This guide gives the order: the work, the data, the numbers, the owner, then a 90-day pilot. A ten-question check scores where you stand.
Exhibit 2
Almost a quarter of SMEs used AI in 2025, but just 11% had a policy for it
Use
23%
SMEs where the owner or team use AI in the business
MYOB survey, April to May 2025.
Policy
11%
SMEs with an AI policy, so the team knows when and how to use AI
Same sample.
Owners, managers and directors of SMEs, n=1,087, self-reported. The ABS 12% counts all businesses, with AI as one of several options, so the two shares do not compare.
The first task has real hours, a cheap error and data in one system
Task
Hours a week
Cost of one error
Data in one system?
Supplier bill entry
6
Low: caught at approval
Yes, in the ledger
Customer quote follow-up
4
Medium: a lost sale
Partly, CRM and email
Payroll changes
3
High: staff underpaid
No, three places
Month-end reconciliations
8
High: wrong reports
Yes, but never checked
Sample business and figures. Hours are time, not cash.
Illustrative figures, not client data.
A tool is an answer. Start with the question: which task costs the most hours, or the most when it goes wrong?
List ten recurring tasks. For each one, write the weekly hours, the cost of one error and whether its data already sits in one system. Time each task for two weeks. A guess is not a baseline.
The best first task has real hours, a cheap error and data that is ready. A task with costly errors and messy data comes later.
Fix the data before the model
Exhibit 4
Only a fifth of surveyed SMBs keep their data in one integrated system
Data
1 in 5
SMBs with a centralised or integrated system for collecting, storing and analysing data
Over 30% keep data across teams or systems.
People
10%
of workforces have advanced AI skill levels
Strategy
32%
SMBs have an AI strategy embedded in core processes and decisions
Deloitte Access Economics for Amazon, 2025: about 1,000 Australian SMBs (question base 1,011). A commissioned survey, so read the shares as a guide.
AI works on what it can read. If a customer sits in the CRM, the job system and a spreadsheet, each with a different address, the tool reads all three and guesses.
Deloitte Access Economics, in a survey for Amazon, found that only a fifth of surveyed SMBs have a centralised or integrated data system. Fix one record first: say where it lives and who owns it. Connecting business systems shows how.
Count hours and dollars apart
Exhibit 5
Bill entry falls from 6 to 2 hours a week, which frees 17 hours a month
Time
17 h
Hours freed a month, after a person’s check
4 hours a week × 52 ÷ 12.
Cost
A$150
Running and software cost a month
A sample. Use your own quotes.
Sample business, excl. GST. Hours are time. They turn into cash only if paid hours go.
Illustrative figures, not client data.
Exhibit 6
The same 17 hours net A$615 a month if paid hours go, and −A$150 if they do not
17 hours at a sample A$45 an hour is A$765, less A$150 a month of running cost. Set-up and your own time are extra.
Show the figures
A$
Paid hours cut
A$615
Hours re-used
−A$150
Illustrative figures, not client data.
Time saved is not cash saved. Hours become dollars only when a paid hour goes: less overtime, no contractor, a hire you no longer need. If the team uses the hours for better work, that is a gain, but it is not a saving.
In my own work, automating payroll, AP and the month-end removed more than 500 labour hours a year. That is time. Show it on one line, the cash on another, and the software and support costs beside both. The cost-to-own guide covers those costs.
The sample below takes one task: bill entry, down from 6 hours a week to 2, including a person’s check.
Why pilots stall
Exhibit 7
A pilot stalls when it has no owner, no baseline, no checker or no rule
01
No owner
Nobody answers for the tool, who may use it or what it produces.
02
No baseline
Without the hours before the pilot, day 90 has nothing to compare.
03
Checking costs more
If a person redoes every output, the hours do not fall. Test on a small sample first.
04
No rule on data
Without a written rule, each person decides what goes into a public tool.
Pilots rarely fail on the tool. Four gaps can stall them, and each is cheap to close before you start.
Deloitte found that 47% of surveyed SMBs have a person responsible for monitoring AI risks. That leaves more than half without one. The OAIC says that AI systems are known to produce inaccurate or false results, and the WA Small Business Development Corporation says to always check the outputs you receive from AI.
Who decides what
Exhibit 8
The owner picks the task, the team proves it works, the board sets the limits
Owner or CEO
Picks the first task from the list
Names one person who owns the pilot
Sets the budget and the day-90 test
Team
Times the task for two weeks
Checks every output before it is used
Flags what the tool gets wrong
Board or adviser
Sets what data may go into AI tools
Reads hours, dollars and costs at day 90
Decides: go on, change or stop
Three groups, three decisions. The owner or CEO picks the task and the budget. The team times the work and checks the output. The board, or the owner acting as one, sets the rule on data and reads the result at day 90.
Keep the decisions in that order. A pilot with a budget but no owner is often how both get wasted.
The rules that apply in Australia
Exhibit 9
Privacy duties follow personal data into AI tools, and a disclosure date is coming
Rule
What it says
Source
Personal information
OAIC best practice: no personal or sensitive information in public generative AI tools
OAIC, updated 17 Jan 2025
Accuracy
AI systems are known to produce inaccurate or false results. Check every output
OAIC; WA SBDC
Small business
A$3 million turnover or less is exempt, except sectors such as health providers
OAIC; Productivity Commission
10 December 2026
APP entities must say where a program makes or informs significant decisions about people
OAIC, 30 Sep 2026
The 10 December 2026 duty covers decisions that significantly affect individuals’ rights or interests (OAIC, 30 September 2026).
Start with the rules you already have. The OAIC says privacy obligations apply to any personal information put into an AI system, and to the output where it contains personal information.
Many small businesses are exempt from the Privacy Act, but not all. A small business is one with an annual turnover of A$3 million or less, and some sectors are covered whatever their turnover, such as health service providers. Check your own position and get advice where it is unclear.
The Australian Government published its Guidance for AI Adoption on 23 October 2025: six practices to plan, manage and use AI in ways that build trust and deliver value.
Every check uses the baseline, and day 90 ends in one of three decisions
When
What happens
Who
Evidence
Before day 0
Time the task for two weeks and note the errors
Team
Baseline hours and errors
Day 0
Name the owner, the budget and the data rule. Pick the tool
Owner
One page, signed off
Day 30
Run on real work with a person checking every output
Owner and team
Hours and errors against baseline
Day 60
Fix what the checks found. Add the running costs
Owner
Net cash and hours, apart
Day 90
Decide: go on, change the scope or stop
Owner or board
One page: hours, cash, costs, risks
One task, one owner, 90 days. The plan has a baseline before day 0 and a go or no-go decision on day 90. Every check uses the same two numbers: hours, and cash.
Day 90 has three outcomes: go on, change the scope or stop. Stopping with evidence is a good result. It costs less than drifting.
Score your starting pointExhibit 11 · Try itScore where your business stands before you startAnswer ten questions about your data, systems, ownership, people and risk. Your score, your weakest area and what to fix first show straight away. It asks for no name or email.Open the check ↓Close ↑
Illustrative scoring for planning, not an audit or a promise of results. The five areas follow Deloitte’s AI maturity index (2025 survey); the score and bands are my own design. It does not replace privacy or legal advice: see the OAIC guidance on AI products or get advice on your own obligations.
Ten questions on data, systems, ownership, people and risk. Each answer scores 2, 1 or 0. You get a total, your weakest area and what to fix first. The score is worked out in your browser, and the check asks for no name or email.
The five areas follow Deloitte’s AI maturity index, which is survey language, not an audit. Use the score to start the conversation with your team.
What to do first
Exhibit 12
Five small steps start a pilot that can be judged at day 90
01List ten recurring tasks
Write the weekly hours, the cost of one error and where the data lives.
02Pick one and time it
Two weeks of real timings is your baseline. Do not estimate.
03Fix its data and name an owner
One record, one system, one person who answers for the pilot.
04Write the data rule
Which information may go into an AI tool, and which may not.
05Run 90 days, then decide
Check at day 30 and 60. At day 90 report hours, cash and costs apart.
Start this week. Each step is small and none needs a purchase.
Common questions
How can AI help a small business?
By taking repeated, rule-based work such as bill entry, enquiry sorting and report drafts, so people spend their time on judgement and customers. Start with one task that has real hours and data that is ready.
Can AI cut staff costs, or does it free up time?
Both can happen and they are different results. Hours freed are time. They become cash only when a paid hour goes, such as less overtime or a hire avoided. Show each on its own line.
How long before AI pays back?
It depends on the task, the running costs and what happens to the paid hours, so no fair figure exists without them. Measure a baseline, then compare at day 90 with the software and support costs included.
Do I need an AI strategy first?
No. Start with one task, one owner and one baseline. A wider roadmap comes after the first pilot shows what works in your business.
Why do AI implementations fail?
Usually because nobody owns the result, there is no baseline, the data sits in several places or checking the output takes longer than the work. Each one is cheap to fix before the pilot starts.
Do I need an AI policy before staff use AI?
Put one in writing early. MYOB found just 11% of SMEs had an AI policy in 2025. One page is enough: which information may go into which tool, and who checks the output.
Chartered Accountant (CA ANZ) and finance executive, based in Canberra. Practical AI and automation for SMEs, and adviser to owners, CEOs and boards on integration and finance.
Is the problem wider than one task? I lead the roadmap, the data and the change across the business, starting with a Diagnostic that ends in clear priorities.
Renzo Demartini