What should an AI implementation plan include?
An AI implementation plan for a small business should cover one job, the person who owns it, the number it is meant to move, the data and rules it needs, and a dated path from first test to a yes or no verdict. If it fits on one page and has dates on it, it is a plan. If it has a vision statement and four workstreams, it is a wish.
Most owners are handed the second kind: a roadmap with phases and a slide about transformation, none of which says what happens on Monday. The plan below is the first kind, with a timeline and goals you can check. It works the same for a plumbing business in Penrith, an accounting practice in Parramatta or a franchise owner with three sites.
Plan one job, not an AI strategy
The first job decides whether there is a second one. Pick the use case behind your loudest pain points: the process that costs real time every week, has clear inputs and a clear finished state, and has someone who knows what good looks like. Our guide to what to automate with AI first has a four-part test for choosing it. Write the plan for that job alone, then scale to the next one with everything the first taught you.
What does the plan look like week by week?
It runs in five stages over roughly two months. Each stage has a short list of things that are either done or not done.
Before week one: choose, name, measure.
- Name the job in one sentence, in the words your staff use for it.
- Name one owner, a person with authority to say yes, no and stop.
- Capture the baseline: how long the job takes now and how often it goes wrong. This is the KPI the whole plan answers to, and a week of honest tallies beats a guess.
Weeks one and two: map it, check the data, write the rules.
- Map the job step by step, including the awkward cases. The National AI Centre publishes a free business process mapping template and a data quality checklist on its planning tools and templates page.
- Find where the inputs live (inbox, job system, shared drive, accounting software) and note any integration the job needs.
- Write down what the AI may do alone, what needs a person to approve, and what it never touches. The same page has a government AI policy template.
Weeks two to four: build and test on your own history.
- Test against real past examples: last month's enquiries, last quarter's invoices, the quotes you actually sent.
- Keep human oversight on every output until the results earn trust. Approval gates come off one at a time.
- Brief the team and do the training while the job is still in test, not on go-live day.
Go-live: write down what "live" means.
- Live means running on real work, every day, with the owner watching. Agree that before the build starts, so nobody declares victory on a demo.
Thirty days after go-live: the verdict.
- Compare against the baseline. Did the number move, by how much, and what broke? One page, a yes or a no, and the start of any ROI conversation. Our post on how to run a 30-day AI pilot covers how to set that verdict up properly.
Who should own the plan in a small business?
One senior person, usually the owner or a partner, who can make decisions without calling a meeting.
The Australian Government's Guidance for AI Adoption: foundations puts accountability first: assign a senior leader as the overall AI governance owner, then make a specific person accountable for every AI system you use. In a ten-person business that may be the same person, which is fine. "The IT guy" or "the vendor" is not, because neither owns the outcome.
What Australian rules does the plan need to account for?
Mainly privacy, plus the government's voluntary guidance on responsible AI use. There is no separate AI licence or registration for a small business to obtain.
- Privacy. If the job touches personal information about customers, patients, tenants or staff, the Privacy Act 1988 and the Australian Privacy Principles shape what you can do with it. Some small businesses are exempt, but the exemption has exceptions. The OAIC's guidance on privacy and commercially available AI products sets out what it expects.
- The Guidance for AI Adoption. Published by the National AI Centre on 21 October 2025, it sets out six essential practices and updates the earlier Voluntary AI Safety Standard. It is voluntary. The useful parts for a small business: name who is accountable, screen each use for risk, keep a simple register of your AI tools, tell customers when they are dealing with AI, and keep a person in control.
- Everything else still applies. Consumer, workplace and anti-discrimination law do not pause because software made the decision. If the job touches hiring, credit or a person's rights, get proper advice first.
This is general information, not legal advice.
Choosing who builds it
Whoever builds it, staff member, vendor or a builder like us, should show the job working on your own data before you commit, and agree the price and the definition of "live" in writing. Our checklist for choosing an AI implementation partner has nine questions to put to anyone, including us. If the job is admin, such as invoices, data entry, reconciliations or reporting, the admin disappears shows how we scope that kind of work.
What usually derails the plan
Scope creep, a missing baseline and staff who heard about it last.
- Scope creep. A second job gets added in week two. Park it on a list for the next plan.
- No baseline. Without a before number, the verdict is an opinion. Measure first, even roughly.
- Data that is not where you thought. The inputs live in three places and one person's head. That is why the map comes before the build.
- Staff left out. The people doing the job today know every exception. Bring them in at the mapping stage and be straight about what changes for them.
Frequently asked questions
Do I need a consultant to write the plan?
No. An owner who knows the job can write it in an afternoon using the structure above and the free government templates. Outside help earns its keep at the build and test stages.
What is the difference between an AI implementation plan and an AI strategy?
A strategy says where AI might fit across the business. A plan says which job, who owns it, which number moves and by when. Three plans in a row usually beat one strategy.
Is the Voluntary AI Safety Standard still current?
It has been updated by the Guidance for AI Adoption, published on 21 October 2025, which condenses its guardrails into six essential practices. Both are voluntary. Read the current version on the National AI Centre's site before relying on any summary, including this one.
See a job working before you plan the rest
The quickest way to sharpen a plan is to watch one job done well. The test bench lets you put the crew to work on a staged business, no sales call attached.