Most Australian businesses move through a predictable sequence as their AI usage matures, from consuming AI features bundled inside SaaS tools, to running a general-purpose assistant like Claude directly, to owning genuinely custom infrastructure. Knowing which stage your business is actually in, rather than which stage sounds most impressive, is what determines whether the next investment makes sense yet.
The four stages, honestly described
Stage one, embedded: AI features bundled inside tools you already pay for, a smart-compose in your CRM, an auto-categoriser in your accounting software
Stage two, direct assistant: using Claude directly for drafting, research and analysis, still largely manual, prompt by prompt
Stage three, connected automation: Claude connected to your real systems via MCP, running defined workflows with approval gates
Stage four, owned infrastructure: custom-built agents, internal tooling, and in some cases self-hosted or fine-tuned components for specific, high-volume needs
Why skipping stages usually backfires
Businesses that jump straight to stage three or four without genuinely working through stage two, direct hands-on use that builds real understanding of what the model does well and where it needs checking, tend to build automations nobody trusts enough to actually rely on, because the team commissioning the build never developed the judgement to specify it properly or catch a wrong output. The maturity model isn't a race to stage four, it's a sequence where each stage teaches something the next stage needs.
Where most Australian SMBs sit today, honestly
The realistic distribution across Australian small and mid-sized businesses in 2026 skews heavily toward stages one and two, plenty of embedded AI features already in daily tools, growing direct use of Claude for drafting and research, and a meaningfully smaller number who've reached stage three with genuine connected automation running reliably. Stage four remains rare outside businesses with dedicated technical resources, and that's appropriate, not a sign of falling behind, since stage four solves problems most SMBs don't actually have yet.
A Newcastle business that progressed deliberately
A twenty-person Newcastle industrial services company spent four months at stage two, staff genuinely using Claude daily for drafting and research, before connecting it to their job-management system for stage three. That deliberate pause meant the team who eventually specified the connected workflow already understood, from months of hands-on use, exactly which parts of the process needed a human check and which didn't, and the resulting automation needed almost no rework after launch, a contrast the operations manager attributed directly to not rushing the stage-two learning period.
How to use the model practically
What the cost curve looks like across the stages
Stage one costs are typically invisible, bundled into SaaS subscriptions you're already paying, effectively zero marginal cost. Stage two adds a direct Claude subscription, typically $30 to $150 a month per active user. Stage three adds one-off connector and skill-building costs, commonly $3,000 to $15,000 depending on scope, plus a modest ongoing usage cost. Stage four is where costs genuinely diverge, ranging from a few thousand dollars for a single custom internal tool to well into six figures for dedicated infrastructure, which is exactly why most SMBs correctly never need to get there.
Mapping your own business honestly against this cost curve, rather than against what a competitor claims to be doing, is the more useful exercise. A business spending confidently and well at stage two is often getting more genuine value per dollar than one that's rushed into an underused stage four build.
A question worth asking before advancing a stage
Before investing in the next stage, a genuinely useful test: has the current stage actually been used well and consistently for at least a full quarter, not just trialled once? A business that's used stage two lightly and inconsistently rarely gets more value from stage three, the underlying habit and judgement gap doesn't close simply by adding more automation on top of it. Consistent, well-used current-stage adoption is the real prerequisite for the next stage paying off, more than budget or technical readiness.
Treat the model as a diagnostic, not a scoreboard. There's no prize for reaching stage four early, and plenty of businesses generate excellent, durable value from a genuinely well-used stage two or three setup for years without ever needing to move further.
Rather than asking 'what stage should we be at,' ask 'what's actually painful right now, and does solving it need the next stage or just better use of the current one.' Most of the value sitting unclaimed in Australian small businesses today is still in stage two, using Claude directly, better, not in racing toward stage four infrastructure that solves problems the business hasn't actually encountered yet.
