How much should we set aside for AI & ICT upgrades?
Marketing has a tidy rule-of-thumb (~8% of revenue); AI has no equivalent yet. This sizes a defensible envelope from your revenue, sector, and — the part that actually matters — how ready your data and processes are. Fix the plumbing first.
The AI licences are the smallest slice. Most of the money is change, training and data work — the part that decides whether the tooling actually sticks. We accelerate the thinking; we don’t replace the thinker.
How this is calculated & what it’s anchored to
Baseline IT intensity by industry. Selecting a sector seeds a current-IT-spend default from published benchmarks: construction and manufacturing run ~1–3% of revenue, healthcare ~3–5%, government ~2–5%, financial services ~4–11%. Cross-industry the average sits ~3–6%, with larger firms benefiting from scale at 2–4%. These are global benchmarks (Computer Economics/Avasant, Deloitte, Gartner) — no public Australian per-industry IT-spend table exists, so treat them as directional and AU-adjusted.
The AI overlay. Rather than treating AI as a fixed slice of revenue, the tool applies a 10–32% multiplier on top of your existing IT budget over a 2–3 year transition, scaled to ambition. The low end brackets what is actually happening — Deloitte tracks AI rising from ~8% to ~13% of the tech budget over two years, and a 2024 survey (ISG) had AI spend planned up ~5.7% while overall IT budgets grew under 2%. The aggressive end reflects the trajectory: AI moving from under 4% of enterprise tech spend today toward ~23% by 2035 (Oxford Economics, 2026).
Plumbing-first adjustment. Readiness does two things. When it is poor, the envelope is nudged up (more remediation needed) and the majority is steered toward data and process work rather than tooling. This matches the evidence on why AI projects fail: RAND (2024) finds the leading causes are leadership and unclear business problems, with data quality a consistent top-two cause — not model quality or compute. For SMEs the binding constraint is evaluation, setup, integration and data hygiene, not licence cost.
The 10 / 20 / 70 split. BCG’s now-standard framework (coined by Sylvain Duranton, 2019): ~10% on the AI itself, ~20% on data and tech, ~70% on people and process change. It is the antidote to the “just buy Copilot seats” instinct.
Build vs run-rate. Year 1 is weighted ~1.3× the transition average (front-loaded remediation and change); run-rate settles to ~0.7× once capability is embedded. The front-loaded shape is well-evidenced (ERP TCO studies); the exact multipliers are planning heuristics, not forecasts. Off-the-shelf SaaS tooling inverts this — flat run-rate, no heavy build year.
Figures are directional planning aids, not financial advice. Industry bands are global benchmarks (no Australian per-industry table is published). Validate against your own GL, contracts and a real readiness assessment before committing budget.