Marketing gets a tidy rule of thumb — roughly 8% of revenue. An AI and ICT budget doesn’t have one yet. Here’s how I’d size a sensible number from your revenue, your sector, and the thing that really decides whether it works: how messy your data and processes are. Spend through the hype, and fix the plumbing first.

Picture it. A board paper in front of you, a vendor quote with “AI” somewhere in the product name, and the board wants a figure for next year. You don’t know if you’re about to throw money at hype or get left behind for standing still. Neither is a great place to decide from.

So let’s be honest: there is no 8% rule for AI. Anyone who hands you a single percentage is selling something. You can still land on a sensible number — but the costliest mistake isn’t getting the figure wrong, it’s answering the wrong question. It isn’t “which AI tool do we buy?” It’s “are we actually ready to make any of it stick?”

Just want the number? Jump to the calculator.

The licence was never the hard part

Software has always been free to copy. Once the thing exists, one more user costs roughly nothing, which is why a thirty-dollar-a-month subscription now does what used to need a junior and a wall of filing cabinets. AI pushes it further: now the building is getting cheap too. The bespoke tool that once cost fifty grand and a three-month wait can increasingly be knocked together in an afternoon.

Which means the thing standing between you and software that helps you is no longer what it costs. It’s whether you can put it to work. And your sector already runs on software whether you call it that or not — rostering, claiming, compliance, ERP, scheduling. So your AI and ICT budget isn’t an optional “innovation” line you drop in a tight year. It’s the cost of staying in business.

You’ve heard of AI. So has the mob down the road.

The real question is whether your competitor is already using it while you’re still deciding. By early 2026 something like 43% of Australian SMEs reported using AI in some form, on the National AI Centre’s tracking — but most of it is shallow. People bang out emails faster and tidy a document; hardly anyone has changed how the work genuinely flows. That shallowness is the opening. There’s a lot of ground still to claim, and most of your competitors haven’t claimed it.

Why does that put pressure on you? Because competition is what drives adoption — one of the sturdier findings in the economics of technology. It also explains why government departments and universities seem so relaxed about all this: nobody’s trying to poach their clients next quarter. A mission-led provider in NDIS, aged care or education doesn’t have that luxury. If you don’t grow into it, someone in your patch will — cheaper, or better, or both.

Right, how much? Start from what you already spend on IT.

Build your AI and ICT budget up from the floor. Before AI enters the conversation, what do businesses spend on IT? Across the board the average sits around 3.6% of revenue, but the range is wild: financial services up near 8%, construction and manufacturing often under 2%, healthcare 3–5%, government 2–5%. For Australian SMEs the quoted benchmark is roughly 2–7% of revenue. Fair warning — and the calculator says the same — there’s no proper published Australian per-industry table anywhere. These are global benchmarks nudged for local conditions. Your own books are the real source of truth.

Now stack AI on top. Don’t treat it as a fixed slice of revenue; treat it as a multiplier on your existing IT budget — somewhere between about 10% and 32% extra over two to three years, depending on ambition. The low end isn’t made up: Deloitte tracks AI rising from roughly 8% to 13% of the tech budget over two years. The high end is the trajectory everyone’s pointing at. Both things are true at once: the spend is real and rising, and a lot of it will be poured straight down the drain — Gartner expects more than 40% of “agentic AI” projects to be cancelled before the end of 2027. Which is the cue for the most important part.

Most of the money isn’t the AI. Fix the plumbing first.

If you remember one line, make it this: the AI licences are the cheapest part of an AI project. The money — and whether it works or flops — lives in your data, your processes, and your people.

Boston Consulting Group has a rule of thumb that’s caught on: 10/20/70. About 10% of the effort on the algorithms, 20% on data and technology, and roughly 70% on people and process — the training, the workflow redesign, the unglamorous slog of getting humans to actually use the thing. The “buy everyone a Copilot licence and call it our AI strategy” instinct gets that ratio backwards.

You’ve probably seen the figure that 95% of AI pilots fail. MIT’s “GenAI Divide” research found only about 5% of integrated pilots pulling out real value — and the gap came down to how people implemented it, not the models. Here’s the way to picture it: getting an AI demo to work 80% of the time is easy. The journey from that 80% demo to a 100% production workflow you’d actually trust is brutal, and it’s filled by your own organisation’s knowledge. Your edge cases. Your “except when the client’s on a transitional plan.” No vendor can sell you that part. You build it.

This is the entire reason RMAI exists, and it’s baked into the calculator. Process first, then AI. When your data and processes are a mess, the sensible AI and ICT budget goes up, because there’s more to fix — and most of that money should buy data and process hygiene, not software.

Spend is lumpy, not a flat line

A practical warning for the board: this spend is front-loaded. Year one is heavy — remediation, integration, change management, training. As a rough planning rule, reckon the first year runs about 1.3× the average; once it’s embedded it settles to a lower run-rate, call it 0.7×. The trap goes both ways: don’t mistake the year-one build cost for the number forever, and don’t try to fund the build year on the settled run-rate — that’s how projects stall halfway. (Go off-the-shelf SaaS instead of building and this flips: flat run-rate, no heavy build year. For plenty of smaller operators, that’s the smart move.)

Isn’t it a bubble? A bit. Spend through it anyway.

There’s real froth, and it’d be insulting to pretend otherwise. But here’s why the bubble question barely matters for you. History is full of investment manias where the speculators got wiped out and the technology still remade the economy. Railway mania ruined a lot of investors — then the railways drove a century of growth. The late-90s telecoms boom laid mountains of “dark fibre,” went bust, and became the backbone of the broadband you’re reading this on. The capital burns; the capability sticks.

So don’t ask “is this a bubble?” Ask three working questions:

1. Does this investment earn its keep even if AI funding dries up next year?
2. Does it set us up to scale if the tech keeps getting better?
3. Could we just buy it cheaper down the track?

If the first answer is yes, the bubble is irrelevant to how you run the place. You’re not buying AI stocks — you’re buying a better way to run your organisation. A workflow that saves your team ten hours a week keeps saving ten hours a week whether or not some valuation in San Francisco holds up. Back the things that compound — clean data, integrated workflows, your team’s fluency — and rent before you build. Spend on the durable layer; stay loose on the disposable one.

Why it’ll be money well spent — especially for you

Cut through the hype and the evidence is strongest in exactly the places mission-led SMEs operate: admin-heavy, compliance-heavy, care-and-service work. The cleanest study going is a randomised trial of more than 5,000 customer-support agents: AI assistance lifted issues resolved per hour by about 15% — and for the least experienced staff, 34%. It didn’t replace anyone; it pulled the newer people up towards the level of the old hands, and customer satisfaction rose with it.

That’s what “saves money, improves services, lets your people do what they’re good at” looks like on the ground. For an NDIS or aged-care provider drowning in claiming and compliance, the win isn’t a flashy chatbot — it’s hours handed back to the people who signed up for the mission, not the paperwork. And with NDIS and aged-care reforms landing through 2026, digital readiness has quietly stopped being a someday-project and become a now-problem.

Get your number. Then pressure-test it.

Enough theory. Here’s what to do.

Get a starting number. Feed the calculator below your revenue, sector, current IT spend, how ambitious you’re feeling, and — the one that matters — an honest read on how ready your data and processes are. It sizes a defensible envelope in dollars and as a percentage of revenue, shows the build-versus-run-rate shape so the board doesn’t mistake year one for forever, and splits it across the 10/20/70 lens. Watch what happens when you tell it your plumbing’s a mess — that’s the whole philosophy in one slider. It’s a sizing aid for a planning conversation, not a quote and not financial advice.

Then pressure-test it against your actuals. Three things you can do today, before you spend a cent on new software:

1. Pull twelve months of statements and tag every software, IT and telco line. Now you know your actual current spend — most owners get a surprise.
2. Score your data and process readiness, and be harder on yourself than feels comfortable. That’s where the budget really gets set.
3. Pick the one workflow you run twenty-plus times a quarter that everyone groans about. That’s your first plumbing job — before any new licence.

When you’re ready to turn the number into a plan, book a call or start with an Operations Assessment.

Process first, then AI. Fix the plumbing before you buy the tooling. We accelerate the thinking — we don’t replace the thinker.