What Claude Actually Does For A Business

If you’ve been hearing the name Anthropic Claude thrown around and wondering whether it’s just another chatbot, the short answer is no, but the longer answer matters more for your business. Claude is a large language model built by the US-based company Anthropic, and what separates it from the older wave of AI tools is how well it handles long documents, nuanced writing, and step-by-step reasoning.

For a New Zealand business owner, the practical question isn’t what the model can do in a benchmark. It’s what it can do in your Tuesday afternoon. Can it draft the reply to a difficult client email. Can it read through a 40-page lease and pull out the bits that matter. Can it turn your meeting notes into something the team will actually read.

We typically see three buckets of use from the small and medium businesses in our network. The first is writing and communication, anything from proposals to internal updates. The second is research and synthesis, which is where Claude’s longer context window earns its keep. The third is workflow support, where Claude is paired with other tools to handle repetitive processes.

What it’s not is a fully autonomous employee. It doesn’t log into your Xero file and file your GST for you. It doesn’t replace the judgement of a good accountant or lawyer. Treat it as a capable junior who reads fast, writes well, and never sleeps, with all the caveats that come with that analogy.

The NZ Privacy Question You Can’t Ignore

This is the part most offshore blog posts skip, and it’s the part that should matter most to you. New Zealand’s Privacy Act 2020 sets out 13 Information Privacy Principles, and the one that catches NZ businesses off guard with AI tools is IPP 12, which covers disclosure of personal information outside New Zealand.

When you send a customer email, a staff file, or a client contract to a US-based AI service, you’re potentially disclosing that information offshore. That isn’t automatically a breach, but it does mean you need to be confident the provider handles the data appropriately, and that you have a lawful basis for sending it.

The Office of the Privacy Commissioner has been increasingly clear that businesses remain accountable for personal data regardless of where it ends up. For most small businesses, this means three practical things. First, check whether Claude or whatever tool you’re using stores or trains on your inputs, and turn off training where the option exists. Second, think hard about what you paste in. Don’t feed it client files, employee health notes, or anything covered by a confidentiality clause without checking the terms. Third, document your decision. A one-page internal note on how your team uses AI tools will save you a headache if someone ever asks.

If you’re in healthcare, you’ll also need to consider the Health Information Privacy Code 2020, which adds another layer. If you’re in financial services, the FMA and Reserve Bank guidance around outsourcing and operational risk are worth reading. This is general guidance, not legal advice, so verify the specifics with your lawyer before you build a workflow around any tool.

What It Costs In Real NZD

Anthropic’s pricing changes reasonably often, so treat any number I give you here as a snapshot rather than gospel. The free tier exists, but for business use you’ll want a paid plan. The Pro plan is roughly USD 20 per user per month, which lands around NZD 33 at recent rates. The Team plan is around USD 30 per user per month, or about NZD 50. The higher-end plans for bigger workloads run from there.

Convert those to annual subscriptions and you’re looking at somewhere between NZD 400 and NZD 800 per user per year, depending on tier and usage. For a five-person team that’s a real line item, but it’s a fraction of the cost of a full-time hire, which is the comparison that actually matters.

The other cost people miss is API usage. If you build Claude into a workflow, you pay per token, which is per piece of text processed. For light internal use this is usually a few dollars a month. For a busy operation running thousands of queries, it can climb into hundreds. Build a rough usage estimate before you commit, and put a spending cap in place. Approximate conversions here are USD 1 to NZD 1.65 as a rough guide only, so verify current rates before you budget.

One thing worth knowing. Unlike some of its competitors, Claude doesn’t have a “pay per feature” structure. You pay for the plan and get access to the model. Where you spend extra is on the API side, which most small businesses don’t need for the first year.

Where Kiwi Businesses Are Getting Real Value

Let me walk you through how businesses in our network are actually using it, because the gap between vendor promises and Tuesday-afternoon reality is real.

Accountancy practices in Auckland and Wellington are using Claude to draft client memos, summarise IRD documents, and prepare first-draft responses to Inland Revenue correspondence. One accountant I spoke with recently told me it cuts their letter-writing time by more than half, but they still review every single line before it goes out, because the sign-off is theirs.

Legal firms are using it for contract review, particularly the first pass on commercial leases and supplier agreements. A Sydney-based commercial lawyer I work alongside uses it to flag unusual clauses before he does his own reading, which means his billable hours go further. Note for NZ readers, your own professional rules around supervision and disclosure still apply, so check with the New Zealand Law Society before you build a process around it.

Tradies and small construction businesses are using it to draft quotes, write job descriptions for Seek, and turn site notes into client updates. Real estate agents working with platforms like REA Group and Trade Me are using it to write listing descriptions faster, though the best results still come from agents who edit ruthlessly.

Retail businesses on Trade Me are using Claude to write product descriptions, respond to common buyer questions, and draft returns policies that don’t put them at risk. None of this is glamorous, but it’s where the actual time savings live.

Connecting Claude To Your Existing Tools

The clever part, and where most small businesses underuse what they pay for, is connecting Claude to the software you already run. If you live in Xero, MYOB, Trade Me, Seek, or REA Group’s backends, the question isn’t whether Claude can talk to them. The question is how you bridge the two.

For most small businesses, that bridge is either Zapier, Make, or a bit of careful prompting. You can build a workflow where a new Trade Me enquiry gets summarised and drafted into a reply, where meeting notes from your project tool get turned into action items, where a job ad draft from Claude gets posted to Seek with a final human check.

This is where the productivity gains compound, but it’s also where the privacy risk grows. Every new connection is another place personal data could end up. Build a quick list of what data flows where, and review it every quarter. The Privacy Commissioner’s expectation, broadly speaking, is that you know where your data is, even if you didn’t build the systems yourself.

If you’re in healthcare or financial services, the regulatory expectations around outsourcing and data handling are higher. APRA’s CPS 234 in Australia sets the bar for information security, and while it doesn’t apply directly to NZ, it’s a useful benchmark. For NZ, look at the FMA’s outsourcing policy guidance if you’re in scope.

Picking The Right Plan And Rolling It Out

Most NZ business owners I work with start with one seat, prove the value, then expand. That’s the right way to do it. Don’t roll Claude out to 20 people based on a demo. Pick one person, give them a real task they currently find painful, and measure what changes.

For a sole trader or very small operation, the Pro plan is plenty. For a team of three to ten, the Team plan starts to make sense because of the shared workspace and admin controls. Beyond that, you’re into enterprise territory where the conversation shifts to contracts, security reviews, and procurement.

The other thing to decide early is your internal policy. You don’t need a 30-page document. A one-pager that says what Claude can be used for, what it can’t, what data stays out, and who to ask if you’re unsure is plenty. Most of the businesses I work with have something they wrote in an afternoon and update twice a year.

The Risks Worth Naming Out Loud

Claude hallucinates, which is the technical way of saying it makes things up confidently. It will give you a case citation that doesn’t exist, a section of an Act that isn’t there, and a number that sounds right but isn’t. For most business uses this is a minor inconvenience. For anything that goes to a regulator, a court, or a client as fact, it matters enormously.

The second risk is over-reliance. We see business owners who start using AI for everything and find, six months in, that they can’t write a client email without it. That’s not a productivity gain, it’s a skill loss. Keep the muscle working, use Claude to draft but not to decide.

The third is vendor lock-in. If your workflows are deeply tied to one AI provider, switching becomes painful. Build with portability in mind, keep your prompts and templates in a place you control, and don’t make a tool central to a process that you can’t easily reverse.

The fourth is the regulation catching up. The Privacy Commissioner, FMA, and various industry bodies are all working out how to handle AI. What’s permissible today might be tightened in two years. Build your processes so they can adapt.

When To Bring In Outside Help

If you’re a five-person team using Claude for email drafts and meeting notes, you probably don’t need a consultant. Read the basics, write your one-pager, and get on with it.

If you’re a 30-person operation with compliance obligations, customer data flowing through the system, and a real risk profile, the conversation is different. You want someone who understands both the AI side and the local regulatory environment, not just a vendor selling you seats.

This is where Enterprise DNA sits. We work with NZ and Australian businesses on exactly this kind of rollout, the boring, important work of making AI tools actually useful inside the constraints you operate under. We won’t sell you more than you need, and we’ll tell you when a tool isn’t the right fit.

A Sensible Next Step

If you’ve read this far, you’re probably closer to using Claude well than most of your competitors. The businesses that get the most from AI in the next two years won’t be the ones that buy the most licences. They’ll be the ones that pick the right handful of workflows, set sensible guardrails, and iterate.

Start small. Pick one painful task. Run it through Claude with a clear prompt. Check the output like you’d check a junior’s work. If it’s better than what you had before, do it again tomorrow. If it isn’t, adjust the prompt or pick a different task.

We’ve put together a free field guide that walks through this in more detail, including prompt templates, a sample internal policy, and the rollout sequence we use with clients. It won’t make you an expert in an afternoon, but it will save you the worst of the trial and error.

Enterprise DNA works with NZ and AU businesses on this challenge. Get the free Working With Claude field guide at https://enterprisedna.co/resources/working-with-claude?utm_source=edna-landing&utm_medium=blog&utm_campaign=nzau