How should a small Australian business start using AI?

Clean up your data and processes, then give the team one business AI assistant with a simple usage policy. Improve two or three repetitive tasks and measure the time saved before adding tools.

Is it safe to put customer data into ChatGPT or Claude?

Not into free or personal accounts. The OAIC recommends keeping personal information out of publicly available generative AI tools; use business plans with training on your data switched off, and only share what your privacy obligations allow.

What is MCP and why does it matter for my business?

MCP (Model Context Protocol) is an open standard that connects AI assistants to your other tools, such as your CRM and documents.

Do Australian businesses need to disclose when they use AI?

From 10 December 2026, APP entities must explain in their privacy policy when computer programs make, or substantially help make, decisions that significantly affect individuals using personal information.

Evolving into AI: a practical roadmap for Australian businesses

Evolving into AI is a staged roadmap for Australian businesses that want AI to do real work, not just sit in a browser tab. It runs in five stages: get your data and processes in order, give the team AI assistants with guardrails, automate repetitive workflows, connect AI agents to your stack, then measure and improve.

I've used AI in my own sales role every day for the last few years. AI won't run your business for you, but it will give you back the hours you lose to research, admin and first drafts. Start small and let real use cases build the stack.

Where Australian businesses actually are with AI

Most Australian small and medium businesses are still early. The National AI Centre's SME AI Pulse survey, run by Fifth Quadrant, found SME AI adoption reached 44% in February 2026 (published May 2026).

The same research shows why:

Trust is the biggest barrier. About 65% of non-adopters cited distrust of AI decision-making or a preference for human control.
Relevance is the second. 54% of non-adopters said AI isn't relevant to their business.
Skills are a gap. 19% said they don't know how to use AI.
Industry matters. Fewer than 30% of construction and agriculture businesses are adopting, against more than half in health, education and services.
Guardrails lag. Only about half of current AI users check outputs before they affect customers.

The encouraging part: broad adoption, with AI embedded across several parts of the business, hit a seven-month high. Businesses that commit are going deeper.

The five-stage roadmap

Stage	What you do	Tools that fit	The main pitfall
1. Data and processes	Clean the CRM, document how work gets done	Your CRM, a shared wiki	Automating a broken process
2. AI assistants with guardrails	Give the team a business AI plan and a usage policy	Claude, ChatGPT, Microsoft 365 Copilot, Fathom	Free personal accounts and pasted customer data
3. Automate repetitive workflows	Connect tools so data moves without copy and paste	Zapier, n8n	Building 40 automations nobody owns
4. Agents via MCP	Let AI read and act across your stack	Claude with MCP connectors, OpenClaw	Giving an agent more access than it needs
5. Measure and improve	Track hours saved, quality and revenue impact	Your CRM reports, a simple scorecard	Never checking whether it worked

Stage 1: Get your data and processes in order

AI is only as good as the context you give it. If your CRM is full of duplicates and your process lives in one person's head, AI will produce wrong answers faster.

What to do: pick one source of truth for customers (usually the CRM), clean it, and write down your core processes. Start an AI inventory too: which tools you use and what data each touches.

Pitfall: automating a process you haven't defined. Fix the process first.

Stage 2: Give the team AI assistants, with guardrails

This is where most early time savings live: call prep, first drafts, summaries and research. Choose one main assistant for the business. Claude and ChatGPT are strong general assistants, Microsoft 365 Copilot suits businesses already living in Outlook, Teams and SharePoint, and Fathom summarises your calls.

What to do: move the team onto a business plan, write a one-page usage policy, and turn repeated tasks into skills. A skill is a saved set of instructions the AI loads when a task calls for it: your pitch structure, pricing rules or tone of voice. Claude calls them skills; in ChatGPT the closest equivalent is a custom GPT or a project. Then give the AI memory of your ideal customer, offers and case studies.

Pitfall: set and forget. Most AI tools don't retrain themselves on your data; they improve because you feed them better context. When the output is wrong, correct it and update the skill or memory. That loop is the training.

Stage 3: Automate repetitive workflows

Look for tasks that happen the same way every time: a form fill that should create a CRM record, or a weekly report someone builds by hand.

What to do: list your ten most repetitive tasks, estimate the monthly hours each takes, and automate the top two or three. Zapier is the easiest start; n8n gives more control and can be self-hosted.

Pitfall: automation sprawl. Every workflow needs an owner, or it will quietly break.

Stage 4: AI agents connected to your stack via MCP

The step change comes when AI can reach your tools. MCP (Model Context Protocol) is an open standard that connects an AI assistant to your CRM, documents and data, so it can research, write and update records in one run.

What to do: start with one defined job, such as weekly pipeline review or CRM hygiene, and connect only the tools it needs. Claude supports MCP connectors, and agents like OpenClaw can send emails, edit documents and update your CRM on a schedule, taking instructions through Slack or WhatsApp.

Pitfall: over-trusting the agent. OpenClaw is powerful and free, but it has real access to your accounts, and early 2026 saw exposed installations and malicious skills in its public skill library. Run it on a dedicated machine with separate accounts, and only install skills you've vetted.

Stage 5: Measure and improve

AI is a numbers game like everything else. If you can't show what it changed, you can't justify the next step.

What to do: set a baseline before each rollout, then track hours saved, turnaround time, rework and, where relevant, pipeline and revenue. Each quarter, retire what isn't working and add a tool only when a real use case justifies it. Our stack calculator shows what your tools actually cost.

Pitfall: measuring activity (prompts sent, automations built) instead of outcomes.

The guardrails

Trust is the main reason businesses hold back, so build it in from day one. This is general information, not legal advice.

Keep personal information out of public tools. The Office of the Australian Information Commissioner's 2024 guidance recommends not entering personal information into publicly available generative AI tools.
Check training settings. Use business plans and confirm in each vendor's settings that your data isn't used for training.
Respect the original purpose. Under the Australian Privacy Principles, using existing personal information in AI is generally limited to the purpose it was collected for, unless consent or another exception applies.
Be transparent. Update privacy policies and collection notices where AI touches personal information. From 10 December 2026, APP entities that use computer programs to make, or substantially help make, decisions that significantly affect individuals must describe this in their privacy policy.
Keep a human in charge. Anything customer-facing, and anything involving pricing, commitments or decisions about people, gets reviewed by a person first.
Use the government's frameworks. The Department of Industry's Guidance for AI Adoption (October 2025) includes a foundations version for businesses just starting out.

Your AI roadmap checklist

[ ] One source of truth for customer data, cleaned
[ ] Core processes written down and an AI inventory started
[ ] One business AI plan chosen, with training on your data switched off
[ ] Usage policy shared and skills built for your three most repeated tasks
[ ] Top three repetitive tasks automated, each with an owner
[ ] Privacy policy reviewed for AI and automated decisions
[ ] Human review step for anything a customer sees
[ ] Baseline measures set and reviewed monthly

How Promptwise helps

Promptwise Advisory helps Australian businesses with the people side of this roadmap. I use AI every day across sales and marketing and coach teams to do the same, through coaching and workshops or as a fractional sales and marketing lead. For deeper AI work, I bring in specialists from my network, and for anything that needs building, trusted build partners. Read The Promptwise Approach to AI, or book a free 30-minute chat.

Sources

Adam Agresta is the founder of Promptwise Advisory in Sydney. He has spent 15+ years in sales, marketing and partnerships, including running ActiveCampaign’s ANZ partner channel and building an AI start-up’s sales function from the ground up. More about Adam.