Mac Sweeny
Quick answer
AI adoption fails on sequencing and people, not on technology. The pattern is consistent: nobody owns it, the first project was chosen because it looked impressive rather than because it hurt, and a tool was bought before anyone defined the job it was doing. A sequence that works is: assess where the hours go, pick one workflow, prove it on real work for four to six weeks, then expand. Adoption consulting is help making those calls. AI Consulting is A$1,195 per month + GST, with a call each fortnight, clarity on what to build in Claude or ChatGPT and practical guidance on how to do it. No lock-in.
From an idea to a useful habit
I would start with one workflow and give someone ownership of the trial.
- 01
Pick the work
Choose a repeated task and record how it works today.
- 02
Try a small version
Use a few examples and keep a person checking the output.
- 03
Review the result
Compare time, rework and usefulness before rolling it out.
If the trial adds more checking than it saves, change the approach before expanding it.
Why AI projects stall
Almost every stalled programme I see fails for one of three reasons, and none of them are about the model you picked.
Nobody owns it
The project is everyone’s priority and no one’s job. It gets discussed at every leadership meeting and moved forward between none of them. A project needs one name against it, with a few hours a week that are genuinely protected, and a decision-maker who will unblock it inside a day.
The wrong first project
The first project is usually chosen because it would look good, not because it hurts. Customer-facing chat is the classic: high visibility, high risk, hard to measure, and it competes with a support process that mostly works. The boring internal one, getting job notes into an invoice, drafting the quote, chasing the follow-up, is worth more and fails more safely.
Tool-first thinking
Someone buys the licences, the licences arrive, and only then does anyone ask what job the tool is doing. Six weeks later the seats are 80 per cent unused and the conclusion is that AI does not work here. The order matters: define the workflow, prove it, then buy the seats for the people who proved it.
A sequence that works
Four steps, in this order. The discipline is refusing to start step four before step three has produced honest numbers.
| Step | What you do | How long | What done looks like |
|---|---|---|---|
| 1. Assess | Map where the hours go across your busiest processes, check whether the data is reachable, and find who would own the change | Depends on the scope | Three to five candidate workflows, ranked, with the reasons written down |
| 2. Pick one | Choose the most repeated, most rule-based workflow that has a willing owner attached to it | A day | One workflow, one owner, one measure everyone agreed to before starting |
| 3. Prove it | Run it on real work alongside the existing process, not on a demo dataset | Four to six weeks | Measured time before and after, how often output needed correcting, and whether people used it unprompted |
| 4. Expand | Only once step 3 pays, take the next workflow and reuse what you learned about rules, prompts and sign-off | Ongoing | A second workflow live and a written rule for what qualifies next |
The people part is most of the work
Technology adoption inside a small business is a permission problem before it is a skills problem. Staff will not put company information into a tool until someone senior has said which information is allowed and confirmed that nobody gets in trouble for trying. Write that down in a paragraph, not a policy, and say it out loud in a meeting.
The training that changes behaviour is a working session on somebody’s real task, in their own system, with their own messy inputs. A demo of what the tool can do in general changes nothing. AI training for employees goes into the format that sticks.
Resistance from the middle of the organisation is usually rational rather than stubborn. The person raising objections is often the one who will be blamed when the output is wrong. Give them the sign-off role rather than arguing with them and the objection tends to disappear.
What adoption consulting actually adds
Three things you cannot easily produce internally. An outside view of which of your candidate projects is genuinely the good one, based on having watched similar ones succeed and fail. A list of what to leave alone, which is worth more than the list of what to do. And a sequence, so the second project benefits from the first instead of starting from zero.
AI Consulting is A$1,195 per month + GST, with a call each fortnight, clarity on what to build in Claude or ChatGPT and practical guidance on how to do it. No lock-in.
This is advice, not delivery. Building the systems is separate from consulting and I do not sell the software I recommend, which is the point. If building turns out to be the right move, that is a separate engagement with its own scope, price and contract, and you are free to take the plan elsewhere.
How to tell whether it worked
Adoption programmes get measured on activity, which is why so many of them report success and change nothing. Measure four things instead, and measure the first one before you start.
- Hours on the task, timed before and after by the person who does it. Not estimated in a meeting.
- Weekly active users, not licences bought. A seat that gets opened twice a month is a cost, not an adoption.
- Rework rate: how often the output has to be corrected before it is usable. If this climbs, the time saving is imaginary.
- Whether the old process got switched off. Until it does, you are running two processes and paying for both.
The switch-off test is the honest one. If the spreadsheet is still being maintained in parallel six weeks in, nobody trusts the new thing, and the reason for that is worth more than any usage dashboard.
A realistic first ninety days
Weeks one and two: assess and choose, with the measure written down. Weeks three to eight: run one workflow on real work, with a named owner and a weekly ten-minute check. Weeks nine to twelve: read the numbers, switch off the old process or stop the project, and only then decide what is next. If you want the assessment step in more detail, the AI readiness assessment guide has a ten-question version you can run yourself.
Official resources
Explore the tools and check current details with the people who make them.
