A services firm's AI budget looks different from a product business's because most of the cost either supports billable work directly or sits as pure overhead -- there's no manufacturing line or inventory cost to blend it into. That distinction should drive how you build the budget line by line, and most Australian services firms we've reviewed haven't made it explicitly.
The categories that actually matter for a services firm
Per-consultant billable-support tools -- anything that speeds up chargeable client work directly (drafting, research, document review).
Practice-wide overhead tools -- internal reporting, admin automation, things that support the whole firm rather than one engagement.
Client-facing tools -- anything a client sees or interacts with directly, which usually carries a higher bar for quality control and data handling.
One-off build costs -- setup, configuration and integration work, budgeted separately from the recurring run-rate.
Why the split matters for the budget conversation
A partner or practice lead reviewing an undifferentiated AI line item -- 'software: $3,400/month' -- has no way to judge whether that's reasonable. The same $3,400 split into '$1,800 in per-consultant tools supporting $180,000 of monthly billings' and '$1,600 in practice overhead' is a number the partnership can actually evaluate, because it connects to something they already understand: utilisation and overhead ratios.
This split also surfaces waste faster. A tool sitting in the 'per-consultant' bucket that consultants aren't actually using to support billable work is an easy cut once it's visible in its own line, in a way it never was buried inside a single combined software total.
A worked budget for a mid-sized firm
A 25-person Adelaide engineering consultancy building their first properly split AI budget landed on: $1,100/month in per-consultant drafting and research tools (roughly $44 per consultant, easily justified against time saved on chargeable work), $650/month in practice-wide reporting and admin automation, and $300/month in a client-facing document portal assistant, for a total of $2,050/month against roughly $310,000 in monthly billings -- well under 1% of revenue, but now defensible line by line rather than a single number partners had to take on faith.
Setting the review cadence
Review this quarterly, not annually, and revisit the per-consultant category specifically whenever headcount changes -- a tool priced per seat needs active management as the firm grows, or you end up paying for licences attached to people who've left. A firm that reviews this line only at annual budget time typically discovers six to twelve months of quietly wasted per-seat spend in one uncomfortable finding, rather than catching it within a quarter.
What to do when the numbers don't line up
Sometimes splitting the budget this way surfaces an uncomfortable finding: a tool sitting in the per-consultant category that, on review, almost nobody's actually using to support billable work. That's not a failure of the budgeting exercise, it's the exercise working as intended -- the whole point of splitting an undifferentiated software line into categories tied to actual firm activity is to make exactly this kind of gap visible instead of letting it hide inside a total nobody questions.
Bring this to the partnership as a specific finding, not a vague 'we should probably review our software spend' comment that tends to go nowhere. 'Tool X costs $340 a month and three consultants have used it twice this quarter' is a concrete decision point a partnership can act on in five minutes, in a way a general sense that spend feels a bit high never quite prompts action on its own.
None of this requires new software to track. A shared spreadsheet with the four categories, reviewed each quarter alongside the normal partnership financials, is enough structure for a firm under fifty people. The value is in the discipline of the split, not the sophistication of the tool used to maintain it.
A note on client-chargeable AI cost
A small but growing number of Australian services firms are starting to itemise AI-assisted work on client invoices the same way they would a disbursement, particularly for document-heavy engagements. Whether that fits your firm is a separate conversation, but it only becomes a real option once you can point to an actual per-engagement AI cost figure -- which is exactly what the category split above gives you the ability to do, and an undifferentiated software line never could.
Start with whatever spend you can already see on this month's statements rather than waiting to build a perfect system first. A rough first-pass split done today beats a comprehensive framework that never gets started.
If you're building or rebuilding a services-firm AI budget and want a second set of eyes on the split before it goes to the partnership, get in touch through /contact.
