Claude for Work and Accredo: practical AI for the back office

There’s a quiet shift happening in NZ back offices. The general-purpose AI tools the team uses for email drafts and meeting summaries have started reaching into the company file. Not in a sci-fi “the agents have taken over” way — in a concrete “the AP team asks Claude what’s overdue and gets a real answer” way.

This post is a practical look at how Claude — Anthropic’s AI assistant, available across a few different surfaces — fits into a real Accredo business as a co-worker. Where it earns its keep, where it doesn’t, and how to wire it into your company file safely.

The four Claude surfaces, briefly

Anthropic ships Claude in a few different forms. They share the same model under the hood; the surface around it changes.

Claude.ai (web) — the chat interface most people start with. Pro, Team, and Enterprise plans add features like longer context, shared workspaces, Custom Connectors, and admin controls.
Claude Desktop — the native macOS / Windows app. Same chat experience as Claude.ai with deeper OS integration and direct MCP server support in Settings → Developer.
Claude Code — Anthropic’s CLI / IDE assistant for engineering work. Good for shell-driven workflows, scripting, and developer tasks.
The API — for programmatic access from any language, with Anthropic’s commercial terms governing data handling (Anthropic does not train on commercial inputs or outputs by default).

Claude for Work is how Anthropic refers to the Team and Enterprise plans — multi-seat, with admin/billing/SSO controls, shared workspaces (“Projects”), and a single-tenant data boundary. If you’re putting Claude in front of customer data or financials, this is the tier that fits.

What Claude can do with Accredo

On its own, Claude is a good general-purpose AI assistant. Where it gets interesting for an Accredo business is when you give it a typed surface to your company file. That’s what the Rapido MCP server does — it publishes Accredo as an MCP server that any Claude surface (and any other MCP-aware client) can connect to.

Once that’s wired up, Claude can answer questions like:

“What’s our exposure on Mainland Steel right now? Pull AR aging plus open OE drafts.”
“How much RHS-50×50×3 do we have on hand across all locations?”
“Find every overdue invoice over $5,000 and draft chase emails — don’t send them yet.”
“What did we sell to the steel-section trade last quarter, by branch?”

The model writes the natural-language answer; behind the scenes it’s calling the MCP server and reading typed JSON back from your live Accredo company file. (More detail: How to connect Accredo to Claude.)

Workflows by role

A few patterns we’ve seen land in NZ Accredo teams:

Accounts Payable. Reading remittances, matching them to invoices, drafting follow-ups when something doesn’t line up. Claude is good at the matching and the drafting; the team approves and posts. Pairs naturally with Rapido’s PO2Order for the inbound side.

Accounts Receivable. Aging runs, customer-by-customer exposure, drafting tone-appropriate chase emails. The hardest thing isn’t writing the chase — it’s writing the chase the right way for this customer. Claude reads the relationship history and pitches appropriately.

Sales. “What did Mainland Steel buy last year?” “What’s their credit available?” “Who haven’t we spoken to in 6 months in the steel-section trade?” These are questions that used to require an Accredo report or a custom SQL query. Now they’re a sentence in Claude.

General Manager / Finance. Standing dashboards in Power BI / Looker / dbt are still the right answer for repeating questions. Claude is the right answer for the one-off — “why did GP% on the steel range drop 3 points last quarter?” — where you’d otherwise burn an analyst.

Engineering / IT. Claude Code is the developer surface. Same MCP server. Engineers can write integration scripts, ad-hoc reports, MaxBasic helpers, with Claude reading from the live company file as it works.

How NZ businesses are getting started

The pattern that works:

Pick one team and one workflow. Don’t try to roll Claude out to the whole company in week one. Pick AP-and-remittances or AR-and-collections — workflows where there’s a clear human in the approval seat.
Start narrow. Connect Claude to the MCP server. Run real queries against real data for a fortnight. Decide whether it earns its place before turning on any writes.
Then turn on drafts. Drafting (an order, a chase email, a remittance) is where the time savings actually live. The human still approves; Claude does the typing.
Then expand. Once the AP team trusts it, AR sees the receipts, then sales asks for an account, and from there it spreads naturally.

We see most customers get to a stable production workflow in 2–4 weeks of part-time effort. No services-engagement, no 3-month rollout.

Privacy and data handling

Three things to know:

Claude for Work customer data is governed by Anthropic’s Commercial Terms. Anthropic doesn’t train on commercial inputs or outputs by default. If you’re on Pro / Team / Enterprise, this applies to you.
The Rapido MCP server runs on AWS Sydney (ap-southeast-2). Tokens are scoped to your company file; rotate them whenever you want.
Writes are explicit. Specific write paths are turned on per workflow, per token. Nothing is inferred.

If your business has a hard rule against US-based AI providers seeing certain data, the same MCP server works with ChatGPT, or with NZ/EU-based models if that’s your preference. The protocol is the same; the client choice is yours.

A note from the founder

Adam Holt, who runs Rapido, is a volunteer Claude Ambassador for New Zealand and convenes the Claude Code Meetups NZ. We see what’s coming next from Anthropic before most NZ businesses do — and we ship the parts that survive contact with reality. The Rapido MCP server is the result of running these patterns ourselves before we put them in front of customers.

If you’re already using Claude in your business and you can see the Accredo-shaped hole — that’s the conversation we’d like to have.

Book a 30-minute call. We’ll connect Claude to a slice of your company file, run a few real queries, and you decide if it earns its place.