How a Small Business Should Actually Start Using AI: The First 90 Days

How should a small business start using AI?

Start with one repetitive, low-risk task and an AI tool you already pay for. Use the assistant built into Microsoft 365 or Google Workspace to draft a quote, summarise a long email thread, or write a first-pass roster, then check how much time it saved. Prove the saving on that single task, write down what worked and what didn't, and only then expand to the next one. You don't need a budget, a developer, or a big platform decision to begin — the first 90 days are about small, measured experiments, not a transformation programme.

In short: pick one task, use tools you already own, measure the hours saved, add a page of simple rules, and decide at the 90-day mark what is worth making permanent. That staged approach keeps the risk low and the learning fast.

Why most small businesses stall before they start

Most owners we talk to have already decided AI is probably worth trying. What stops them isn't doubt, it's the gap between 'we should use this' and 'what do I do on Monday'. Small businesses rarely have a data team, a spare budget for a six-month project, or the time to test twenty tools. The fear of getting it wrong, whether that's entering the wrong information or trusting an answer that turns out to be false, is reasonable and worth respecting.

The fix is to stop treating 'adopting AI' as a single, large decision. Treat it as a series of small experiments instead: one task, one tool you already have, one week to see if it helped. If it does, keep it. If it doesn't, you've lost an afternoon, not a quarter. Adoption here is still early — most Australian businesses have not yet adopted AI in a routine way — which means a well-run experiment is still a genuine edge, not a late catch-up.

The first 90 days: a staged roadmap

Days 1–30: Prove it on one task

Pick a single task that is done often, takes real time, and won't cause harm if a draft needs fixing. Good candidates: writing quotes and proposals, summarising long email threads or meeting notes, drafting supplier or customer replies, or turning rough site notes into a tidy job summary.

Do it in a tool you already own. If you have Microsoft 365, that's Copilot; on Google Workspace, that's Gemini; and the free or low-cost tier of a tool like ChatGPT covers plenty on its own. The point of month one is a single, honest question: did this save time and hold up to a quick human check? Keep the task small enough that one person can own it and report back.

Days 31–60: Write the rules, measure the hours

Once one task works, do two things. First, write a short, plain set of usage rules — one page is enough. Spell out what information staff may and may not paste in, that every AI draft gets a human check before it leaves the business, and who to ask when unsure. Second, extend to two or three tasks and measure. Track roughly how many hours a week the tool is saving across them. Rough is fine; you want a signal, not a research project.

Why bother measuring? Because it turns 'AI feels helpful' into 'AI saves us four hours a week on quoting', which is the number that tells you whether to expand. A 2023 field experiment by economists Erik Brynjolfsson, Danielle Li and Lindsey Raymond found that giving customer-support staff a generative AI assistant raised productivity by about 14% on average, with the biggest gains, up to 34%, going to newer, less-experienced workers. Your numbers won't match a study, but the pattern you're testing for is the same: real, measurable time back, often most of all for less-experienced staff.

Days 61–90: Decide what to formalise, and where a human stays

By the third month you'll know which tasks earned their place. Formalise those: make them the standard way you work, fold them into how you train new staff, and consider a paid plan if the free tier has become the bottleneck. Just as important, name the places where a human must stay in the loop — anything that goes to a client, touches money, or involves personal or health information. Formalising isn't about locking everything down; it's about keeping the good habits and putting a clear boundary around the risky parts.

Where to start by sector

The right first task depends on the work. Some Australian starting points:

• Trades and construction: draft quotes and variation letters from a few bullet points, or turn rough on-site notes into a clean job summary for the client.
• Cafes and retail: first-draft rosters from your usual constraints, and polite, on-brand replies to supplier emails and online reviews.
• Accounting and bookkeeping firms: first-draft client communications and plain-language explanations, or a starting point for workpaper narratives that a qualified person then checks and signs off.
• Clinics and allied health: admin and intake wording, letter templates, and plain-language explanations. Do not put patient-identifying or health information into a consumer AI tool — keep it to general admin text until you have a tool and process that meets your privacy obligations.

In every case, start with one use case, not the whole list.

What to avoid in the first 90 days

• No custom builds. You don't need a bespoke system or an integration project to start. If a vendor is pitching a big build in month one, that's a later conversation.
• No customer, client or health data in consumer tools. Treat anything you paste into a free consumer tool as potentially leaving your control. Under the Australian Privacy Act, personal and health information carries obligations you shouldn't hand away by accident. Keep experiments to non-sensitive text.
• Don't replace a person before you understand the process. AI is useful for the repetitive parts of a role long before it can stand in for the judgement in it. Cut a role too early and you lose the person who knows when the AI is wrong.

Keeping these three lines clear is what makes the whole thing credible to staff and clients.

How to tell if it's working

You don't need analytics to read the signals:

• Hours saved per week. Are the two or three chosen tasks measurably quicker? Even a rough 'about half a day back' is a result.
• Error and rework rate. Are AI drafts needing less fixing over time, or more? Rising rework means the task or the rules need tightening.
• Actual use. Are staff reaching for it without being told? Voluntary use is the strongest sign it's earning its place.

When two of these move the right way, expand to the next task. When they don't, stop, change the task or the tool, and try again.

None of this needs to be perfect. The businesses that get value from AI aren't the ones with the biggest budgets; they're the ones that ran small experiments, measured honestly, and kept what worked. If you'd like a second opinion on where to start, that's what our Fractional AI Officer engagement and AI advisory approach are built for, and you can book an intro call to talk it through.

Frequently asked questions

How much does it cost to start? Often nothing beyond what you already pay. The assistants in Microsoft 365 and Google Workspace, and the free tier of tools like ChatGPT, cover most first experiments. Only pay for an upgrade once a task has proven it saves more than the subscription costs.

Do I need a developer? No. The first 90 days use off-the-shelf tools through their normal interface. You need someone curious and organised, not someone technical. A developer only matters much later, if you decide to build or integrate something, and that's a decision to make with advice.

Is my data safe? It depends on the tool and what you put in. Treat anything pasted into a free consumer tool as potentially outside your control, and never enter customer, client or health information there. Paid business tiers usually offer stronger data handling; check the terms and keep sensitive data out until you've confirmed the tool meets your obligations under the Australian Privacy Act.

What's the single best first task? The one that is repetitive, time-consuming, and low-risk if a draft needs editing — commonly drafting quotes, summarising emails, or writing first-pass replies. Pick the task that annoys your team most and won't cause harm if the first version is imperfect.

How long until I see a return? Many owners see time saved on a single task within the first few weeks. A clear, business-wide picture usually takes the full 90 days, once you've measured two or three tasks and know which to keep.

When should I bring in outside help? When you've run a few honest experiments and either can't get traction, or you're ready to move from one-off tasks to a considered plan across the business. That's the point where a Fractional AI Officer or similar advisory engagement earns its keep.