How a suburban Melbourne accounting firm won back ~$300K a year of staff time it can bill to clients — without changing a single core system.
Harwood Lane Partners is an accounting and advisory practice in Melbourne's south-east: $8.5 million in annual fees from roughly 1,900 client groups — small businesses, family groups and self-managed super funds. Of its 38 people, 28 are fee earners: the accountants whose hours are billed to clients. In a firm like this, time quite literally is money — every hour freed up is an hour that can be sold.
That last number is what brought the managing partner to the table. The firm had no idea what was flowing into free chatbots from nine different desks — client work, possibly client data.
The Tax Practitioners Board's draft guidance on AI under the Code of Professional Conduct makes two things plain: practitioners remain fully accountable for AI-assisted work, and Code item 6 confidentiality obligations apply to anything pasted into a third-party tool. For a partnership, "should we look at AI?" became "we need a defensible policy this quarter."
The firm also sat with the majority described in Thomson Reuters' Future of Professionals research: among the 78% of firms without a defined AI strategy — while firms with one are twice as likely to see AI-driven revenue growth.
Crucially, nothing here required technical work. Everything ran on ready-made AI features inside software the firm already owned, plus approved AI assistants for drafting and research — no custom software, no system rebuilds, no IT project. That's what makes the Sprint possible: 40 days, fixed fee, nothing that can break.
A half-day session with the partners, plus an anonymous staff survey — which surfaced the unapproved AI use without pointing fingers. Twenty-one possible AI applications were scored on value, ease and risk using the firm's own billing rates. Six made the cut.
A clear, written policy for AI use, built on the firm's real obligations — the tax regulator's draft AI guidance, client confidentiality rules, and privacy law. The hard lines: no client tax file numbers in free public tools, no AI-generated advice without partner review, nothing lodged on an AI's say-so. Then, and only then, the software: Microsoft Copilot for 30 staff (~$16K/yr) and a business-grade AI assistant for 10 heavy users (~$4.5K/yr).
Each team practised on its own genuine tasks (with client data kept inside the approved tools): the accounts team used AI to summarise working files, the tax team drilled the discipline of "the AI drafts, you verify and cite", and the advisory team rebuilt its client-report process around a draft–review–polish loop. Every team left owning its own library of proven instructions, with a named senior responsible for it.
Usage data from the tools, a re-run of the staff confidence survey, and a comparison of timesheets against the same fortnight last year. The numbers went to the partners with a single decision attached.
| Use case | Function | Level |
|---|---|---|
| Client correspondence, engagement letters, fee proposals | All | I / II |
| Workpaper summaries and year-end file notes | Business services | II |
| Management letters and advisory report first drafts | Advisory | II |
| Tax research first-pass, mandatory source verification | Tax | II · Amber |
| Meeting transcription → file notes | All | I |
| Client newsletter and seasonal tax alerts | Practice | I |
The maths, in plain terms: 28 fee earners, each saving 2.5 hours a week, over 46 working weeks ≈ 3,220 hours a year handed back to the firm. Not every freed hour becomes a billed hour — people catch up, go home on time, turn work around faster — so the model assumes only 35% of those hours get re-sold to clients, at the firm's average $265/hr. That alone is worth ~$299K a year, against ~$39K of total first-year cost. And beyond the arithmetic: a written AI policy aligned to the regulator's draft guidance, a documented answer for the firm's insurer, and a routine that doesn't depend on the consultant coming back.
Drag the sliders to match your firm and watch the ledger update. It uses the same assumptions as the case study — and you can set the most important one, how many freed hours actually get billed, as low as you like to stress-test the result.
Everything runs inside software the firm already owns, plus an approved AI assistant. No system rebuilds, no IT project — nothing that can break.
A managing partner with timesheet data in hand doesn't need a board cycle to approve a fixed fee with a measurable day-40 checkpoint.
The tax regulator's draft AI guidance turned "should we look at AI?" into "we need a defensible policy this quarter." The Sprint delivers it in week two.
In a firm that sells time, "2.5 hours per person per week" needs no translation to become money. The value is visible on the very timesheets the firm already keeps.
The Sprint is the entry point, not the destination. Firms that complete it have proof in their own timesheets — exactly the position from which to plan the next stage: AI in client-facing work, new advisory services, and the deeper workflow automation that genuinely does need a bigger program.
And one thing worth saying plainly: a wave of "autonomous agent" products is coming for professional-services back offices — software that promises to do the admin by itself. Some of it will be excellent. The firms that benefit won't be the ones that bought first; they'll be the ones with the policy, the data discipline and the measurement habit to choose well and adopt safely. That's what the Sprint builds. We maintain a continuously updated radar of AI tools and agents available to Australian practices — so when the right one arrives for your firm, you'll hear it from us before you hear it from a salesperson.