Industry guide - accounting practices
AI for accountants: the practice build sheet
The professional bodies cover what AI means for the profession. This page covers what to actually build in one practice: the six workflows worth automating first, what stays behind the accountant, and what the hours are worth.
By Tylar Edwards - the engineer who scopes and builds these workflows for professional practices. Updated August 2026.
Why practices feel the drag more than most
An accounting practice is a document-chasing machine wrapped around a judgment business. Every job waits on client paperwork; every inbox mixes advice questions with routine confirmations; every workpaper starts with re-keying figures from source documents. The drag is not the accounting - it is the process around the accounting, and that process is exactly what AI automates well. And because practice hours are chargeable, recovered admin time converts to revenue at a rate most industries cannot match.
Six practice workflows AI automates today
| Workflow | What AI does | What stays with the accountant |
|---|---|---|
| Client document chasing | Runs the missing-information ladder per job checklist, reads what arrives, files it against the job and updates the status | Almost nothing - this is the purest recovered time in a practice |
| Inbox and ATO correspondence triage | Sorts mail by client, job and urgency; drafts routine acknowledgements and lodgment confirmations for review | Anything that constitutes advice, and every ATO position taken |
| Workpaper preparation | Extracts figures from bank statements and source documents into your workpaper templates and flags anomalies for attention | Review, judgment and sign-off - AI prepares, the accountant concludes |
| Client onboarding | Drafts engagement letters from your templates, assembles entity checklists, sends portal invitations and chases signatures | Engagement scope, pricing and acceptance |
| Lodgment reminder cycles | Runs BAS and return reminder sequences from lodgment dates, escalating tone on your schedule | The escalation call when the ladder runs out |
| Job status upkeep | Keeps practice-management status current from email and document activity, and nudges stalled jobs | Reprioritising the queue |
Around the practice stack you already run
Microsoft 365 or Google Workspace
Email and documents
Client context is commonly distributed across inboxes and files.
HubSpot or a practice CRM
Pipeline and client records
Lead and engagement status need a shared system of record.
Xero, MYOB or QuickBooks
Billing and accounting
Approved scope and time need a reliable billing handoff.
What it is worth
Using the capacity model behind our free assessment at professional services rates: eight hours a week of recovered admin is an indicative $13,140-$20,610 a year at $90/hr. The boundary we never move: require human review for advice, proposals and external commitments.
Common questions
- What should an accounting practice automate first?
- Client document chasing, in almost every practice we assess. It consumes admin hours year-round, it is pure process rather than judgment, and the AI failure mode is harmless - a reminder not sent - rather than an advice risk.
- Does AI replace accountants or bookkeepers?
- Not in a practice context. AI prepares: it chases, extracts, drafts and files. The accountant concludes: judgment, advice, positions and signatures stay human, and every workflow we build enforces that boundary with explicit approval points.
- Does this work with Xero, MYOB and our practice software?
- Yes - that is the design constraint, not an afterthought. Workflows are built around the tools a practice already runs: Microsoft 365 or Google Workspace, the practice CRM, and Xero, MYOB or QuickBooks for the billing handoff.
- What is it worth to a practice?
- At professional services capacity rates, a practice losing eight hours a week to admin recovers an indicative $13,000-21,000 a year of chargeable capacity - and unlike most industries, recovered hours in a practice convert directly to billable time.
- What about client confidentiality?
- Confidential client material stays permission-aware: workflows run inside your existing accounts and access model, and nothing client-identifiable trains anyone's model. The data-handling design is agreed in scoping, before the build.