What “AI native” actually means for a small accounting firm

Somewhere this month a software rep, a conference speaker or a firm down the road will describe something as “AI native”. If you run a ten-person firm, the honest reaction is a question: what would that actually mean for us, on a Tuesday in March, with a BAS due and a junior off sick?

This explainer is that question, taken seriously. It covers where the term comes from, what is real underneath it, what is marketing, and a ten-question test you can run on your own firm in an afternoon. Dave Sellick has made his own case against the label, and our piece on his deterministic approach covers it; this one is the general test.

“AI native” is mostly a funding word. When the audit technology company Modus announced $85 million in seed and Series A funding on 7 April 2026, CPA Practice Advisor’s report described it as an AI-native audit technology platform. In February 2026 Kacee Johnson and Bebe Kim launched an AI-Native Accounting Foundation, and its announcement did not define the term either. It is used confidently and defined rarely.

Borrowed from “cloud native”, it implies a business designed around AI from the start rather than one with AI added later. For a software company starting from nothing, that is a coherent idea. For an existing accounting firm, with fifteen years of client files, a ledger it does not own and an engagement letter template from 2019, it needs translating.

Strip the marketing off and a useful definition is this: a firm is closer to AI native when its work is designed so that AI can do the routine parts reliably and a person reviews the output, rather than people doing the work and occasionally asking a chatbot for help.

That definition says nothing about which tools you buy. It is about three things you either have or do not.

1. Process redesign

AI speeds up a documented process and makes an undocumented one more chaotic. If each manager does year-end their own way, AI will produce three versions of the same mess faster. The firms getting value have written down how the work is done, then changed the steps to suit what AI is good at: drafting, comparing, summarising and checking against a rule.

Xero’s Modern Practice Playbook survey of 524 Australian senior accountants and bookkeepers, released on 29 September 2026, found 94% of practices using AI day to day rate their core processes as well-documented, compared with 40% of practices not using AI. It is a Xero-commissioned survey and shows correlation, not cause, but the direction matches what you would expect.

2. Systems of record

Every fact the firm relies on needs one home: the ledger for the numbers, the practice management system (Xero Practice Manager, FYI, Kloud Connect or whatever you run) for jobs and deadlines, a document store for workpapers. AI should read from those systems and write drafts back for review. It should not become a second, unofficial place where the truth lives, scattered across individual chat histories.

I haven’t really changed the core of how I run my practice over the last few years.

3. Data access

AI is only as useful as what it can see. A model that can read your job list, client notes and the client’s ledger can draft a useful review. One that sees a pasted paragraph cannot. That access now comes through official connectors and protocols such as MCP, which our MCP guide explains. It also raises the stakes: once an AI tool can read client data, your privacy obligations, your professional standards and your insurer all care how it is set up. Our client data guide covers the controls.

AI needs the full context, and without it, it can’t be that powerful.

Some claims attached to “AI native” deserve a raised eyebrow, whoever makes them.

• “Agents run the work.” AI agents can take actions, and our agentic AI explainer sets out what they do today. A small firm that lets one post journals or lodge without review is carrying a risk no vendor has offered to insure.
• “You need a new platform.” Most of the gain comes from process and data, which you can improve on the stack you already have. A new system can help; it rarely fixes an undocumented process on its own.
• “Everyone else has done it.” The published numbers say otherwise. In the same Xero survey, 85% of Australian practices had tried AI but only 52% used it day to day.
• “It replaces judgement.” Your registration, your name on the return and your insurance do not transfer to a model.

Answer each one yes or no. Be strict: “sort of” is a no.

Self-check

Score your firm: 10 questions

Answer Yes or No to each. Your answers stay in this browser.

1. 1Is each of our five highest-volume jobs written down as a process someone new could follow?

   Yes No
2. 2Does every client fact have one home, and does the team know where it is?

   Yes No
3. 3Does the team use AI through firm business accounts, not personal logins?

   Yes No
4. 4Do we have a written AI policy that says what client data can go into which tool?

   Yes No
5. 5Can our AI tools read our practice data through an official connector, without copy and paste?

   Yes No
6. 6Is there a named reviewer for every piece of AI-drafted work that reaches a client or a regulator?

   Yes No
7. 7Do our engagement letters and privacy notice say we use AI?

   Yes No
8. 8Have we measured the time a job takes before and after AI, on at least one job type?

   Yes No
9. 9Have we changed how we price or package at least one service because AI changed the work?

   Yes No
10. 10Has someone in the firm been given time each week to improve how we use AI?

    Yes No

0 of 10 yes

10 questions to go.

What your score means

• 0 to 3start with questions 1 to 4.
• 4 to 7you are using AI, not yet designing around it.
• 8 to 10you are closer to AI native than most firms that use the phrase.

Copy the questionsStart again

Notice that only one question (number 5) is about technology in the narrow sense. The rest are about how the firm is run. That is the honest answer to what AI native means for a small firm: it is mostly management.

Pick your lowest-scoring cluster and fix that, in this order. Accounts and policy first, because they are quick and they make everything else safer. Then document one high-volume process and redesign it with AI in the loop and a reviewer at the end. Then connect your data. Then reprice the work it changes. One job type done properly teaches you more than ten tools trialled at once, and our first AI project guide covers how to choose it.

And ask the question Dave Sellick put to the June summit audience: “The first question should be: do you need to use GenAI at all?” Often the right answer is a rule, a template or a connector that behaves the same way every time.

Dave returns to the subject at The Firm’s AI in Practice Summit this week: “‘AI native’ is a lie: here’s how I’m actually transforming my accounting practice with GenAI” is on the agenda for 12 November. Notes from the summit will be at thefirm.media/summit/november-2026/notes after the event.

The assumption most likely to age badly is that review stays the bottleneck. If AI tools become reliable enough on routine compliance that review can be sampled rather than total, firms built around AI from day one could run with far fewer people per client, and “AI native” would stop being mostly a funding word. We have not seen independent evidence of that at small-firm scale as at November 2026, and the survey figures quoted here are from a Xero-commissioned study that partly drew on Xero’s own customers. This is general information, not legal or regulatory advice; check your professional body and privacy regulator for your obligations.

Is “AI native” a regulated or defined term?

No. No regulator or professional body we know of defines it, and it is used loosely in vendor and funding announcements. Ask whoever uses it what, specifically, would be different in your firm.

Can an existing firm become AI native, or only a new one?

A new firm can design around AI from day one, which is easier. An existing firm can get most of the way by documenting and redesigning its high-volume processes, but it will always carry older clients, files and habits.

Do we need to change our practice management system?

Usually not first. Check whether your current system has an official AI connector or API, and get your processes and data into it properly. Switching platforms on top of an undocumented process moves the mess.

How many people does an AI-native ten-person firm need?

Nobody has published reliable small-firm data on that. The more useful measure is whether revenue per person rises while quality holds, tracked over several years in your own firm.

Is it risky to let AI read client data?

It can be done safely, on business plans with training switched off, under a written policy and with a reviewer on every output. The risk sits in personal accounts, unclear policies and unreviewed work, not in the idea itself.