AI decision systems · Melbourne
AI decision support for small business: software does the homework, your people make the call
Most business decisions aren’t slow because they’re hard. They’re slow because of the homework: the PDF, the spreadsheet called FINAL_v3_USE_THIS_ONE, the job system and the phone call to check. Lumeio builds AI decision systems that gather, check and rank that information, then hand each decision to a named person with the reasons attached.
Fixed price. Written report in 5 business days. Credited if you go ahead.
Software already helps decide things about staff, clients or applicants? Check your ADM exposure
Book your Pain Point Audit
No obligation. We reply within 1 business day. Your details stay private. See the privacy policy.
What is AI decision support?
AI decision support is software that collects the information a decision needs, checks it against your rules, ranks or flags the cases that need attention and shows why. A person then decides faster and misses less. The software prepares the decision. A named person makes it, and the system records who decided and when. Lumeio builds these systems for Australian businesses with roughly 10 to 200 staff.
The symptoms
The decision takes a minute. The homework takes the morning.
Picture Monday, 8:40am. Your operations manager has to decide which three of 22 open jobs get a crew today. The decision itself takes about a minute. Getting to it means the job system, a supplier’s email about a late delivery, the variations spreadsheet, and a text to the site lead who replies “on it” (which is not information).
Now multiply that across the decisions your team makes every week. Do any of these sound familiar?
- Which supplier invoices get paid this week, and which three have a price that doesn’t match the purchase order.
- Whether a subcontractor can go on site Monday, given one certificate quietly expired last Tuesday.
- Which client files are missing an ID document or a sign-off before the lodgement deadline.
- Which incoming request is genuinely urgent, and which one just has a lot of capital letters in the subject line.
If you said “oh no, we do exactly that”, good news: you don’t have a decision problem. You have a homework problem, and homework is what software is good at.
Here’s where a lot of AI tools get pitched badly. They’re sold like a Magic 8-Ball with a monthly subscription. Ask the question, give it a shake, read the answer, hope. That’s fine for choosing lunch. It is not how you want to decide who gets a shift, which claim gets paid or whether a builder goes on site.
We build the other kind: software that does the homework and shows its working, and a person who signs their name to the decision.
The decision ladder
How much should software decide? Use the decision ladder
Not every decision deserves the same amount of automation. Before we build anything, we put each decision on one of four rungs. The higher the rung, the more the software does on its own.
- Rung 1
Inform
Decisions about people- Software:
- gathers the facts from email, documents and systems and shows them on one screen.
- Who decides:
- a person, every time.
- Example:
- a subcontractor’s licence, insurance and induction status on one page, with the source document one click away.
- Rung 2
Recommend
Decisions about people- Software:
- ranks, scores or flags cases and shows the reason for each one.
- Who decides:
- a named person, who can disagree with the ranking.
- Example:
- overdue invoices sorted by amount and age, with the ones that don’t match the purchase order flagged and the mismatch shown.
- Rung 3
Act unless stopped
Low stakes only- Software:
- acts on a plain rule, and a person reviews the log and can reverse it.
- Who decides:
- the rule, with a person watching.
- Example:
- a reminder email goes out 30 days before a certificate expires.
- Rung 4
Act alone
Trivial and reversible only- Software:
- decides and acts with nobody checking.
- Who decides:
- the software. This is where the Magic 8-Ball lives.
- Example:
- filing an incoming invoice email into the right folder.
Lumeio’s rule: decisions that could significantly affect a person, such as their work, money, care or access to a service, stay on rungs 1 and 2. A named person makes those calls. Rungs 3 and 4 are for low-stakes, reversible tasks that follow plain rules, and they rarely need AI at all.
The split
What an AI decision system does, and what stays human
We don’t start with AI. Most of a decision system is ordinary rules, because a rule gives the same answer every time and you can check it. AI earns its place where the input is messy.
| Part of the decision | Who does it | Why |
|---|---|---|
| Pull the facts from email, PDFs and your systems into one record | Rules | Moving data between systems is predictable. No AI needed. |
| Read documents and free text: classify, extract, summarise | AI | This is where AI is genuinely better than a rule. Every value stays linked to the page it came from. |
| Check values against your thresholds, dates and amounts | Rules | “Expired means blocked” is an IF statement, not a job for a language model. |
| Rank and flag cases, with the reason shown | Rules and AI | A flag without a reason is just a nag, and nags get ignored by week three. |
| Make the decision | A named person | Accountability can’t be handed to software. |
| Handle anything the system can’t validate | A named person | Unclear cases go to a person with the reason. They are never guessed. |
| Record who decided, when and why | Rules | So you can answer the question when a client, auditor or insurer asks it later. |
“A person decides” only counts if the oversight is real. Our working test is a named reviewer, a timestamp and a genuine right to say no. Here’s what human in the loop means when it’s done properly.
Worked example (illustrative)
Example: can this subcontractor go on site Monday?
A Melbourne builder needs every subcontractor’s insurance and licence checked before they start. It’s a yes or no decision with real consequences, which makes it a good rung 2 decision.
- Certificate arrives as a PDF attachment
- Someone opens it and reads the expiry date and cover amount
- Checks the compliance spreadsheet
- Finds the licence on file is out of date
- Emails the subcontractor and waits
- Updates the spreadsheet, if they remember
- Friday, 4pm: “Are we good for Monday?” Nobody is sure
- Certificate arrives and is picked up automatically
- AI reads the insurer, policy dates and cover amount
- Rules check them against your minimums and the job dates
- Each subcontractor shows as ready, needs attention or blocked, with the reason
- Site manager decides on every “needs attention” case
- The decision, the name and the time are recorded
Friday, 4pm: one screen, one answer, and the reasons behind it.
Say the builder onboards 25 subcontractors a month and each check takes 20 minutes of opening, reading and cross-checking. That’s about 8 hours a month of homework before anyone makes a single decision. [Quietly closes 14 browser tabs.]
Illustrative arithmetic, not a client result. Your Pain Point Audit measures your real volumes and times. For the method, see how to calculate the ROI of automation, and for more on site paperwork, read construction paperwork automation for site managers.
Where we use it
AI decision support in the five industries we work in
Lumeio works in regulated, paperwork-heavy industries. In each one, the software prepares the decision and the professional keeps it.
Building surveying
Is this permit application complete enough to assess?
Software checks the documents against your checklist and lists what’s missing. The surveyor makes every professional call.
Waste and EPA compliance
Which dockets don’t reconcile?
Software matches dockets to weighbridge records and flags the mismatches. A person decides what gets corrected and what gets escalated.
Aged care
Which records need a closer look today?
Software flags care and incident records with missing fields so nothing sits unread. A care or clinical lead decides what happens next, including any report.
Construction
Who is cleared to start on site?
Software reads certificates, SWMS and inductions and shows each subcontractor’s status with the reason. The site manager clears them.
Accounting
Which client files aren’t ready to lodge?
Software checks for missing identification, source documents and sign-offs. The accountant decides what’s ready.
Not in one of these? The same approach applies to any business that makes repeat decisions from documents and spreadsheets. See all the industries we work with.
Is it worth it?
Is AI decision support for small business worth it?
It’s worth it when a decision repeats often and gathering the information takes longer than deciding. It usually isn’t worth it for decisions that come up a few times a year or rest entirely on judgement.
Good candidates
- The decision comes up weekly or more.
- Gathering the facts takes longer than the decision.
- Most of the rules are written down, or could be.
- A late or wrong call costs money, a client or an audit finding.
Leave these alone
- Decisions made a few times a year.
- Calls that rest on relationships or gut feel, with no data behind them.
- Anything where nobody can say what a good decision looks like yet.
- A process that’s broken. Software makes a broken process faster, not better.
Privacy Act · APP 1.7
[Switches to serious face.]
AI decisions about people and the 10 December 2026 privacy change
From 10 December 2026, organisations covered by the Privacy Act must describe in their privacy policy when they use personal information in automated decisions that could significantly affect people. That includes software that does something substantially and directly related to a decision, such as scoring, ranking, filtering or flagging people. A person making the final call does not by itself take the system out of scope.
Businesses with annual turnover of $3M or less are generally exempt. Some, such as health service providers, are covered regardless. The OAIC explains which small businesses the Privacy Act covers.
In ladder terms, a rung 2 system that ranks or flags people can be in scope even though a person decides. That is a good reason to know exactly which of your systems do it.
So every AI decision system Lumeio designs comes with an automated decision-making (ADM) review. We list each system that touches decisions about people, rate it for risk, record where the data lives and where a person steps in, and give your privacy lawyer a plain-English brief. The review is a A$750 + GST add-on to the Pain Point Audit, included free in audits booked before 10 December 2026.
Six questions, about five minutes. Or read how the automated decision-making review works.
Lumeio maps the systems. A privacy lawyer writes the policy. This section is general information, not legal advice.
Why the person stays in charge
Most businesses don’t trust AI to decide. Fair enough.
About 65%
of Australian SMEs that aren’t adopting AI cited either distrust in AI decision-making or a strong preference to keep human control of their processes, according to the National AI Centre’s AI adoption insights for December 2025 to February 2026. Trust was the biggest single barrier, ahead of cost.
We think those businesses are right to ask. The National AI Centre’s Guidance for AI adoption sets out six essential practices:
- Decide who is accountable
- Understand impacts and plan accordingly
- Measure and manage risks
- Share essential information
- Test and monitor
- Maintain human control
A Lumeio decision system builds the first and last of those into the workflow itself. Every decision has a named owner, the software shows its reasons, and the person can always overrule it. The same guidance points out that an internal chatbot carries very different risks from AI used in hiring or customer decisions, which is exactly why we sort decisions on the ladder before building anything.
How it works
How Lumeio builds an AI decision system
- A$1,950 + GST
Pain Point Audit
A fixed-price, half-day look at how work really moves through your business. Within 5 business days you get a written report ranking your top five time drains, what AI should and shouldn’t handle, and where a person must stay in charge. Read what the audit covers.
- Fixed price per phase
Build
We build the decision system from what the audit found, quoted before anything starts. The audit fee is credited in full against a first build of A$5,000 + GST or more signed within 60 days of your report.
- Rung 1 or 2 first
Run it next to your team
The system starts by informing or recommending. Your team checks its flags against their own judgement, so you can see where it’s right and where the rules need work.
- Optional
ADM review
If software touches decisions about people, the automated decision-making review gives your lawyer a brief before 10 December 2026. A$750 + GST, or free in audits booked before 10 December 2026.
One fixed price, agreed before we start. The report is yours whether or not you go ahead.
Common mistakes
Four ways AI decision support goes wrong
Automating the decision instead of the homework
The expensive part is usually gathering and checking the information. Hand that to software and keep the decision, and you get most of the time back with far less risk.
A human in the loop who approves 400 items before morning tea
If the reviewer can’t realistically say no, the oversight is a turnstile with a login. Fewer, better flags beat a long queue of rubber stamps.
Using AI where a rule would do
If the rule is “certificate expired means blocked”, you don’t need AI. You need an IF statement. Rules are cheaper, faster and easier to explain to an auditor.
Answers with no reasons
When a system isn’t sure, it shouldn’t say “Reply hazy, try again” like a Magic 8-Ball. It should send the case to a person and say why. A flag without a reason teaches your team to ignore flags.
About the author
FAQ
Questions about AI decision systems
What is an AI decision system?
An AI decision system is software that prepares a business decision for a person. It pulls the information together from email, documents and your existing systems, checks it against your rules, ranks or flags the cases that need attention and shows the reason for each flag. A named person makes the decision, and the system records who decided and when.
Is AI decision support worth it for a small business?
Usually, when your team makes the same kind of decision many times a week and spends longer gathering the information than deciding. Approving invoices, checking subcontractor compliance and reviewing incoming requests are typical. If a decision comes up twice a year, it probably isn’t worth building for. The Pain Point Audit measures which of your decisions are worth it before anything is built.
Will the AI make decisions without a person?
Not decisions that could significantly affect someone, such as their work, money, care or access to a service. Those stay with a named person who can disagree with the software. Software acts on its own only for low-stakes, reversible tasks that follow plain rules, such as sending a reminder before a certificate expires.
What is the difference between AI decision support and automation?
Automation does a task: it moves data, sends a reminder or builds a report. Decision support prepares a choice: it gathers the facts, applies your rules, ranks the options and explains why, then leaves the choice to a person. Most Lumeio decision systems use ordinary automation for the rules and AI only for reading messy documents and free text.
Does the Privacy Act’s automated decision-making rule apply to AI decision support?
It can. From 10 December 2026, organisations covered by the Privacy Act must describe in their privacy policy when they use personal information in automated decisions that could significantly affect people. That includes software that does something substantially and directly related to a decision, such as scoring, ranking, filtering or flagging people, even when a person makes the final call. Businesses with annual turnover of $3M or less are generally exempt. Some, such as health service providers, are covered regardless. Lumeio maps the systems. Your privacy lawyer advises on the policy.
What does an AI decision system cost?
It starts with the Pain Point Audit at a fixed A$1,950 + GST. Builds are quoted at a fixed price per phase from what the audit found, so you see the cost before anything is built. The audit fee is credited in full against a first build of A$5,000 + GST or more signed within 60 days of your report. The ADM review is a A$750 + GST add-on, included free in audits booked before 10 December 2026.
Do we have to replace the software we already use?
No. The starting assumption is that the decision system reads from and writes to the email, spreadsheets, accounting and job systems you already run. The audit checks what each system allows. If one can’t be connected, the report says so before you pay for a build.
Keep the decisions. Lose the homework.
No Magic 8-Ball. Just the information gathered, checked and ranked, and the call left with the person who should make it.
Fixed price A$1,950 + GST. Written report in 5 business days. Credited if you go ahead.
- All solutions
- Pain Point Audit
- Automated decision-making review
- Human in the loop, defined
- Automation ROI
- Industries
Sources and further reading
- National AI Centre: Guidance for AI adoption, implementation guidance
- National AI Centre: AI adoption insights, December 2025 to February 2026
- OAIC: new resources on transparency for AI and automated decision-making
- OAIC: privacy guidance for small business
Examples on this page are illustrative. The privacy section is general information, not legal advice. Lumeio maps systems and does not give legal advice or certify compliance.