Start With Two or Three Jobs, Not a Strategy
Claude is an AI assistant you talk to in plain English. For a small business, the useful way to think about it is narrow: it is very good at work that is made of language. Drafting things, rewriting things, reading long documents and telling you what is in them, turning messy notes into something structured. It is not a bookkeeper, it is not a database, and it does not know anything about your business until you tell it.
So the sensible way in is not a strategy. It is picking two or three recurring jobs that eat time every week and are made of words, then doing those jobs with Claude for a month and seeing whether the output is good enough to send. Almost every business has candidates: the same quote covering letter written from scratch every time, the forty page contract nobody has read properly, the meeting notes that never get turned into actions, the policy document that has been on the list for two years.
The rest of this guide covers what it handles well, where it will let you down, what Australian businesses need to be careful about with client information, and what a realistic first month looks like. One note before you start: Anthropic publishes current model names, plan tiers and pricing on their own site, and those change often enough that you should check there rather than trust any article, including this one.
What It Is, in Practical Terms
In day to day use, Claude is a conversation. You open it in a browser or a desktop app, type or paste what you want, attach documents if they are relevant, and read what comes back. You can then push back, ask for changes, or ask it to explain its reasoning. There is no setup, no configuration, and no integration required to get value on day one.
A few practical characteristics matter.
It works from what you give it. If you attach the contract, it can answer questions about the contract. If you do not, it will answer from general knowledge, which is not the same thing and is where most bad experiences start.
Each conversation is its own context. Longer running work is easier if you keep a single conversation going rather than starting fresh each time, and the paid plans offer ways to keep reference material available across conversations. Check the current plan features rather than assuming.
It does not act on its own. Out of the box it does not send your emails, update your accounting file, or touch your booking system. It produces text and analysis that you then use. Connecting it to business systems is possible, but it is a genuinely different project with its own governance questions and it is not where a small business should start.
There are free and paid tiers. The free tier is enough to find out whether the work you have in mind is a fit.
The Admin Jobs It Handles Well
Four categories cover most of the value small businesses actually get.
Drafting. Anything you write repeatedly from a blank page: quote covering letters, service descriptions, job advertisements, client update emails, first drafts of policies and procedures, tender responses, website copy. It is fastest when you give it your own previous example and say "in this style, for this situation". A first draft you then edit is a genuinely different task from writing from nothing, and it is usually where the biggest time saving sits.
Summarising and reading. Long documents are where it earns its keep. Contracts, leases, insurance product disclosure statements, tender documents, supplier agreements, long email threads. Ask what the obligations are, what the termination terms say, what changed between two versions, what a clause means in plain English. It will not replace your solicitor on anything that matters, but it will tell you which three clauses to take to your solicitor, which is worth a lot.
Turning mess into structure. A rambling email thread becomes a list of decisions and open items. A page of handwritten job notes becomes a table. A pile of enquiries becomes a categorised summary of what people keep asking for. This is quietly one of the most useful things it does, and the one people think of least.
Spreadsheet reasoning. It is good at explaining what a formula does, finding the logic error in a broken one, writing the formula you are trying to describe, and reading an exported CSV to tell you what patterns are in it. One caution covered in the next section: get it to write the formula and let the spreadsheet do the arithmetic.
There is a fifth use that is harder to measure but real. It works as a second reader. Paste the proposal you are about to send and ask what a sceptical client would push back on, or what you have left out. For an owner who has no colleague to check their thinking, that is a genuine gap filled.
Where It Falls Down
Being honest about the limits is what stops a trial failing for the wrong reasons.
It does not know your business. It has no idea what you charge, who works for you, which suburbs you service, or what your refund policy says, unless you tell it in that conversation or store it somewhere it can reach. Most disappointing first sessions come down to this and are fixed by supplying context rather than expecting recall.
It can be confidently wrong about specifics. Names, dates, figures, clause numbers and citations are all things it can get wrong while sounding entirely certain. The rule that makes this manageable is simple: anything checkable, check. Use it to find things quickly, not as the final word on facts.
Long arithmetic is a weak spot. It reasons about numbers well but is not a calculator, and errors creep in over long columns of figures. Have it write the formula, the query or the method, then run that in the tool built for it.
Its general knowledge has a cutoff date. For anything recent, in particular tax rates, award rates and regulations, treat it as a drafting aid and verify against the source.
Without instruction, the writing sounds generic. Everyone can spot the default register now, and it is not how your business sounds. The fix is to feed it two or three things you have actually written and ask it to match them.
And it does not have judgement about your situation. It will tell you what a clause says. Whether to sign it is your call, or your advisers.
Data Privacy Basics for Australian Businesses
This is the part small businesses skip and then worry about later. It is not complicated, but it does need a decision made deliberately.
Start with whether the Privacy Act 1988 and the Australian Privacy Principles apply to you. Broadly they cover organisations with an annual turnover over three million dollars, but they also cover all health service providers regardless of size, businesses that trade in personal information, and several other categories. Many small clinics and allied health practices are covered even though they are tiny. If you are not sure, the Office of the Australian Information Commissioner publishes guidance on who is covered, and it is worth ten minutes of reading.
Even if you are not covered, your clients still expect their information to be handled sensibly, and a preventable breach damages a small business far more than a large one.
The practical rules are short. Do not paste identified health information, tax file numbers, full bank details, passwords or identity documents into any AI tool. Where you need to work with real material, remove the identifying details first, because for most tasks the AI does not need the name to do the work. Read what your plan says about whether your inputs may be used to improve the models, because consumer and business tiers commonly differ on this and the answer should drive which one you buy. Know where the data is processed and stored, and ask for that in writing if it matters to your obligations or your clients.
Then write it down. One page is enough: what may go in, what must never go in, who in the business is allowed to use it, and what to do if someone makes a mistake. Tell your team, and put it somewhere they will find it. Most privacy incidents in small businesses are not malice, they are a well meaning staff member pasting something into a tool because it was the fastest way to get an answer, without ever being told not to.
If your work involves health records, legal files, or anything covered by a professional obligation, check the guidance from your professional body and what your insurer expects before you roll anything out broadly.
How a Realistic First Month Runs
A rollout that works tends to look the same in every business, and it is smaller than people expect.
Week one is one person and two jobs. Pick whoever is most comfortable with new tools and give them a specific brief: use it for these two tasks only, keep the output, keep the failures. Trying to use it for everything at once produces an impression rather than evidence.
Week two is turning what worked into something reusable. When a set of instructions produces a good result, save it, along with the context that made it work. A shared document with a dozen working prompts is the difference between one person being good at this and the business being good at it.
Week three brings in two or three more people, using the saved prompts rather than starting from scratch. Expect this week to surface the practical questions: who pays for seats, what goes in the shared prompt file, what happens when someone gets a bad answer.
Week four is the privacy page and the decision. Write the one page policy from the previous section, then look at the evidence: which tasks got faster, which produced output good enough to send, which were abandoned and why. Decide what to keep, what to stop, and whether to pay for more seats. Months two and three are for expanding into a few more tasks and checking whether the early wins held up once the novelty faded. Some do not, which is useful to know.
One thing that separates rollouts that stick from those that fade: someone has to own it. Not a committee, one person who keeps the prompt file current and is the one people ask when they get a poor result.
Getting Better Output
The single biggest difference between people who find it useful and people who do not is the quality of the instruction. A few habits are worth teaching everyone on day one.
Say who it is for and what it is for. "Write a quote covering letter" produces something generic. "Write a quote covering letter to a facilities manager at a mid sized aged care provider, for a $14,000 electrical maintenance job, warm but businesslike, under 200 words, ending with a clear next step" produces something you might actually send.
Give it your own examples. Two or three letters you have sent before will do more for the tone than any amount of describing the tone.
Give it the source material rather than relying on its memory. If the answer depends on the contract, attach the contract. If it depends on last quarter numbers, paste the export.
Say what you do not want. Length limits, banned phrases, no bullet points, no marketing language. Constraints improve output more than encouragement does.
Work in rounds. The first response is a draft. Tell it what is wrong and ask for a revision. Two or three rounds usually beats one long, elaborate instruction, and it is faster than it sounds.
Ask it to critique its own work. Questions like what is weak about this draft, or what a sceptical reader would challenge, often catch the thing you would have noticed only after sending.
And when something works, save the instruction. That is the whole compounding mechanism.
How to Tell If It Is Worth Keeping
Judge it on named tasks, not on general impressions, because the general impression is always more positive than the reality.
Before you start, write down roughly how long each of your chosen tasks takes today. Not precisely, just honestly. Two hours a week on quote letters, half a day a month on the board report, whatever it is. Without that baseline you will be comparing against a feeling in four weeks.
Then track three things for the month. How long the task takes now. Whether the output was good enough to use with light editing, heavy editing, or not at all. And how many attempts were abandoned because the answer was wrong or unusable.
That third number is the one people forget, and it is the one that tells you whether a task is genuinely a fit. A task that produces a usable result eight times in ten is a keeper. A task that needs three attempts and still gets rewritten is costing you time while feeling productive.
The cost side is simple. Compare the monthly cost of the seats against the hours saved, valued at what those hours are actually worth in your business. If it does not clear that bar, that is a legitimate result worth acting on rather than persisting out of a sense that you ought to be using AI.
One quality check worth building in: for any output that goes to a client, someone who knows the business reads it first. Not forever, but certainly for the first few months. The failure mode you are guarding against is not bad writing, it is confident, plausible detail that happens to be wrong.
When to Get Help
For the use cases in this guide, you do not need help. A capable person with a paid seat and a few hours of practice will get most of the value, and paying someone to teach you to write prompts is rarely money well spent.
There are three situations where outside help genuinely earns its cost.
The first is connecting AI to your business systems. The moment you want it reading from your job management software, writing into your CRM, or answering customers without a person in the loop, you have moved from a productivity tool to a system that needs design, testing, error handling and someone accountable when it gets something wrong. That is a project, not a subscription.
The second is consistency across a team. One person getting good results is a personal productivity gain. Fifteen people getting reliably good results on the same tasks requires shared instructions, agreed standards for what gets checked, and a decision about what is allowed to leave the building unreviewed.
The third is anything with a compliance edge. Health records, legal files, financial advice, or personal information at scale. The technology question is usually the easy part and the governance question is the real work.
If you are somewhere in that territory and want a straight assessment rather than a pitch, our consultations are free for businesses with 20 or more full time staff, and $200 including GST for smaller teams. And if the honest answer is that you should spend a month with a paid seat and a list of tasks before spending anything with us, that is what we will tell you.