Why a roadmap comes first
If you run a business in Melbourne in 2026, you have probably had the same week a hundred owners have had. A supplier mentions their quoting is “done by AI now”. A staff member asks whether they can use ChatGPT for customer emails. Microsoft keeps putting Copilot buttons in front of you. A software vendor offers an “AI agent” for a monthly fee. Every one of those might be useful. None of them tells you where to start, what it will cost, or how you will know it worked.
The businesses that get value from AI tend to do three unglamorous things well. They pick one job to improve rather than “doing AI”. They set rules for data and staff use before something goes wrong. And they give it enough time with real work that people stop treating it as a novelty. A 30/60/90-day roadmap is just a way of making those three things happen in order, with dates and names next to them.
We are a consulting service, so we will be upfront about what we are selling. The first conversation is free and short. If it makes sense to go further, SG1 Consulting runs a paid two-week audit that produces the written roadmap, then a paid pilot that builds the first one or two automations. You can also take the plan on this page and run it yourself. It is written to be usable either way.
The plan, stop by stop
Think of it like a tram route rather than a project schedule. There are three stops. You can’t skip one, and at each one you get off, look around and decide whether to keep going.
Stop one: decide what you are actually trying to fix
The first month is about choosing, not building. The output is a short written document, not software.
- List where the hours go. Sit with the people doing the admin, not just the managers, and write down the repeated jobs: quotes, purchase orders, rosters, patient letters, enrolment paperwork, invoice chasing, compliance forms. Note roughly how often each happens and how long it takes.
- Score each job on three things. How much time it eats, how predictable the inputs are, and what happens if the AI gets it wrong. A job that is frequent, fairly predictable and easy to check by a person is a good first candidate. A job where a mistake reaches a customer or patient without anyone seeing it is not.
- Take stock of what you already pay for. Many businesses already hold Microsoft 365 or Google Workspace licences, an accounting system and an industry tool, and some of those now include AI features. The cheapest first step is often switching on and governing what you have.
- Find out what staff are already doing. Ask, without blame, which AI tools people use today and for what. You need this to write a sensible policy, and it usually reveals the best use cases.
- Write a one-page interim rule. Until the full policy exists: no client, patient or student personal information in free public AI tools, and AI-drafted material is checked by a person before it leaves the building.
- Pick one or two pilots and name an owner for each. Not “IT”. A person who does the work and wants it to be easier.
What you should have at day 30: a ranked list of candidate jobs, a chosen pilot with a named owner, a baseline measurement of how long that job takes now, an inventory of tools and licences, and an interim AI use rule that staff have actually read.
Stop two: prove one thing with real work
The second month is a controlled pilot. The goal is evidence, not a launch.
- Build the smallest version that does the real job. If the pilot is order entry, that means reading real order emails and drafting the order in the real system for a person to approve, not a demo on sample data.
- Put a person at the approval point. Every automated step that changes a record, sends a message or commits money waits for a human tick, at least during the pilot. You can relax this later, step by step, once you have evidence.
- Keep a log. Every draft, every approval, every correction. The corrections are the most valuable data you will get: they show where the AI is reliable and where it is not.
- Measure against the day-30 baseline. Time per item, error rate, and how often a person had to rewrite the output. Use the same measure you took before, not a new flattering one.
- Train the pilot group properly. An hour of hands-on time with the actual workflow beats a generic AI course. People need to know what the tool is good at, where it fails, and what they are still responsible for.
- Hold a stop-or-go review at day 60. Is it saving time? Is the error rate acceptable with a person checking? Do the people using it want to keep it? If the answer is no, you have spent a month and learnt something. That is a cheap lesson.
Pilots fail for predictable reasons: the inputs turn out to be messier than anyone said (handwritten notes, scanned faxes, five different spreadsheet layouts), the owner gets pulled onto something else, or the approval step becomes a bottleneck because only one person can do it. Planning for those at day 30 is cheaper than discovering them at day 55.
What you should have at day 60: a working pilot used on real work, a log of approvals and corrections, before-and-after numbers, and a written go, change or stop decision.
Stop three: make it normal, then choose the next stop
The third month turns a pilot into how the work is done, and sets up the next round.
- Write the proper AI use policy. Replace the interim rule with something specific: approved tools, what data can go where, who approves what, how mistakes are reported, and how often the policy is reviewed. The governance page walks through it.
- Roll the pilot out to the rest of the team that does that job. Staggered, with the pilot group helping the next group, not all at once on a Monday.
- Decide which approval steps can loosen. If a person has approved 400 order drafts and corrected three, maybe low-value orders from regular customers can go straight through while anything unusual still waits. Write the rule down.
- Record what the AI is allowed to touch. A simple register: each automation, what systems it reads and writes, who owns it, when it was last reviewed. This is one of the practices in the Australian Government’s Guidance for AI Adoption, and it is easy to keep at small-business scale.
- Choose the next one or two jobs from the day-30 list. With the first pilot done, the second is faster: the policy exists, staff know the pattern, and the integrations are partly in place.
What you should have at day 90: one automation in normal use, a real AI policy, an AI register, a trained team, numbers you can show your accountant or board, and the next pilot chosen.
Where Microsoft 365 Copilot fits in the plan
Because so many Melbourne businesses run on Microsoft 365, the Copilot question usually comes up in the first week. Our honest view is that Copilot and custom automation solve different problems. Copilot helps an individual with their own work inside Word, Outlook, Teams and Excel, and it can only see what that person already has access to. Custom automation takes a whole process, such as order entry or invoice chasing, and runs it across systems with approval steps. Many businesses end up with both. Some need neither yet.
The plan handles this simply: in the first 30 days, you check whether your permissions and file sharing are tidy enough for Copilot to be safe (Copilot surfaces anything a user can technically open, including the folder someone shared with “everyone” in 2019), and you decide whether your top-ranked job is an individual productivity problem or a process problem. The Copilot or custom page goes through the decision in detail.
The part most plans skip: people
Most AI projects that stall do so because people don’t use the thing, not because the technology failed. Staff worry that it is a step towards replacing them, or they tried it once, it got something wrong, and they quietly stopped. A good roadmap treats adoption as work with its own tasks:
- Tell people early what the plan is, what it is for, and what it is not for. If the goal is to stop your admin team working back on Thursday nights, say that.
- Choose pilot users who are respected and a bit sceptical. If they are convinced, others follow.
- Make it easy to report a bad output, and act on the reports visibly.
- Keep the human responsibility clear. The person who approves a quote is still responsible for the quote.
Why the plan looks different in different Melbourne industries
The structure stays the same, but the first pilot changes a lot by industry. A manufacturer in Dandenong South will usually start with orders, supplier paperwork or quality records. A freight business near the port or in the western warehouse belt around Truganina and Laverton North tends to start with booking emails, proof-of-delivery paperwork and customer status updates. An allied health practice needs to start with admin that sits away from clinical judgement, under tighter privacy rules for health information. A training provider looks at enrolment and evidence paperwork. A hospitality group looks at rosters, supplier invoices and enquiries. The Melbourne sectors page has a starting point for each.
Running the plan yourself
You do not need a consultant to follow this. A capable operations manager with protected time, a supportive owner and a willingness to measure honestly can run all three stops. Where outside help tends to pay for itself is in the first two weeks (an outsider is often better at spotting which jobs are worth automating, because they are not used to the workarounds) and in the pilot build, where connecting systems and setting up approval steps is technical work.
If you want to try it alone first, start with the day-30 checklist above. If you get stuck, or you want a second opinion on your ranked list, book a 15-minute call and bring it with you.
Common questions
Do we need a 90-day plan, or can we just start using ChatGPT or Copilot?
Plenty of Melbourne businesses already have staff using public AI tools without a plan. That is usually where the trouble starts: client data pasted into personal accounts, no agreed rules, and nobody able to say whether it saves time. A 90-day plan does not slow you down; it gives the experimenting a direction and a finish line.
What does the first 15-minute call cost?
Nothing. It is a short conversation about one bottleneck in your business and whether AI is a sensible fix for it. If it is not, we will say so.
How much does the roadmap itself cost?
The written plan comes out of SG1 Consulting’s AI Opportunity Audit, which takes about two weeks and starts from $2,500. A pilot of one or two automations runs 30–60 days and starts from $5,000. Ongoing scale-up work is priced to the work. The engagements page has the detail.
Is this only for Microsoft 365 businesses?
No. A lot of Melbourne SMEs run on Microsoft 365, so Copilot is often the first question, but the plan works the same way on Google Workspace, Xero, MYOB, a CRM such as HubSpot or Pipedrive, or industry software with an API.
Do you come on site in Melbourne?
The work is run remotely by SG1 Consulting, an Australian business, over video and phone. Melbourne and Sydney share a time zone, so calls land in your normal working day. We do not claim a Melbourne office.