Your team is moving faster. The engineers say they are shipping more. The analysts are turning reports around in hours instead of days. Everyone agrees it feels productive.
Your board wants a number.
McKinsey's State of AI, published 25 August 2026 from a survey of 1,719 respondents across 97 countries, puts the gap in one line: 80 per cent of respondents say AI has improved their own productivity, while only 37 per cent can attribute any EBIT impact to it, and just 6 per cent qualify as high performers. Both of those last two figures are flat against 2025. PwC's 29th Global CEO Survey from January 2026, covering 4,454 CEOs, found 56 per cent saw neither increased revenue nor reduced costs from AI over the prior twelve months, with only 12 per cent seeing both.
The problem is not that AI delivers nothing. It is that individual productivity is not an enterprise metric, and boards fund enterprise metrics. This article sets out the framework that gets AI spend approved and kept: headcount avoidance. Before you build the case, make sure you can see the cost accurately, which is what the companion guide on how to budget for AI tools covers.
Published: September 2026
The usual approach is to claim a percentage. Our team is 30 per cent more productive. We save ten hours per person per week.
Three problems.
You almost certainly did not measure a baseline. If you were not tracking stories shipped per sprint, reports produced per week or turnaround time before the rollout, you are comparing a feeling to nothing.
Self-report is unreliable in a specific and well-documented way. Randomised trials of experienced open-source developers in 2025 found participants believed they were around 20 per cent faster with AI assistance while measured task completion took roughly 19 per cent longer. That is one study on one population, and it is not evidence that AI does not work. It is strong evidence that asking people whether they feel faster tells you nothing about output.
And boards do not fund feelings. A six-figure annual line needs to connect to the financial statements.
Instead of proving existing people are more productive, which is hard, you demonstrate that AI tooling has removed or deferred the need for planned hires. This is not about cutting people. It is about redirecting hiring budget.
Most people underestimate what an employee costs. In Australia the loaded cost includes base salary, superannuation, payroll tax, workers compensation, recruitment, equipment and software.
Two things to get right, because both are commonly wrong in AI business cases.
Superannuation guarantee is 12 per cent of qualifying earnings, and has been 12 per cent since 1 July 2025. There is no further legislated increase. What changed on 1 July 2026 was Payday Super, which requires the contribution to reach the employee's fund within seven business days of each payday, calculated on qualifying earnings rather than the old ordinary time earnings base. If your model still says 11.5 per cent rising to 12 per cent, every number downstream is understated. Our guide to what employers actually pay under the super guarantee has the detail.
Payroll tax is state-specific and NSW is 5.45 per cent, not the 4.85 per cent figure that circulated during the pandemic reduction. Revenue NSW has published $1,200,000 and 5.45 per cent every year since 1 July 2022, unchanged for 2026 to 2027. The threshold is tested on total Australian wages, not just NSW wages, and it is apportioned if you employ interstate. If your grouped wages sit below the threshold, load nothing. Check your own state with the payroll tax threshold calculator and the state by state payroll tax thresholds and rates guide.
Illustrative loaded cost for a senior software engineer in Sydney, assuming grouped wages above the NSW threshold. Base $160,000, super at 12 per cent adds $19,200, payroll tax at 5.45 per cent adds $8,720, workers compensation around $1,600 to $3,200 depending on industry rating, equipment and licences $8,000 to $12,000 in year one, and recruitment at 15 to 20 per cent of base adds $24,000 to $32,000 in the first year only.
That gives roughly $195,000 to $210,000 ongoing and $220,000 to $240,000 in year one. A financial analyst on $90,000 to $110,000 base lands around $115,000 to $145,000 ongoing. A customer support lead on $75,000 to $95,000 base lands around $95,000 to $125,000.
Run your own roles through the employee cost calculator rather than using these bands, and cross-check the method against our guide to the true cost of hiring an employee in Australia.
Go back to the hiring plan in your last board-approved budget. Which roles were funded? Which of those have become unnecessary or deferrable because AI tooling absorbed the capacity gap they were meant to fill?
Common cases: a second analyst no longer needed because the first can process data faster, a junior developer hire deferred because senior developers with coding agents are covering the workload, an extra support hire avoided because per-person ticket throughput rose.
When this framework does not apply. If the role was never budgeted, you cannot count the avoidance. No budgeted role, no saving. A CFO will find that hole in ninety seconds, so name it before they do.
The old version of this slide showed one net number. That version gets approved and then blows up, because it assumes the AI line is fixed. It is not.
Show both:
The second net number is the point. Gartner reported in June 2026 that AI coding costs above $2,000 per developer per month are already appearing among its clients, with outliers at $20,000 and $32,000 in a single month, and projects that coding token costs will match global-average developer monthly pay by 2028. An uncapped agent rollout can consume the hire you avoided. Showing that yourself is what gets the governance framework approved alongside the spend.
The framework only holds if structural changes stop the saved budget leaking elsewhere.
Justify to replace. When someone leaves, require the hiring manager to make the case for backfilling rather than doing it automatically. The question becomes whether AI tooling can absorb some of the capacity or whether the role needs judgement and presence that cannot be replicated. This is not a hiring freeze. It makes every hire a decision.
Formal budget reallocation. If you budgeted for three hires and are making one, the remaining amount should explicitly offset the AI line in the financial model rather than disappearing into the general pot. Without that link, there is nothing to point at next year.
Quality guardrails. Track reverts, error rates and customer satisfaction alongside output. Headcount avoidance that ships worse product is not a saving, it is a deferred cost.
The measurement plan is what separates a business case from a hope. Make it hard to fake.
Baseline in the two weeks before rollout, using metrics that already exist in your tools rather than survey questions. Engineering: pull requests merged, cycle time, revert rate. Finance: days to close, time to report, journals per close, error rate. Support: tickets per person per day, first response time, customer satisfaction. Owners and managers: hours on email and proposals from the calendar, or a written diary started before the tool is switched on.
Measure again at 30, 60 and 90 days. Then apply a decision rule you have written down in advance.
If output per full-time equivalent is up by 15 per cent or more and quality has not fallen, convert the pilot into a hiring-plan adjustment.
If people feel faster but output is flat, you have a novelty effect. Cut seats back to the power users.
If tokens per unit of output are rising, you have an agent loop or a model-default problem, not an adoption success.
That third rule is the one most companies do not have, and it is the one that catches the expensive failure mode early. If you do not have a reporting rhythm capable of producing these numbers monthly, our guide to setting up a weekly KPI dashboard is the place to start.
There is a case for spending more. In March 2026, Nvidia chief executive Jensen Huang proposed giving engineers token budgets worth roughly half their base salary, describing tokens as a recruiting tool. Worth context: Nvidia supplies the infrastructure AI runs on, its engineers work on some of the strongest AI use cases in the world, and for an Australian company of 80 people the proposal is neither realistic nor necessary.
There is also a case for governing hard. Datadog published first-party figures in August 2026 showing it saved more than $1 million per month by changing two defaults, moving its default model down a tier and dropping coding-agent effort from high to medium, without banning anything.
Huang is the spend-more pole. Datadog is the govern-it pole. The useful question is not how do we spend less, and it is not how do we spend more. It is how do we spend the right amount in the right places, with a cap.
The data now separates the 5 to 8 per cent capturing real value from everyone else, and the pattern is consistent.
PwC found high performers were far more likely to have applied AI to products, services and customer experience: 44 per cent against 17 per cent of the rest. Companies with strong foundations, meaning responsible AI frameworks, defined roadmaps and integrated technology environments, were around three times more likely to report meaningful returns.
McKinsey adds the budget dimension: 28 per cent of organisations already spend more than 10 per cent of their enterprise ICT budget on AI, 60 per cent expect to spend more next year, and high performers are more than twice as likely to be spending above 15 per cent. Notably, 20 per cent say AI operating costs including tokens have constrained further use. Your board is not imagining the line item.
The takeaway for an Australian SME: isolated experiments do not generate measurable returns. Deployment with measurement built in from day one does. Headcount avoidance lets you capture the return from the deployment you are already making, without having to claim AI transformed your business model.
One slide is enough:
If nobody internally owns the modelling behind that slide, that is the work our fractional CFO services do, and the complete guide to fractional CFO costs in Australia sets out what it costs.
How do I justify AI spend to my board?
Use headcount avoidance. Calculate the fully loaded cost of hires you have deferred or removed because AI tooling absorbed the capacity, then present the delta against your AI spend both capped and uncapped. It converts a soft productivity argument into a line-item swap any director can evaluate.
What is headcount avoidance?
Using AI tools to fill capacity gaps that would otherwise require new hires, without cutting existing staff. Money earmarked for salaries and recruitment is redirected to fund tooling that delivers equivalent or greater capacity at lower total cost. It only works if the role was actually budgeted.
What percentage of companies get measurable ROI from AI?
McKinsey's 2026 survey found 37 per cent could attribute any EBIT impact and 6 per cent qualified as high performers, both unchanged from 2025. PwC found 12 per cent of CEOs reported both revenue gains and cost reductions, while 56 per cent reported neither. Estimates of companies achieving substantial return at scale cluster between 5 and 8 per cent across independent surveys.
What is the fully loaded cost of a senior software engineer in Australia?
Roughly $195,000 to $210,000 ongoing on a $160,000 base in Sydney, including 12 per cent super, 5.45 per cent NSW payroll tax where grouped wages exceed the $1.2 million threshold, workers compensation and equipment. Year one runs $220,000 to $240,000 once recruitment at 15 to 20 per cent of base is included.
Is superannuation 11.5 per cent or 12 per cent?
Twelve per cent. The rate reached 12 per cent on 1 July 2025 and remains 12 per cent for 2026 to 2027 with no further legislated increases. The change on 1 July 2026 was Payday Super, which altered the timing and the earnings base, not the rate.
Should I measure AI ROI on time saved or headcount avoided?
Both, but lead with headcount avoided. Time saved is self-reported and hard to verify, and the research on developer self-report is unflattering. Not filling three budgeted roles is concrete and connects directly to the profit and loss. Use time-saved metrics as supporting evidence from pilot teams.
Does saved time count as a saving?
Only if you act on it. Ten engineers saving five hours a week at a loaded hourly rate around $116 is roughly $278,000 of annual capacity, but that is a speed-to-market argument until you defer a hire or cut contractor spend. Say so on the slide. It makes the rest of the case more credible, not less.
How long before AI shows measurable ROI?
Most companies need 90 days of measured data to build a credible case and six to twelve months for headcount avoidance to produce real savings. Median time to positive return across enterprise surveys sits around 14 months. Treat year one as a learning investment with a decision point at day 90.
Could the AI bill exceed the hire we avoided?
Yes, if agents run uncapped. Gartner has documented individual monthly bills above $20,000, and projects coding token costs matching average developer pay by 2028. That is why the slide shows two net positions, and why the cap goes to the board at the same time as the spend.
Scale Suite is a Sydney-based provider of outsourced finance teams and fractional CFO services for Australian SMEs. We deliver weekly bookkeeping, payroll, BAS/IAS lodgement, cashflow reporting, management accounts, and strategic fractional CFO oversight, all as a fully embedded team that works inside your business.
CA-qualified, Xero Certified, and registered BAS Agents, we replace fragmented bookkeepers and once-a-year accountants with one responsive finance function at a fraction of the cost of full-time hires. We serve growing businesses across Sydney, Melbourne, Brisbane, and Perth, with packages starting from $1,500 per month and no lock-in contracts.
See reporting for a raise or a board.
We review and check this guide periodically. At the time of writing (September 2026), all information was current. Scale Suite is a registered BAS Agent, not a licensed tax advisor or financial advisor. This content is general information only and does not constitute professional tax, financial, or legal advice. Some details may change over time. Salary bands, AI pricing and state payroll tax settings all move, so verify current figures before making hiring or investment decisions.
Scale Suite is a Sydney-based provider of outsourced finance and HR services for Australian SMEs. We deliver bookkeeping, financial reporting, payroll processing, fractional CFO support, recruitment, employee onboarding, people and culture support, and fractional HR oversight, all as a fully embedded team that works inside your business.
Employment Hero Gold Partner, CA-qualified, and Xero Certified, we replace fragmented finance and HR processes with one responsive, senior-level function at a fraction of the cost of full-time hires. We serve growing businesses across Sydney, Melbourne, Brisbane, and Perth, with packages starting from $1,500 per month and no lock-in contracts.
30 minutes with our team.
We will review your current finance setup, compare the full cost of an internal hire against our embedded team, and show you exactly what your finance function should cost at your stage of growth.
You will leave with a clear view of what is working, what is missing and where you would save.
No lock-in contracts. 30-day money-back guarantee.
Prefer to book directly? Grab a time here.
