How to choose the right AI tasks — and the tools to run them
8 minute read · Written by James Clements, Skills for AI, Williamstown
The hard part of AI is not the technology, it is deciding what is worth paying for. Here is the same process we use in the 45 - 60 minute audit, written so you can run a first pass on it yourself.
The short answer
Start with tasks, not tools. Identify three to seven tasks AI can readily carry out with a clear return — hours saved, faster turnaround, more output, better quality or higher accuracy — put an estimated benefit against each, and only then work out the smallest set of tools that can run them. Typically one or two tools are enough; more only when the tasks are genuinely diverse. Then save the task into the tool so staff run the same process every time.
Six steps to your task list
1. Write down the tasks, not the tools
List the five tasks that eat the most hours in your week — quoting, chasing invoices, writing proposals, answering the same customer questions, rostering, reporting. Put a rough number of hours next to each. This list, not a software review site, decides what you buy.
2. Sort each task into repeatable, judgement or messy
Repeatable tasks with a clear input and output are where AI pays fastest. Tasks that need your judgement can be assisted but not handed over. Messy tasks that live across paper, memory and three inboxes need tidying up before any tool will help.
3. Put a number on the benefit of each task
Hours saved a week, faster turnaround, more jobs out the door, fewer errors or a better standard of output. Three to seven tasks with an estimated return is the shortlist worth acting on — this is exactly what the audit report sets out.
4. Then work out the smallest set of tools that covers those tasks
One paid seat on a general assistant like ChatGPT or Claude usually handles most of the list. Add a specialist tool only where a task genuinely cannot be done in what you already have. Typically one or two tools are enough; more only when the tasks are diverse.
5. Count the whole cost, not the sticker price
Per-seat subscription, the hours it takes to set up, the training time, and the cost of the change not sticking. A $30 a month tool that nobody adopts is more expensive than a $90 one that saves four hours a week. Match the mix of tasks and tools to your budget against the benefit.
6. Turn the winners into a repeatable process
This is the step almost everyone skips. Instead of leaving staff with a blank prompt box, build the task into the tool — a Skill, CustomGPT, Claude Project, Gem, Workflow, Artifact or small App that already knows the context, steps and output format. Then the job is a button, not a guess, and results stay consistent when you are not watching.
Five traps that cost owners money
| Trap | Do this instead |
|---|---|
| Buying on a demo | Trial it on one real job from your own list, with your own messy data. |
| Ten tools at once | One tool, one job, one person. Add the next only after the first sticks. |
| Free tier forever | Pay for the seat once a job depends on it, so you get the better model and saved setups. |
| Pasting client data anywhere | Decide up front what may go into a public AI tool, and which tools are approved for client or staff information. |
| No owner for the change | Name one person responsible for each tool, even in a team of five. |
Common questions
- How much should a small business budget for AI tools?
- Most of the businesses we work with land between $50 and $300 a month in total, and a good share of that is subscriptions they already pay for. The number of tasks you automate is usually far higher than the number of tools you buy — one or two paid seats commonly cover three to seven tasks. Start with one, prove it saves time, then add another only if a task genuinely needs it.
- Is the free version of ChatGPT or Claude enough?
- For trying things out, yes. For a business process you rely on, the paid tiers are worth it because they give you the better models, file handling and the ability to save reusable setups your staff can open and run. That is where the repeatable time savings come from.
- Should I buy one AI tool that does everything?
- Start with as few tools as the tasks allow. One general assistant, properly set up for your specific jobs, handles most task lists. Add a specialist tool only when a task cannot be done well in what you already have — not because a platform promises everything.
- How do I stop staff quietly ignoring the new tool?
- Give them a setup that already knows the task instead of a blank prompt box. When the file, template or workflow is built for the job, the output is consistent and people keep using it. That is the part we build for clients during the audit and install.
What this looks like in practice
A local builder was losing Sundays to quoting. We set up a simple app in Claude holding his labour, equipment, support and overhead rates plus his standard proposal template — he enters material sizes and quantities, it checks current material costs, prices the job and generates the proposal. See more client stories.
Choosing AI tools in your suburb
Every suburb page sets out the AI tasks, tools and ROI we typically start with for businesses in that area:
Next steps
Where most owners go from here.
- How we augment AI tools for repeatable resultsWhat we build inside ChatGPT or Claude so a task runs the same way every time.
- AI for tradies and building businessesQuoting, variations and site reports, with estimated time saved.
- AI audit FAQsPlain-English answers on the 45 - 60 minute audit, cost and support.
- Areas we serve across Melbourne's westFootscray, Sunshine, Werribee, Truganina and surrounds.
Want a second opinion on your own task list?
Bring the jobs that eat your week and we'll talk through which of them AI can genuinely take on, and what that would be worth to you. The 45 - 60 minute AI audit is $599 incl GST, one-off, and includes your task and ROI report, guided set-up and AI basics training.
Prefer to talk first? Call or email us.