How to use Claude in your business: AU SMB guide

What is Claude, and what are the four product lines?

Claude is the AI assistant built by Anthropic, and as of 2026 it ships in four product lines that confuse almost every Australian SMB owner.

Most people meet Claude through claude.ai, assume that's the whole product, and miss the parts that actually run a business.

Product	What it is	Best for	Cost (AUD)
Claude.ai	The chat website	Solo research, drafting	Free or ~$30/mo
Claude Code	Terminal agent running scripts and skills	Owner-operators automating	~$30-300/mo
Claude Cowork	Shared team workspace with skills, projects, memory	2-10 person teams	~$45/seat/mo
Claude for Small Business	The packaged 2026 plan (Cowork + admin tools)	SMBs without an IT team	~$70/seat/mo

The answer to "how do I use Claude in my business" depends on which product you start with. Claude.ai is toe-in-the-water. Claude Code is where productivity gains compound. Cowork and Claude for Small Business are how you stop being the only person who knows how. See our BPA guide.

How to use Claude in your business: six use cases that pay back

The honest list of where Claude pays back for an Australian SMB is shorter than the marketing pages suggest, but each use case is worth real money over a quarter.

These six survive the three-month "is this still being used" test in client builds:

Admin and email triage, drafting replies, summarising threads, turning a voicemail into action.
Customer communications, quote follow-ups, booking confirmations, post-job check-ins, complaint responses.
Document drafting, proposals, SOPs, onboarding packs, website copy, training material, e-invoicing templates.
Operations and scheduling, turning messy job notes into records, prepping the next day's run sheet, BAS prep.
Research and decisions, competitor scans, supplier comparisons, sizing a new service line, SEO planning.
Content and ads, first drafts of blog posts, social captions, Google Ads variants.

All six are now augmented by Claude's native Connectors, direct integrations into Intuit QuickBooks, PayPal, HubSpot, Canva, Docusign, Slack, Google Workspace, and Microsoft 365, so Claude reads and writes inside your stack without custom plumbing.

The ones that don't survive: anything client-facing without human review (legal, medical). For trades, our tradie AI guide covers the same six.

How do you set up Claude Code without being a developer?

You don't need to be a developer to run Claude Code, but you do need a weekend to learn the terminal, and that's the gap most owners fall into.

The setup isn't hard. The friction is unfamiliar tooling and no-one to ask when a skill breaks.

The actual setup, top to bottom:

Sign up for a Claude plan that includes Code (Pro at ~$30/mo or higher).
Install Claude Code via a single terminal command, under five minutes.
Authenticate with your Claude login.
Point it at a folder where your business lives (proposals, SOPs, spreadsheets, customer notes).
Build your first "skill", a saved instruction set for a recurring job.

Sample skill prompt you can paste into Claude.ai or save in Claude Code today:

You are a quote follow-up assistant for [BUSINESS NAME].
Context: we send 30-50 quotes a month, 60% never respond on their own.
Task: draft a follow-up email to a prospect who hasn't replied in 5 business days.
Tone: friendly, brief, low-pressure, mention one specific detail from the original quote.
Output: subject line + 3-paragraph email body, sign off as [OWNER FIRST NAME].

Most owners give up between step 4 and step 5, the gap our AI training and done-for-you builds at UnderCurrent Automations close.

Behind the chat: a Claude architecture for production

Once Claude is doing real work in your business, it's no longer a chat window, it's a service running between your stack and your customers, and the architecture matters more than the prompt.

Here's the shape of a typical production Claude workflow for an Australian SMB:

[Inbound lead/email] → [n8n or Make trigger]
       → [Claude (Cowork or API) + Supabase context store]
              → [CRM: drafted reply + tagged task + follow-up scheduled]

Each box is a checkpoint, not a black box. The trigger fires deterministically. Claude operates as an un-monitored microservice with a locked system prompt. Supabase (or any vector store) supplies the context Claude needs without you re-pasting it into every chat. The CRM receives a structured output, not a wall of prose.

A production-grade system prompt looks more like this than the quote-follow-up sample above:

You are a sales-ops assistant for [Business Name], an Australian [vertical] business.

ROLE: triage inbound leads, draft a first-touch reply, log the lead to the CRM.

CONTEXT (pulled from Supabase per lead):
- Lead source, industry, size, original enquiry text
- Past interactions if any
- Active offers + capacity for the next 14 days

FORMAT:
- Subject line under 60 chars
- 3-paragraph email body, plain text, no markdown
- Sign off as [Owner First Name], [Business Name]
- Currency in AUD, dates DD/MM/YYYY, times in AEST

CONSTRAINTS:
- Never invent past interactions
- Never quote a price or guarantee a delivery date
- If a context field is empty, write "[needs input]" rather than guess

OUTPUT: JSON with keys subject, body_text, crm_tag, follow_up_in_days.

The CONSTRAINTS block is what stops hallucinated promises. The OUTPUT schema is what makes the result safe to wire into a CRM API.

What are Australian businesses actually doing with Claude?

Australians use Claude harder than almost anyone else on earth, per the Anthropic How Australia Uses Claude report, but the mix is office, sales, and management work, not coding.

Australia takes 1.6% of global Claude.ai traffic and ranks 11th worldwide; per capita we use Claude at over 4× the rate population alone would predict. NSW (37.2%) and Victoria (30.8%) account for two-thirds of activity. Use mix: 46% work, 47% personal, 7% study. Computer and mathematical tasks sit eight percentage points below the global baseline; workplace correspondence, business documents, and financial guidance run higher. The Australian autonomy score is 3.38 out of 5, we prefer Claude as a co-pilot, not an autonomous agent. The average prompt assumes 11.9 years of schooling and represents work a skilled professional would spend 2.7 hours on, per the Anthropic March 2026 Economic Index. Over 100 Australian companies are confirmed Claude Code users, and the Australian Government signed an MoU with Anthropic in 2026.

How to use Claude in your business safely: the ACSC angle

Before you put a single customer record into Claude, read the Australian Cyber Security Centre's small business AI guidance and the business.gov.au AI page, together they're the closest thing to an Australian compliance checklist.

The ACSC's working rules for SMBs using Claude (and any cloud AI):

Don't paste sensitive data without assessing risk first. PII, financials, health, NDA contracts need a deliberate decision.
Check what the tool stores and trains on. Cowork, enterprise, and API tiers carry stronger controls than the free tier.
Match the plan to your data sensitivity. Sometimes Cowork, sometimes an enterprise contract.
Train your staff first. Most breaches are people, not platforms.
Have an incident process. What if someone pastes a customer list into a free account?

One nuance most SMB guides miss: SMBs under $3M turnover (not health-service or data-trading) are technically exempt from the Privacy Act 1988, per the OAIC's commercially-available-AI guide. But the Privacy Amendment Act 2024 brings automated-decision-making rules in on 10 December 2026, so plan as if they apply.

Government engagement with Anthropic isn't blanket approval (per business.gov.au). Your obligations sit on you. Pair with our service area business explainer if you're remote-first.

What's the real ROI on Claude for an Australian SMB?

Australian SMBs typically spend $200-$500 a month on Claude-plus-automation and recover 30-60 hours of staff time worth $1,500-$2,500, in line with the Anthropic March 2026 Economic Index.

Scenario	Monthly cost	Hours saved	Net monthly benefit
Just Claude.ai Pro	~$30	5-10	~$200-400
Claude Code + 2 skills	~$80	20-30	~$800-1,200
Claude Cowork (3 seats) + workflows	~$350	40-60	~$1,500-2,500
Done-for-you (Cowork + custom skills)	~$500-800	60-100	~$2,500-4,000

Two examples from Flowtivity's Australian SMB automation dataset (third-party cases, neither is a UnderCurrent Automations client): one 12-staff Melbourne plumbing firm spent $350 a month and recovered $12,000 in lost quotes, with quote-win climbing to 31%. Another, an 8-practitioner Brisbane allied-health clinic, cut no-shows down to 11% (a 40% relative reduction), recovering $3,400 a month, consistent with the Anthropic Economic Index September 2025 report. See cost of manual processes and the top 5 SMB automation tools for 2026.

Your first month with Claude: a four-week playbook

Use this four-week playbook if you want a concrete plan you can start tomorrow.

No team rollout, no consultants, no software shopping. Just you, Claude.ai, and one workflow (the shape we use in AI training for SMBs):

Week 1, pick one task. A recurring weekly job that takes you over an hour. Quote follow-ups, weekly report prep, customer email triage, or BAS-prep checks all work.
Week 2, run it through Claude.ai daily. Use the chat tier only. Time yourself before and after. Measure minutes saved.
Week 3, turn the working prompts into a saved "project" on Claude.ai (or a skill if you've upgraded to Code). Add context once, reuse forever.
Week 4, decide. Stay on Claude.ai, upgrade to Claude Code, or get help via UnderCurrent Automations.

If you can't carve four hours a week, skip to a workshop. That's a signal. The same playbook scales for tradie operators and service-area firms losing time to manual work.

What we learned auditing 199 articles on AI in business

We've run 199 articles through the UnderCurrent Automations Article Reviewer rubric (version 2.0.0) across the AI-in-business vertical, and the gap between competent and great is bigger than the gap between bad and competent.

The vertical mean across 118 articles from 55 distinct Australian hosts sits at 60.7 out of 100, competent, but not the work people share or cite.

Our own internal benchmark across 33 published articles averages 87.8 out of 100, mostly because we score every draft against the rubric before publishing and rewrite the failures. The pattern in the bottom half is consistent: thin first-party data, no comparison tables, citations that exist but don't link out, FAQ sections full of marketing questions instead of real search queries, and a generic-AI tone any rubric flags inside ten seconds. The top half does the boring stuff: original numbers, named sources, structured comparisons, real Australian context.

Crawl, walk, run: how UnderCurrent Automations works with clients

This is the approach, not a slogan: every client we put on Claude goes through crawl, walk, run so they're never overwhelmed, and every dollar spent has paid back before the next one goes in.

Three stages, three graduation gates, three pay-back bands we've seen across the work:

Crawl (week 1-4): one owner, one task, Claude.ai chat. Typical lift we see: 5-10 hours a week off a single task, around $200-$400/mo of recovered time at AU labour rates. Graduates when the lift sticks for two weeks straight.
Walk (month 2-3): Claude Code with 3-5 saved skills, owner-led, with light AI training across the team. Typical lift: 20-30 hours a week, $800-$1,200/mo. Graduates when two skills run unattended and a second person uses them daily.
Run (month 4+): Claude Cowork or Claude for Small Business, team-wide skill library, integrated with your business tool stack. Typical lift: 60-100 hours a week, $2,500-$4,000/mo, in line with Anthropic's Economic Index productivity benchmarks and Master of Code's $3.50-per-$1 ROI data.

The point: no overwhelm, no skipped lessons. The 60.7/100 vertical mean above is what happens when SMBs jump straight to run.

Should you DIY, sit a workshop, or have it built for you?

There are three honest ways to put Claude into an Australian business, and the right one depends on whether your time is worth more than the learning curve, per adoption patterns published by Anthropic.

Most owners try DIY first, hit the wall at skill 3, then choose between a workshop and a done-for-you build from UnderCurrent Automations:

DIY (the crawl stage), hands-on owner, comfortable with the terminal. Weekend setup, ~6 weeks before compounding. Cost: the Claude subscription.
Workshop (accelerates the walk), team of 2-15 needing the same mental model. Two-day session, your real workflows, three skills shipped. Cost: a few thousand.
Done-for-you (gets you to run faster), buy the outcome. We scope, build the skills, train your team, monitor month one. Cost: setup fee plus the Claude plan.

Not sure which fits? Book the free 30-min UnderCurrent AI Workflow Audit and Luke will personally map your two-to-three highest-ROI Claude wins on the call, plus tell you straight when DIY is the smarter call (consistent with Anthropic's autonomy data). Worst outcome: a $5k build no one uses. Best outcome: a costed path to your first compounding skill inside the month.

Frequently asked questions

How much does Claude cost for an Australian small business in 2026?

Claude pricing for Australian SMBs ranges from free (Claude.ai basic) to roughly $70 per seat per month for the Claude for Small Business plan, with Claude Code at $30 to $100 per month for individual use and Cowork at around $45 per seat for teams. Most SMBs land between $30 and $250 a month total. Pricing is in US dollars on Anthropic's site, so add roughly 50% for the AUD equivalent before card surcharges.

Is Claude safe to use with customer data in Australia?

Claude can be used safely with customer data in Australia, but only on the right tier and with the right controls. Free Claude.ai accounts may use your inputs to improve models and aren't appropriate for sensitive data; Cowork, enterprise, and API tiers carry stronger data-handling commitments. The Australian Cyber Security Centre's guidance is to treat any AI tool as a third-party data processor, never paste anything covered by the Privacy Act 1988 without a deliberate decision.

What's the difference between Claude.ai, Claude Code, and Claude Cowork?

Claude.ai is the chat website where you type questions and get answers, useful for one-off help but limited as a business tool. Claude Code is a terminal-based agent that reads your files, runs scripts, and builds automations, which is where real productivity gains live for an SMB. Claude Cowork is a shared team workspace launched in 2026 that adds projects, memory, and team skills on top of Claude. Most businesses end up using a combination of all three.

Do I need to be a developer to use Claude Code in my business?

You don't need to be a developer to use Claude Code, but you do need to be willing to learn the terminal and spend roughly a weekend on initial setup. The friction isn't conceptual, Claude Code does the technical heavy lifting itself, but the first three or four skills take a few attempts to get right. Most non-technical SMB owners either sit a workshop or hire someone to set it up before training their team.

How many Australian businesses are actually using Claude?

Australia ranks 11th globally for Claude.ai usage and uses Claude at more than 4x the per-capita rate population alone would predict, according to the Anthropic How Australia Uses Claude report. NSW and Victoria account for 68% of that activity. Over 100 Australian companies are confirmed Claude Code users as of early 2026, including names across banking, design, consulting, and software. Adoption is heavier in office, sales, and management tasks than in coding.

What's the fastest way to start using Claude in a business that's never used AI before?

Sign up for a Claude Pro subscription, pick one recurring weekly task that takes you over an hour (quote follow-ups, weekly report prep, customer email triage), and use Claude.ai to do only that one task for two weeks before adding anything else. After two weeks you'll know whether the productivity lift is real, and you'll have learned enough to make the next call: stay on Claude.ai, upgrade to Code, or bring in UnderCurrent Automations.