Claude for New Zealand businesses.

What Claude is good and bad at for NZ business work

Claude is Anthropic’s AI assistant. For New Zealand businesses its strengths are long documents, careful writing and structured analysis: reading a 200-page tender pack and listing the obligations, turning meeting notes into a board paper, or reconciling two versions of a contract. Every current model, Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, has a one-million-token context window, so large document sets fit in one conversation.

The app has also grown. Claude Cowork, Anthropic’s agentic product for handing over bigger tasks, merged into the main Claude app on 16 September 2026, rolling out first to Pro and Max. Since 8 October 2026, Claude Docs, Slides and Design are out of beta on every plan, including Free.

Where Claude is weaker:

• Usage limits. Every plan has limits on a rolling five-hour window, and paid plans add weekly limits. Chat, desktop, mobile and Claude Code all draw from the same pool, so one heavy user can stall.
• Local knowledge. Like any model, it can state NZ employment, tax or tenancy rules wrongly. Give it the source.
• Processing location. There is no New Zealand or Australian processing option on Anthropic’s own platform.

What Claude costs in New Zealand

Viewed from New Zealand, claude.com/pricing shows prices with a bare “$” and the line “Prices include 15% GST”. It never labels the currency as NZD, so confirm on your first invoice.

Worked example: a six-person practice on Team Standard seats with annual billing pays 6 × $40 = $240 a month as displayed, with GST stated as included. Enterprise has a floor of 20 × US$20 = US$400 a month in seat fees (US$4,800 a year) before any usage, which is then billed at API rates and scales with the model and task. That makes Enterprise cost harder to forecast than Team, so meter a pilot before committing.

How Claude handles your data

• Team, Enterprise and the API: not used for training by default. The exceptions are opting in, or a user pressing thumbs up or down, which stores that conversation for up to five years. Owners can switch the feedback button off.
• Free, Pro and Max: chats are used for training only if the user allows it in the model-improvement setting. Anthropic’s plan table lists model training as “Opt-out” for these plans, so check every account rather than assuming it is off. If allowed, de-identified data may be kept for up to five years.
• Retention: Team and Enterprise chats stay until deleted, then leave back-end storage within 30 days. Enterprise adds custom retention controls. Using Claude Fable requires 30-day retention for safety monitoring by default.
• Location: Anthropic’s own platform offers only “global” or “us” inference and US workspace storage.

For the Privacy Act 2020, the line that matters is whether Anthropic handles your data only for you or also for its own purposes. On Team with feedback off, it is processing on your behalf, and you remain responsible for security and accuracy. A sole trader on Pro with training allowed is in a different position: customer data used to improve Anthropic’s models is used for Anthropic’s purposes, which can make it a disclosure, and an offshore one under IPP 12. Put client work on Team. Developers who need processing kept in the region can use Claude through Amazon Bedrock’s Australian profile, covered in our data residency guide.

Claude with Xero, MYOB and Microsoft 365

Claude connects to other systems through connectors, many built on the Model Context Protocol. Anthropic’s directory listed 974 connectors on 19 October 2026, including Microsoft 365 (SharePoint, OneDrive, Outlook and Teams), Google Drive, Gmail, Notion, Slack, HubSpot and Asana.

• Xero offers a connector in Claude that answers questions from live Xero data with a link back to the source. Xero says it is read-only, so Claude can see your numbers but cannot change them. Xero announced its Claude integration as available at Xerocon London in July 2026. More in AI and Xero.
• MYOB has a read-only MYOB Connector for Claude and ChatGPT. The business owner must allow AI access in My Account first.
• ServiceM8 has an official MCP server you can add to Claude as a custom connector.
• Microsoft 365: Pro and Team include Claude for Microsoft 365. If you instead use Claude models inside Microsoft Copilot, Microsoft excludes them from its EU Data Boundary and in-country processing commitments.

Custom connectors are called from Anthropic’s cloud, not from your laptop. Any MCP server you build must be reachable from the public internet on Anthropic’s IP ranges, which matters if your systems sit behind a firewall. Our guide to MCP for business systems covers the design choices.

Starter project: a month-end debtor review with the Xero connector

This replaces a generic tutorial with one task most NZ businesses repeat monthly: working out who owes money and what to do about it. A debtor list that includes sole traders or individuals is personal information, so use a Team workspace with feedback switched off.

1. Connect Xero. Follow Xero’s instructions to connect the Xero connector in Claude, signed in as a Xero user who already has access to sales invoices. Nothing new is exposed beyond that user’s rights.
2. Write your credit control policy. One page: payment terms, when you send a reminder, when you phone, when you stop work. Save it in a Project called “Month-end”.
3. Run the prompt below in that Project. Use it on the last business day of the month.
4. Check against Xero. Open the aged receivables report and tick off every invoice Claude listed, using its links back to Xero. Note anything missed or wrong.
5. Edit and send reminders yourself. Claude drafts; a person sends. Record how long the whole review took compared with last month.
6. Repeat for three months. Keep it if the list is complete each time and the review is clearly faster.

Using the Xero connector, list every unpaid sales invoice more than 30 days past due as at today. For each give: contact name, invoice number, amount as shown in Xero, days overdue, and the link back to Xero.

Apply the credit control policy in this project to suggest one next step per customer: reminder email, phone call, or escalate to me. Never suggest a step the policy does not allow.

Draft a short, polite reminder email in New Zealand English for each invoice that needs one.

Mark any figure you could not confirm from Xero with [CHECK]. If you give a total, list the invoices it comes from.

Choosing Claude over ChatGPT, Gemini or Copilot

Choose Claude Team when your work is document-heavy, you want the Xero connector, or you have developers who will use Claude Code from the same subscription. Compare it directly in ChatGPT versus Claude and Claude versus Gemini. If your whole business runs on Microsoft 365 and needs AI inside Outlook and Teams, Microsoft Copilot may suit better. Our Claude Code guide covers the coding side.

Limits and failure modes

• One shared usage pool. A developer running long Claude Code sessions can exhaust limits for their own chat too. Give heavy users Premium seats.
• Xero’s developer server is different. The read-only connector cannot change data, but Xero’s separate remote MCP server for developers can write, and its version 1 tools do not ask for confirmation first. It also needs Xero’s Core developer plan (A$35 a month, excluding tax) for real organisations, and Xero’s terms ban using API data to train AI models.
• Fable retention. Choosing Fable means 30-day retention by default, even where other settings are stricter.
• Rollout gaps. Merged Cowork features reached Pro and Max first, with Team and Free following later. Check what your plan has before training staff on it.

If you want the Xero, Microsoft 365 or custom connectors set up with the right permissions from the start, that is part of our AI integration services.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.