How do you find every automated decision in your business before 10 December 2026?
Start from your process inventory rather than your IT asset list, and at every step of every process that touches an individual ask three questions: is a decision made here, does a computer program make it or do something substantial towards it, and does it use personal information. Record each answer in a register, then write the privacy policy statements from the register rather than the other way round.
Written for privacy officers, operations leaders and anyone who has been asked to list the organisation's automated decisions. Reviewed by Mike Bentley, founder, .
What exactly are you looking for?
From 10 December 2026, Australian Privacy Principle 1.7 requires an APP entity's privacy policy to describe certain uses of computer programs. The Office of the Australian Information Commissioner's fact sheet, published 30 September 2026, puts the test in three parts: the entity "has arranged for a computer program to make, or do a thing that is substantially and directly related to making a decision", "the decision could reasonably be expected to significantly affect the rights or interests of an individual", and "personal information about the individual is used in the operation of the computer program".
All three must be true. In plain words, an automated decision for this obligation is a decision a computer program makes, or substantially and directly helps to make, using personal information about the individual, where the decision could reasonably be expected to significantly affect that person's rights or interests. When all three are true, APP 1.8 requires three things in the privacy policy: the kinds of personal information used in the operation of such programs, the kinds of decisions made solely by them, and the kinds of decisions for which a program does a thing substantially and directly related to the decision. The inventory therefore has to tell those last two kinds apart.
Two points in the Act itself matter. The Privacy and Other Legislation Amendment Act 2024, assented to on 10 December 2024, applies the new paragraphs to decisions made after commencement whether the arrangement for the program was made before or after that date, so nothing already running is exempt. The same Act adds APP 1.7 to section 13K of the Privacy Act, the civil penalty provision for which infringement notices or compliance notices can be issued. This article is a method, not legal advice.
Why start from processes rather than systems?
Decisions live inside processes, not inside an asset register. An approval, an eligibility check, a price, a triage order, a schedule, a fraud flag: each is a step in a process an individual passes through, and the program that makes or shapes it is often a feature of a system nobody would list as a decision-making tool.
An IT asset list fails in both directions. It misses the spreadsheet whose formula ranks clients, the configured feature inside a vendor platform, and the chatbot that decides whose query reaches a person. It also says nothing about whether a decision is made, which is the only question that matters here. The OAIC's first worked example of a program in scope is a spreadsheet in a health and aged care provider whose formula scores and ranks people for contact, according to its fact sheet.
So begin with the process inventory, filtered to the processes that touch individuals: customers, applicants, employees, claimants, patients, students, members. If no inventory exists, that shorter list is the one to build first, because it holds nearly everything in scope. Expect one platform to hide several decisions: the OAIC's example of a single eligibility report describes outputs from different programs running in the background, some in rapid succession, each a decision in its own right.
What three questions do you ask at each step?
Walk each process in the order the individual experiences it, and at every step ask the three questions below. Write down the answer even when it is no, because a documented no is what shows the search was complete.
Ask the second question, whether a computer program makes the decision or does something substantial towards it, of every kind of program, not only anything labelled AI. The fact sheet lists pre-programmed rule-based processes, artificial intelligence and machine learning, software, apps or word-processing tools, and generative AI including chatbots. In practice that means rules engines, spreadsheets with formulas, and features inside vendor software: credit scoring, fraud flags, CV screening, dynamic pricing, chatbots that triage. The fact sheet also counts configuration or customisation of off-the-shelf or packaged software, and reliance on advisory outputs, as ways of having "arranged for" a program, so a feature you switched on counts as much as one you built.
- Is a decision made here? Under APP 1.9, making a decision includes refusing or failing to make one, and a decision counts whether it affects the individual adversely or beneficially. Narrowing, ranking, scoring, prioritising, flagging and routing all determine an outcome for someone.
- Does a computer program make it, or do something substantial towards it? The fact sheet says "substantially" means the program is a key factor in facilitating the human's decision making, and "directly" means it has a direct connection with making the decision. The fact sheet says a program may be substantially and directly related to a decision whether its output is advisory or determinative, so an advisory output can be in scope as much as a determinative one.
- Does the program use personal information about the individual? If the inputs include anything about the person the decision concerns, the answer is yes.
What do you record for each decision you find?
Each decision found becomes a row in the register, with enough detail to write the three APP 1.8 statements and to defend the judgement later. The OAIC's guidelines suggest considering which stage of the decision the program supports, what information it uses, what parameters it was given, and how much its output shapes the human decision-maker's judgement.
Keep the rows where you found no decision as well, with the date and who looked, so the register shows where you searched and not only what you found.
- The decision in plain words: what is granted, refused, ranked, priced or flagged, and for whom.
- The program: the system, the vendor where there is one, and the specific feature or rule set, since one platform may hold several.
- The inputs: the kinds of personal information used, with sensitive kinds such as health or biometric information called out, because the OAIC expects the policy to make those clear.
- Whether the program decides solely, or does something substantial towards a human decision: the split between APP 1.8(b) and 1.8(c).
- What the human actually does, not what the procedure says. The OAIC's remuneration example treats a recommendation managers rely on as substantial even with director sign-off, and names what might change that: interrogating outputs, reviewing the underlying factors, setting narrower parameters, using additional evidence and documenting why the decision diverged.
- Who is accountable for the decision, as one named person or position, and who can change the rule or the configuration.
- Which processes the decision sits in, since one program often appears in several and the policy describes kinds of decisions, not systems.
How do you find the ones nobody documented?
The decisions most likely to be missed are the ones nobody describes as decisions. To the people running them, the system simply works that way.
When the walk is done, cross-check the result against the OAIC's non-exhaustive list of decisions it would generally consider in scope: recruitment software that sorts candidates, differential pricing for significant goods, prioritising health or disability services, loan and credit approval, insurance eligibility, AI reports that rank employee performance. If your organisation does any of these and the register shows nothing, the register is wrong. Four sources surface the decisions the walk missed.
- The people who do the job. Ask what the screen tells them before they act, what they would do if it said nothing, and how often they go against it. The honest answer to the last question says whether the program is substantial.
- A recording of the job done once. Watching a case travel from intake to outcome shows every point where a system filled in, ranked, scored or routed something, including steps that happen between clicks and never reach a procedure document.
- Meeting transcripts and support tickets. A transcript in which someone says the system will not let them approve something names a rule. A complaint asking why an application was declined or a price was higher names a decision, and often the individual's own view of its significance.
- Vendor documentation and contracts. The fact sheet says providers of third-party software should give clear, high-level information about how their software can be used to make decisions, and the guidelines say contracts should be clear about which entity keeps ultimate decision-making capability. Ask every vendor for both, and read the configuration screens rather than the brochure.
How do you turn the register into privacy policy statements?
Once the register exists, the policy text is a grouping exercise. Group the rows into kinds of personal information and kinds of decisions, and write the three statements APP 1.8 requires: the kinds of personal information the programs use, the kinds of decisions made solely by programs, and the kinds of decisions where a program does something substantial towards a human decision.
The fact sheet says entities may group categories of decisions and types of personal information, but the level of detail "must be meaningful to a reasonable person", and where sensitive information such as health information or biometric templates is used, that should be clear. Its worked examples use a plain structure: a heading for decisions made by computer programs, a list of the kinds of personal information, then a list of the kinds of decisions made with it.
Commercial-in-confidence information about the programs is excluded, according to the fact sheet, but only the narrow category whose release could harm commercial interests. How a fraud model weights its data points can stay private; that personal information drives fraud decisions cannot, and information that would merely expose the entity to criticism is not commercial-in-confidence. The fact sheet also says an entity in doubt about whether a use is in scope "should take a cautious approach and include information in their APP privacy policy". Government agencies have a supplementary fact sheet, published the same day, on how the obligation sits with the Information Publication Scheme.
How do you keep the register current?
The obligation applies to every decision made after 10 December 2026, so the register has to track change rather than describe a moment. Re-ask the three questions whenever something moves: a new vendor or module, a feature switched on, a rule or threshold edited, a process redesigned, or a team that starts relying on an output it used to ignore. Procurement and change approval are the two gates where the questions can be made mandatory.
The OAIC's guidelines say entities should actively identify, assess and keep oversight over third-party computer programs, during and after procurement, and recommend a program of proactive review and audit of the privacy policy's adequacy and currency. A register with a review date on each row, and a named owner who answers when it falls due, satisfies both.
A spreadsheet and a disciplined owner are enough for a small organisation. Where processes are many, span several systems and are not written down, the walk itself is the cost, and a connected model of processes, systems and data turns it into a query rather than a survey. Omni by Valstra is one option: it builds that model from the documents, meeting transcripts and wiki pages an organisation already holds and from a consented recording of a job done once, and its AI system register records each AI use with its purpose, its risk classification, who oversees it and what was approved, so the three questions can be asked of the model and the answers kept beside the processes they belong to.
Sources
- OAIC, New resources on transparency for use of AI and automated decision-making (30 September 2026)
- OAIC, APP 1.7 to 1.9 Transparency Obligation Fact Sheet (PDF, September 2026)
- OAIC, APP 1.7 to 1.9 Transparency Obligation Flowchart (PDF)
- OAIC, APP 1.7 to 1.9 Transparency Obligation Supplementary Fact Sheet for Government Agencies (PDF)
- OAIC, APP Guidelines, Chapter 1: APP 1 Open and transparent management of personal information (version 2.0, updated 30 September 2026)
- Privacy and Other Legislation Amendment Act 2024 (Cth), text as made
Common questions
- Does this apply if we only use off-the-shelf software?
- Yes, it can. The OAIC's fact sheet counts procuring a third-party program, and configuring or customising off-the-shelf or packaged software, as ways of having "arranged for" a program, and its flowchart notes that entities that procure, configure, authorise, integrate or rely on a third-party program may still be in scope. The obligation typically stays with the entity using the personal information, so ask the vendor what the feature does with it and record the answer.
- Does a human in the loop take a decision out of scope?
- Not on its own. Record what the reviewer actually does with the output: how often they go against it, whether they see the factors behind it, and whether they can change its parameters. The OAIC's remuneration example treats a recommendation managers rely on as substantial even with director sign-off, so a signature at the end is not the test.
- What if we are not sure whether an effect is significant?
- The OAIC says significant means more than trivial, with the potential to considerably influence the individual's circumstances or outcomes, that vulnerability can make a decision weigh more heavily, and that small price differences can compound into a significant effect. Its published position for an entity in doubt is to take a cautious approach and include the information in the privacy policy.
- What happens after 10 December 2026 if the list is incomplete?
- The Act adds APP 1.7 to the civil penalty provision for which infringement notices or compliance notices can be issued, but the OAIC's resources of 30 September 2026 set out no enforcement approach for this obligation, and as at 9 November 2026 we found no OAIC publication with enforcement detail specific to it. Its published position is the cautious one: if in doubt, include it, and keep the register that shows how you looked.
See it against your own processes
A demo runs on your material, not a canned dataset. Bring a process you find hard to explain.