documents covered different periods? Did a junior colleague spend time interpreting a note that should have been clarified with the client?

These are investigation questions, not claims about every practice. Follow the job from engagement and information collection to review. Separate client waiting time from staff effort. Examine who prepares, who checks and who resolves uncertainty. Standard document requests, clearer upload instructions or structured checklists may help before AI is involved.

Choose a process with repeatable inputs and an identifiable reviewer. A trial should help that person see the evidence more clearly. It should not produce confident conclusions from incomplete records or move difficult judgement into a tool that cannot own it.

A fictional running example

Fictional example, not an Alongside People client: A Melbourne accounting practice prepares a monthly review pack for a small business. Staff receive an agreed list of documents and explanations, mark what has arrived and assemble questions for the accountant. The accountant verifies the records and decides what further work or advice is needed.

The proposed tool prepares a source-linked document index and a draft list of unresolved information requests. It distinguishes “not supplied”, “unclear” and “requires practitioner judgement”. It cannot declare a transaction deductible, determine a tax position, post entries, lodge a return or send a client request without review.

Initial examples are synthetic. Any move to real records requires the firm's approval and appropriate client-information permissions. The fictional practice has no reported savings or observed results.

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 practitioner-accepted document index and list of unresolved questions.
• Who judges it: The reviewing accountant, with feedback from the preparer and client-request owner.
• Hold and return to the manual route when: A source is invented, a period mismatch is hidden or an unresolved matter becomes a tax conclusion.
• Evidence for another step: Compare total preparation, checking and follow-up effort for packs of similar scope; record omissions and repeat requests.

A useful fictional flag reads: “The supplied statement covers a different period; the practitioner needs to confirm which record is required.” “This expense is deductible” is an unsupported conclusion outside the proposed tool’s job.

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

Five stages from preparation to dependable use

Understand. Enter with the review-pack job, its practitioner owner and the people gathering documents. Map the request, collection, indexing, queries and review. Agree what “ready for review” means. The benefit hypothesis is fewer avoidable follow-ups and less reconstruction by the reviewer. Exit with a process map, document requirements, baseline, permission boundaries and one question to test. If readiness depends on unwritten knowledge, make that knowledge explicit first.

Try. Enter after agreeing permitted data, tool, reviewer, acceptance criteria and stop conditions. Use synthetic packs containing duplicate documents, mismatched periods and incomplete explanations. The hypothesis is reduced preparation effort while preserving traceability. Exit with observed defects, checking time and the practitioner's proceed/revise/stop decision. Stop if the tool invents a document, treats an unresolved matter as settled or suggests an unsupported tax conclusion. Return to the existing checklist.

Repeat. Enter when a useful trial merits further work. Establish consistent folder structure, period labels, source references and request wording. Have another preparer follow the instructions. The hypothesis is more complete packs with fewer repeated requests. Exit with reproducible instructions, documented exceptions, a manual route and a baseline comparison. The practitioner verifies that the change helps the review rather than simply moving work from preparation into correction.

Operate. Enter only after routine use in the bounded scope is approved. Identify the practitioner owner, operational backup and people allowed to change templates or access records. Retain evidence of what the tool prepared and what people corrected. The hypothesis is steadier turnaround and more usable staff capacity. Exit with operating guidance, visible failures, authorised access, change control and exercised recovery. Pause if permissions lapse or staff cannot recover a pack manually.

Improve. Enter with enough observations for a meaningful review. Compare accepted packs, total staff effort, repeat questions and reviewer experience. The hypothesis is learning which preparation changes genuinely help clients and staff. Exit each review with a retain, revise, expand or retire decision and a next trigger. A structured form may prove better than generated requests. Broader actions need separate authority and evidence; dependable human review can remain the endpoint.

Understand → Try → Repeat → Operate → Improve is our original per-workflow discussion framework. It is not a professional standard, validated readiness assessment or certification. A practice can operate one workflow dependably and still be exploring another.

Professional responsibility and client information

The Tax Practitioners Board's final AI guidance statement, TPB(GS) 55/2026, was issued on 22 July 2026. It concerns registered tax and BAS agents providing tax agent services. It retains practitioner responsibility and calls for professional judgement and output review. Its confidentiality discussion addresses permission before client information is disclosed to a third party, including AI configurations that involve such disclosure.

APESB's technical alert on ethical AI use reminds members of the Australian Accounting Bodies that their ethical obligations under APES 110 continue when using AI. Membership, service type and other applicable standards need to be identified; do not treat tax-agent guidance as the complete rulebook for all accounting work.

OAIC's commercial AI privacy guidance supports product due diligence and privacy assessment where personal information is involved. Determine the firm's actual obligations with appropriate advisers. This paper does not provide tax advice or legal assurance.

A proposed 30-day pilot

This is a planning outline, not a delivery commitment. Start only once the firm has approved the scope, environment and information. Client permissions, professional review and access decisions may take longer.

Days 1–7: Follow recent review-pack preparation, establish baseline effort and agree the definition of readiness. Create a synthetic example set, including missing and contradictory material. Document the restricted actions and approval route.

Days 8–14: Build the index-and-question draft. Practitioners check it against the source records, record defects and inspect whether uncertainty remains visible. Improve templates before adding more automation.

Days 15–23: Repeat in shadow mode using authorised examples. The existing process remains authoritative. Test whether another staff member can prepare and recover a pack using the instructions. Review every outgoing request; no unattended client messaging.

Days 24–30: Compare accepted packs, review effort and remaining problems. Produce a decision record, operating proposal and handover material. Choose continued testing, an authorised bounded implementation, revision or stopping. The end of the month does not automatically establish readiness.

Baseline and guardrails

The primary outcome could be total staff effort to produce a practitioner-accepted review pack. Record preparation, checking, correction and client-request effort. Track elapsed turnaround separately, including time awaiting client responses. Compare similar periods and pack complexity rather than attributing every change to AI.

Guardrails include source traceability, incorrect missing-document flags, repeated requests, period mismatches and unauthorised disclosure. Include practitioner confidence and the client's experience of the questions. Preserve agreed criteria; do not count a pack as complete by silently excluding difficult matters.

Keep estimates separate from observations and record the limits of a small sample. Time released is capacity to allocate, not automatically profit or cash saved. Financial amounts should remain tied to verified records and the firm's established checking processes; fluent wording cannot validate a balance.

Record vendor access, retention, storage, subprocessors, permitted uses and change notifications. Masking a name does not by itself establish that records are anonymous or authorised for processing. Keep the person approving the use accountable and the manual process available.

How our team helps

Alongside People provides advice and hands-on implementation. Paul Volpato, our founder and CEO, connects operational purpose with practical systems work. Our team can map the preparation bottleneck, design the evidence plan, implement an agreed workflow and prepare instructions, checks and handover.

The firm's professionals retain judgement and acceptance. We agree scope and ongoing responsibilities rather than imply a tax-agent service, professional qualification or open-ended support. Bring one recurring job and a representative example. Together we can decide what is worth trying and what should stay with a person.

Official references

Accessed 3 October 2026; recheck before consequential use.

• TPB(GS) 55/2026, issued 22 July 2026: final guidance for registered tax/BAS agents, not the earlier exposure draft.
• APESB AI technical alert, 31 October 2025: professional-body guidance concerning members' continuing ethical obligations.
• OAIC commercial AI guidance: regulator guidance, not a comprehensive statement of all applicable obligations.

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.