The systems map: the one page that shows where AI can and cannot go in your organisation

Summary

This article gives you the systems map: one table, one row per system, that shows what you run, who owns each system, where its data sits and where AI is switched on. It shows how to build one in three passes, how to keep it, and what it tells the board. It is one of the Putting AI to Work series for Australian businesses and organisations, and the other articles in the series are mapped below. It is written for small, medium and large organisations alike.

Skip to the Systems Map Template ↓

Background

An organisation that puts AI into its software stack has a solution architecture whether or not anyone has drawn it: the systems it runs, what each is the official home of, what connects to what, and where AI can now read and write.

Boards and executives have to understand it in a way they never had to before, because AI features arrive inside ordinary software, switched on by the vendor rather than bought, and because AI tools can now read from and write into the accounting system, the customer relationship management (CRM) system and the email suite. The systems map is the first page of that architecture, and implementing an AI policy is the moment to draw it. The policy cannot name the approved tools until somebody has listed the systems they sit in.

Three Australian sources ask you to know where AI is used. The first governance question the Australian Securities and Investments Commission (ASIC) asks licensees (the financial services and credit businesses it licenses) is “Do you know where AI is used in your organisation?”¹ The Australian Institute of Company Directors (AICD) opens its checklist for small and not-for-profit boards with “Understand where AI is already being used in the organisation, including AI features in third-party products and services.”² The National AI Centre’s answer is “Create and maintain an AI register”, and its template for one has twelve columns.³

None of them records where a system’s data is stored, when its subscription renews, or whether the AI in it can write into another system.⁴ The map adds those. A business of a few dozen people keeps the map instead of a separate register. A large organisation keeps both, and the map is the index.

The columns were first set out in Choosing AI software and connecting it to your systems, which also carries the nine vendor questions and the six rules before AI writes into a core system. This piece is the map itself.

Checked against the National AI Centre’s register page and template, its Guidance for AI Adoption, ASIC’s Report 798, the AICD’s director’s guide and board checklist, the Australian Signals Directorate’s AI and cloud guidance for small business, and the OAIC’s privacy principles guidelines on 6 November 2026. This covers Australian businesses and organisations of any size that use AI tools bought from a vendor, not businesses that build them. It is not legal advice.

Methodology

A long article made quickly with AI can be slop, and the risk is real. We would rather say how this one was made than have you guess.

1. The questions came first. The issues this article addresses are ones the author has worked through with clients facing them, in his consulting practice.
2. AI did the research and the first draft. Claude Code, running Claude Fable 5.1, read the primary sources and drafted the article to the house standard: every claim tied to a source, and every source opened by a person before the article leaves draft. Where an article says "we read" a document, or all 121 of something, the model read every one of them in full and did the counting; the author checks the method and a sample, not every page.
3. The author did the first review. The author read the draft end to end, cut what did not hold, and edited the body.
4. Other AI tools checked it. A second and third model read each article for errors and for one model's blind spots. This article has not had that check yet.
5. Specialists review it last. Reviewers across technology, law and human resources read the final draft. This article has not had its specialist review yet, and no law firm has reviewed it.

Why use AI at all, given the risk? Because at the rate AI is moving it would be remiss not to use the top frontier models to synthesise the current state of play, so that a business decides from the most up-to-date information there is. Feedback and corrections are welcome, and so are volunteers, law firms among them, who want to make this the best resource available to Australian businesses: write to the editor.

The map is one table, with a row for every system

The map is a table, kept where everyone can see it, with one row for each system the organisation runs: the accounting system, the CRM, the enterprise resource planning (ERP) system or job management system, the online store, email and documents, payroll and HR, the meeting tool, and the smaller tools around them. The columns are the nine from Choosing AI software and connecting it to your systems, unchanged:

Column	What goes in it
System	Its name and plan
Source of truth for	The records it is the official home of: invoices, customers, stock, prices, staff records. Each kind of record has one home
Owner	The person who knows when it is wrong and decides about it, not necessarily the person who logs in most
Administrator	Who adds and removes users and connections
Personal information	Whether it holds information about customers or staff, and what kind
Where the data is stored	The country, from the vendor’s terms
AI features	Which are switched on, and whether they read or write
How other tools reach it	Export, API, connector, or not at all
Renewal	The date, the notice period, and what it costs a year

Two words in that table need explaining. An application programming interface (API) is a door a software product opens so that other software can ask it for data or give it instructions. The product decides what the door allows, which may be to read a record, to write one, or both. A connector is a ready-made link that uses that door for you, so a general-purpose assistant or an automation can reach the product without anyone writing code. It carries only the permissions it was set up with, like a key cut for one room.

The map and the National AI Centre’s register template do different jobs. The template’s twelve fields are name and version, owner, status, source, purpose and business goals, intended use cases, known limitations and prohibited use, foreseeable misuse, data sources and type, registered date, screening outcome, and the people affected.⁴ Nine of the twelve describe one AI use: why it exists, what it must not do, how risky it is.

Two of the map’s columns match the template’s (system and owner) and one overlaps in part (personal information, against “data sources and type”). The other six, source of truth, administrator, where the data is stored, read or write, how other tools reach it, and renewal, appear in no federal column. They describe a system rather than a use, and they are what a board needs when its question is “what can this AI reach?” rather than “is this AI safe?”.

In Australian practice the whole picture is the organisation’s solution architecture, singular: the systems, the records each is the home of, the connections between them, and now the AI that reads and writes across them. In a company of a few hundred people somebody in IT keeps that picture. In a business of a few dozen people nobody does, and the map is the whole of it. Either way it is the first page, and everything else in this series points at it.

Build it from the money, then the logins, then the switches

A map built from memory misses the tools nobody remembers paying for and the ones nobody admits using. Build it in three passes, in this order, a week each.

Pass one: what the organisation pays for. The accounting system already lists most of it. Pull every repeating bill, direct debit and card transaction for the last 12 months, because annual subscriptions appear once, and sort the software from the rest. Add the subscriptions a staff member pays on their own card and claims back, and the ones bundled into an outside IT provider’s invoice.

Each becomes a row with a renewal date, a notice period and a yearly cost before you know anything else about it. The Australian Signals Directorate (ASD) tells small businesses to “review the AI vendor’s terms and conditions related to data ownership, protection, usage and storage”, and this pass is where you find out how many vendors that means.⁵

Pass two: what staff log into. Ask every person, in writing, what they sign into for work, including on their phone and including anything free. The single sign-on list (the one company login that opens the other systems), where you have one, is the answer to check against.

This pass finds the shadow tools: a free meeting note-taker two people installed, a personal chat assistant used for drafting, a spreadsheet add-on someone built a report on. The AICD’s 2026 guide says “employees may also be using AI tools on a ‘shadow’ basis, without formal oversight”, and its checklist lists “‘Shadow AI’ use by staff, reducing control and visibility over how AI is used” among the risks small and not-for-profit boards face.⁶²

A tool found here gets a row like any other, with the owner column blank until someone approves it on a business plan (a paid, company-owned subscription) or retires it. Do not punish the person who told you.

Pass three: the AI already switched on. Open the administrator settings of each system on the map and find the AI features: the writing assistant in the email suite, the meeting transcription, the invoice-coding suggestions in the accounting system, the drafting and summarising in the CRM. For each, record whether it is on, whether it reads only or can also write or send, and who switched it on. The ASD notes that whether a provider trains on what you submit “can depend on the configuration settings or the type of subscription”.⁵

The AICD guide says it from the board’s side. As AI features “are increasingly embedded within broader software and digital platforms”, the AI functionality may need to be “highlighted or accounted for in the organisational digital and data inventory or register”.⁶ That sentence is the map’s AI-features column.

A worked example. The composite below is a professional services practice of under 20 staff: a principal, a practice manager, and fee-earning and administrative staff. Four of its rows after the three passes:

System	Source of truth for	Owner	Administrator	Personal information	Where stored	AI features	How other tools reach it	Renewal
Accounting system, business plan	Invoices, bills, payroll, the chart of accounts	The principal	The practice manager	Yes: staff pay and bank details; client billing details	Australia, per the vendor’s terms, checked [date]	Bank-feed coding suggestions, on by default: reads. Receipt capture: reads	Export; API; the practice system pushes invoices in	Yearly in July, 30 days’ notice, [cost]
Practice management system	Client records, matters, time recorded	The practice manager	The practice manager	Yes: client names, contact details, identity documents	United States, per the vendor’s terms, checked [date]	Assistant that drafts client emails and summarises files, switched on by the vendor in [month]: reads; can send email if enabled (off)	API; connector to the accounting system and to email	Yearly in March, 60 days’ notice, [cost]
Email, calendar and documents suite	Correspondence, documents, the calendar	The principal	The outside IT provider	Yes: copies of everything above	Australia for mail and files; AI requests processed overseas, per the vendor’s terms, checked [date]	Writing assistant and meeting transcription, on by default: read. The note-taker records meetings	Everything connects to it; staff phones	Yearly in [month], [notice], [cost]
Free meeting note-taker, found in pass two	Nothing: it is not the official home of any record	Nobody yet	Nobody: two personal accounts	Yes: recordings and transcripts with client names	Unknown	Transcribes and summarises; stores the recordings	Nothing reaches it; nothing can get the data out	No subscription. Decision due [date]: approve on a business plan, or retire

The fourth row is the one the map exists for. Before the map, the recordings were in two personal accounts and nobody’s job. After it, there is a decision with a date, and until then the practice’s staff AI policy, written from A free staff AI policy template, says the tool is not approved.

Keep it: one owner, and a review before every notice period

One named owner. In a business of a few dozen people that is the operations or office manager, who in A free staff AI policy template already owns the AI policy and the list of approved tools. In an organisation of 200 or more it is whoever keeps the solution architecture, usually in IT, and the map is the page the executive reads. The National AI Centre’s template asks for the same thing under another name: “Your organisation should appoint a register administrator to own and manage this document”.⁴ The AICD guide adds that “There should be a process to regularly update the inventory or register.”⁶

Three triggers to review it. The template’s spreadsheet says to review the register when “your organisation changes the way you use an AI system” and when “systems change”.⁴ We add the one trigger that comes with a date:

1. Every subscription’s yearly review, before its notice period opens. The owner of each row answers the five questions in Choosing AI software and connecting it to your systems, and the answers go in the row, dated.
2. Every new tool. Nothing is bought, renewed above a threshold you set, or given company data without a row, and the row is filled in before the decision to take the tool on. The template says the same: anyone procuring a new AI system “should log it on the AI register once the AI system or tool is approved”.⁴
3. Every automation switched on. When an AI tool starts writing into a core system, its one page of rules, thresholds and spot-checks, signed by the process owner, is filed beside the row. The six rules and the sign-off are in Choosing AI software and connecting it to your systems.

How the series points at it. A free staff AI policy template records approved tools and shared-task automations in the map, with the person who owns each. The choosing-software guide’s sign-off rules live beside each automation’s row. How to build an AI roadmap runs its shared-task layer from the map. The first shared process is chosen from systems the map says are connectable, and a candidate process waits if its data sits in a system with no export. One document is kept, and four read from it.

Four Australian sources ask for something like it, and the register is the closest

Source	Date	What it asks for	What the map adds
National AI Centre, Guidance for AI Adoption: Foundations, and the AI register template34	Foundations PDF published 5 May 2026; template 22 April 2026	A register of “all your AI systems”, including AI “embedded in other systems”; twelve fields per AI use; a register administrator	Where the data is stored, read or write, connections, renewal; the systems that have no AI in them
ASIC, Report 7981	29 October 2024	“Do you know where AI is used in your organisation?” and “Do you have an AI inventory, and are you confident that it is being adequately maintained?”	A dated review per row, so “adequately maintained” has a test
AICD and Human Technology Institute, Director’s Guide, v2, and the small and medium enterprise (SME) and not-for-profit (NFP) checklist62	June 2026	“Understand where AI is already being used in the organisation, including AI features in third-party products and services”; a “data inventory, stocktake or mapping exercise” before an AI strategy; an inventory or register that “should track all approved AI systems or tools”	The same list, kept by the person who keeps the subscriptions
ASD, Artificial intelligence for small business, and the cloud shared responsibility guidance57	14 January 2026; 20 October 2025	Nine checklist questions of each AI tool, among them where its data is stored and who owns it; review the shared responsibility model (which security tasks the vendor does and which are yours) “for each cloud service you use”	A place to write the answers, one row per tool, so the nine questions are asked once per system rather than remembered

The ASD never asks for the list. Its AI checklist for small business is nine statements, each about one tool: “I have verified what data the AI tool collects and where it is stored”, “I know who owns the data: my business or the AI vendor”.⁵ Its cloud guidance says to review the shared responsibility documentation “for each cloud service you use or plan to use”, because “If you and the CSP are unclear on who is responsible for what, it can leave gaps in security.”⁷ A CSP, in the ASD’s term, is a cloud service provider, the vendor that runs the software and the servers for you. Both assume you know which services those are. The map is the list the ASD’s questions are asked of.

The register describes a use, and the map describes a system. The National AI Centre says a register “should capture all AI systems, including AI features embedded in common software packages”, so staff can “check the register before deploying, buying or using any AI system”, and that it “will help to reduce fear, misinformation and ‘shadow’ AI use”.⁴

The map does that job in a business of a few dozen people or a not-for-profit of the same size, where each system carries one or two AI features and the AI-features column can name them. Where one system carries many AI uses, each with its own purpose, screening and affected people, those are register rows, and the map’s row for that system is the index that points to them.

ASIC’s review of 23 licensees and 624 AI use cases shows what keeping neither well looks like. Some licensees “required extra time to collate use cases” when asked, ASIC suspected “a lack of an AI inventory, or the recording of models in several dispersed model registers” was the cause, and one licensee found “models missing from the register” when it answered.¹ ASIC’s heading for the finding was “Licensees and their boards may not have clear visibility of their AI use”.¹ Those were financial services firms, and nothing in the finding depends on the sector.

What the map shows the board and the executive

The map is read at four columns.

Where personal information sits, and so where the Privacy Act bites. The Act’s Australian Privacy Principles (APPs) apply to every system marked “yes” in the personal-information column, wherever the vendor keeps the data, if the Act covers you (the test is turnover, and the businesses covered regardless of turnover are listed in AI regulation in Australia in 2026).

The guidelines of the Office of the Australian Information Commissioner (OAIC) say an organisation “holds” personal information if it “has possession or control of a record that contains the personal information”, and that “holds” extends “beyond physical possession of a record to include a record that an APP entity has the right or power to deal with”, for example where “the entity has outsourced the storage of personal information to a third party but it retains the right to deal with it”.⁸

Every cloud system on the map is that example, because the vendor stores the information and you keep the right to deal with it. That is why the where-stored column says “per the vendor’s terms”. The ASD puts the customer’s side in one line: “You can’t outsource that risk to a CSP.”⁷

Which tools can watch staff. An administrator console or audit log that records what staff type into an AI tool can be computer surveillance under NSW law and data surveillance under the ACT’s, and both need written notice before it starts. The AI-features and administrator columns show which rows that applies to. The rules, the 14 days’ notice and the state-by-state table are in How to write an AI policy. The map only tells you which systems to read them against.

Which systems AI can write into. The AI-features column separates a feature that reads from one that writes or sends, and the how-other-tools-reach-it column shows which systems a connector or automation can reach. Read together, they show a board, without a technical briefing, every place an AI tool could post an invoice, change a customer record or send an email in the organisation’s name.

The ASD’s cloud guidance names the risk the customer always carries, of “your data being stolen, changed without your approval, or you losing access to it”.⁷ Each write-capable row has a signed page beside it or a decision date for one. The AICD guide asks that board reporting “provide a clear view of where AI is being used across the organisation”, and these rows are the part of that view that carries money.⁶

Which report is the official source of each headline number. A headline number is any figure the organisation steers by and reports upward: a margin, a headcount, a debtor total, an occupancy rate, a waiting time, a case load, a donor count. Once AI makes analysis cheap, anyone can produce one in a minute, from whichever system they have open, and two people will bring two different figures to the same meeting. The source-of-truth column decides it. Each kind of record has one home, and the report from that home is the official number.

Write the headline numbers the board sees against their rows, with the report name and the owner. A figure from anywhere else is a draft until it is reconciled to the row. The AICD guide’s governance red flags include that “There is no data stocktake or inventory and limited understanding of internal data quality”.⁶ The map is the stocktake, and it is how a board finds out what has already happened in the software it pays for.

The Systems Map Template

Copy everything in this section into a document and replace each item in square brackets. Write the owner’s name and role once, at the top. Add a row for each system as the three passes find it, and tick off each pass below the table with the date it was done.

What changed

This is version 1.0, first published 6 November 2026 and last updated 6 November 2026. The version moves at a major change, such as a law or a piece of guidance the guide rests on changing; a smaller fix moves the updated date only. Earlier versions are kept from the first major change on.

No changes since first publication.

What we could not confirm

• An Australian source that uses “systems map” or “solution architecture” in this sense. None of the documents we read does. The nine columns, the three passes and the yearly review before the notice period are ours. The table above shows where each source asks for part of it.
• A review frequency for a register or map. No source we read sets one. The National AI Centre’s spreadsheet says to review when “your organisation changes the way you use an AI system” and when “systems change”, and the AICD guide asks for “a process to regularly update” the register without a period.⁴⁶ The yearly review before each notice period is our proposal.
• The Word version of the register template. In the copy we extracted on 6 November 2026, the description column of the field table appears shifted down one row from the eighth field, so that “Registered date” is described as a “Risk level assessment result”. The spreadsheet’s descriptions line up with its fields, and the twelve fields above are quoted from it. We may have misread a merged cell.
• Whether ASIC’s question applies beyond licensees. Report 798 reviewed 23 financial services and credit licensees and addresses its questions to them.¹ We borrow the question. Nothing in the report extends it.
• The OAIC chapter’s currency. Chapter B is version 1.4 of 21 December 2022, and the page says the OAIC “is currently updating this chapter” for terms about collection. We quote the version online on 6 November 2026 and found no newer text.⁸
• The worked example’s where-stored entries. They show what a filled column looks like, not findings about any vendor. Your vendor’s terms supply the country, and the date you read them goes in the cell.

We will re-check this piece every quarter, and on 10 December 2026 when APP 1.7 to 1.9 commence. When the register template, the AICD guide or the ASD’s guidance changes, the date at the top of this page changes with it.

Notes

1. ASIC, Report 798: Beware the gap: Governance arrangements in the face of AI innovation, 29 October 2024: p 4 (23 licensees, 624 use cases), Finding 6, p 27 (“Licensees and their boards may not have clear visibility of their AI use”), and “AI governance: Questions for licensees”, p 35, question 1, “Taking stock”. Read 6 November 2026. ↩ ↩ ↩ ↩ ↩

2. Australian Institute of Company Directors and Human Technology Institute, A Director’s Guide to AI Governance: SME NFP Board Summary and Checklist, version 2, June 2026: “Opportunity and risk” (p 1) and the board considerations table, “AI Strategy” and “AI Governance Practices” (p 2). Read 6 November 2026. ↩ ↩ ↩

3. National AI Centre, Guidance for AI adoption: foundations, ai.gov.au, practice 4 (“Share essential information”), “Getting started”, first action; the PDF on the page is marked “Published 05 May 2026”. Read 6 November 2026. ↩ ↩

4. National AI Centre, AI systems register, ai.gov.au, with the AI register template as a Word document and a spreadsheet, both marked “Published 22 April 2026”: the page’s “Why use an AI systems register?”; the template’s “Recommended fields”, “Appoint a register administrator”, “Register all new AI systems” and, in the spreadsheet, “Review and update the register”. The twelve field names and descriptions are quoted from the spreadsheet. Read 6 November 2026. ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩

5. Australian Signals Directorate, Australian Cyber Security Centre, with NCSC-NZ and COSBOA, Artificial intelligence for small business, published 14 January 2026: “Data leaks and privacy breaches”, “Supply chain vulnerabilities” and “AI cyber security checklist for small businesses”. Read 6 November 2026. ↩ ↩ ↩ ↩

6. Australian Institute of Company Directors and Human Technology Institute, A Director’s Guide to AI Governance, version 2, June 2026: p 6 (embedded AI and shadow use), p 29 (“Baseline”: the data inventory, stocktake or mapping exercise), p 33 (what board reporting should provide), p 38 (“AI inventory or register”), p 52 (Appendix C, the SME and NFP director checklist), p 53 (glossary, “AI inventory (or AI register)”) and the governance red flags under “Stakeholder impact”. Page numbers are the PDF’s printed numbers. Read 6 November 2026. ↩ ↩ ↩ ↩ ↩ ↩ ↩

7. Australian Signals Directorate, Australian Cyber Security Centre, Cloud shared responsibility model: Guidance for individuals and small and medium businesses, published 20 October 2025: “Key actions”, “Understanding responsibilities”, “Foundational responsibilities”, “Your responsibilities” and “Coverage of responsibilities”. Read 6 November 2026. ↩ ↩ ↩ ↩

8. OAIC, Australian Privacy Principles guidelines, Chapter B: Key concepts, version 1.4, publication date 21 December 2022, paragraphs B.82, B.84 and B.147; the definition of “holds” is in s 6(1) of the Privacy Act 1988 (Cth). Read 6 November 2026. ↩ ↩