You don't need to replace your stack. You need a thoughtful layer on top.

A clinic owner from Whangaparāoa rang us last month. A consultant had quoted her $42,000 to "implement an AI-powered practice management platform". She wanted a second opinion.

Twenty minutes of conversation revealed two things. Her existing system — Cliniko — works fine and she likes it. Her actual problem is that her front desk spends two hours every morning typing yesterday's voicemails into the booking system.

She did not need a $42,000 platform. She needed a $2,500 layer.

This post is about that distinction.

The replatform pitch

The shape of the bad pitch goes like this:

"Your current system is from 2018. The new wave is AI-native. You can either bolt AI on top of legacy software, or you can replatform onto a modern stack that has AI built in. We recommend the latter. Here is our quote."

Two things wrong with this pitch.

First, your current system probably isn't legacy. Xero, MYOB, Cliniko, Vend (now Lightspeed), Workflow Max, simPRO, Timely, Ascend — the tools small New Zealand businesses actually use — have all shipped serious AI features in 2024–25. Xero alone acquired Syft in 2024 and integrated its analytics into the core product for millions of customers within a year ¹. JAX, their financial assistant, launched the same period ².

Second, replatforming has a cost the quote rarely mentions: you have to migrate data, retrain staff, rebuild every workaround you've spent five years getting right. The 2025 Datacom State of AI Index found that 22% of organisations cite data quality or integration as a main barrier to AI adoption — i.e. the most common technical problem isn't the AI itself, it's the connection between AI and the data you already have ³.

Replatforming makes that problem worse before it gets better.

What "a layer on top" actually means

The honest opportunity for most NZ SMEs is small and specific. It looks like this:

        ┌─────────────────────────┐
        │   Your existing system  │
        │   (Xero, Cliniko, etc.) │
        └───────────┬─────────────┘
                    │
                    │  read & write
                    │  via API
                    │
        ┌───────────▼─────────────┐
        │     AI translation      │
        │  (chat, summarise,      │
        │   extract, classify)    │
        └───────────┬─────────────┘
                    │
        ┌───────────▼─────────────┐
        │       Your team         │
        │   (still in charge)     │
        └─────────────────────────┘

The AI doesn't store your data. Your existing system does. The AI doesn't make decisions. Your team does. The AI translates between the two — taking messy human input (a voicemail, an email, a photo of a receipt) and producing structured output your system already understands.

Three patterns that we build over and over again:

Pattern 1 — The natural language report

Your bookkeeper opens Slack and types: "Which customers are more than 30 days overdue and haven't been chased this week?"

A small assistant queries Xero, cross-references your chase log, and answers in plain English. Nothing in your system has been replaced. You just got a new way to ask a question you used to need a report for.

Justin Gray, Datacom NZ's MD, put it well in 2025: "We're seeing organisations starting to think about... the interfaces between their existing applications and AI." ³ That's exactly the work. The interface is the project.

Pattern 2 — The smart inbox

Customer emails arrive in a shared inbox. An assistant reads each one, classifies it ("new enquiry", "complaint", "supplier invoice", "spam"), drafts a reply if it's confident, and routes the rest to the right person.

Your team still hits Send on every reply. The assistant just removed the worst part of the morning — the triage.

Pattern 3 — The auto-extraction

Your tradesperson takes a photo of a job sheet at the end of the day. An assistant reads it, extracts the materials list, the labour hours, the address, and creates a draft invoice in your existing system. You check the draft on your phone on the drive home. One tap to send.

Note what just happened: the photo didn't replace your invoicing system. It fed it.

What this costs

The AI Forum's third productivity report, released August 2025, found that 75% of New Zealand AI setup costs are now under $5,000 ⁴. That's the order of magnitude you should expect for any of the three patterns above, provided the underlying system already exists and has an API.

A clean rule of thumb we use internally:

• If your existing tool has an open API and good docs → a layer is small (typically $2,000–$6,000).
• If your existing tool has a partial API → still doable, slightly more discovery ($4,000–$10,000).
• If your existing tool has no API at all → consider whether the tool needs to be replaced first, but for different reasons than "it's not AI-native".

Most of the systems Kiwi SMEs use have good APIs. Xero's is excellent. MYOB's is solid. Cliniko's is well-documented. The "legacy" critique doesn't hold up.

When replatforming is the right call

To be fair: there are situations where the consultant in the opening paragraph would have been right. If your business runs on a custom-built system from 2009 that nobody supports, or on a stack of disconnected Excel files, or on a vendor who's been acquired and is winding down support — yes, replatform. But replatform because the foundation is failing, not because the surface isn't "AI-powered".

The conversation we'd rather have

If you've been quoted a five-figure number for an "AI transformation" and something about it feels off, send us the proposal. We'll tell you, in writing, what we think the smaller version of the project looks like. No charge. We do it because we'd rather your $42,000 went on a holiday than on a platform you didn't need.

If the smaller version looks like work we can do, we'll quote it. If it doesn't, we'll send you to someone who can.

The clinic owner from Whangaparāoa got a $2,400 voicemail-to-Cliniko integration. It runs every morning before her front desk arrives. She kept her existing system, her existing workflows, and most of her $42,000.