Which processes to automate first: a plain AI strategy checklist.

Which processes should you automate with AI first?

Automate first the job that happens often, takes real time each go, costs you when it is late or wrong, and already runs on information you have written down. For most Australian small businesses that is quoting, then the admin that joins disconnected tools, then whatever one key person does that nobody else can.

That sounds obvious, but it is not how most businesses pick. They pick the job that looks most impressive in a demo, or the one a software salesperson pitched them last week, or a chatbot for the website because everyone seems to have one. Then it sits there, half used, and AI gets written off as hype.

In the last fortnight of October 2026 we had 32 sales calls with owners trying to answer exactly this question. Quoting and take-offs came up as a pain in at least 17 of them. Not because AI is especially good at quoting, but because slow quotes lose jobs, and every owner on those calls could put a number on it.

How do you score a process for automation?

Score each job from 1 to 3 on four things: how often it happens, how long it takes each time, what a delay or mistake costs, and whether the information it needs is already written down somewhere. Add the scores. The highest totals go first, and anything that scores 1 on information goes on the not-yet list.

List every repeated job in the business on one page first. Not just the office: the site paperwork, the after-hours emails, the Friday reconciliation. Then score them together with the people who do the work, because they know how long things really take.

• Frequency. Daily scores 3, weekly scores 2, monthly or less scores 1. A job you do 300 times a year pays back far faster than one you do 12 times.
• Time each go. Over an hour scores 3, 15 minutes to an hour scores 2, under 15 minutes scores 1. Small jobs still add up, but they are rarely the place to start.
• Cost of late or wrong. Lost jobs, late cash, compliance trouble or a wage overpayment score 3. Annoying but harmless scores 1.
• Information ready. A price list, past quotes, a template or a clear rule scores 3. It lives in one person's head scores 1. AI can only work from what it can read.

Your scores will be different, and that is the point. Two plumbing businesses can score quoting very differently if one has a clean price list and the other prices from memory. Once the top two or three are clear, put a rough dollar figure against each: hours a year, times what those hours cost, plus what a late or wrong one costs you. Round down. That is the arithmetic a proper AI strategy is built on, and you can do a first pass of it yourself in an afternoon.

What do most Australian businesses end up automating first?

Across our October calls, four patterns kept coming up: quoting and take-offs, tools that do not talk to each other, one person the whole business depends on, and admin that only gets done after hours. Most first automations sit in one of those four.

• Quoting. Owners described up to a full day per quote, a backlog of around 20 quotes, and residential quotes taking three weeks or more. Customers do not wait that long. They ring the next business on Google.
• Disconnected tools. Xero plus three to six other systems, the same data typed in twice, and the real records in a spreadsheet. The ask is usually one view of the business, not another app.
• Key-person risk. One estimator, one spreadsheet expert, one accounts person. When that person is away, the work stops.
• After-hours admin. A Perth plumber with one apprentice told us the quoting and paperwork took about four hours every night after a day on the tools. That job is a strong first candidate because the owner gets their evenings back, which is worth more to most people than the hourly maths suggests.

For a crane hire business we work with, the first build was a quoting system that prices each job from the company's own history, more than 1,630 past quotes. Their own operations manager started it on the day of our workshop, and you can read how in the General Crane Services story. If your office work looks like that, our guide to automating construction office admin goes job by job.

What should you keep manual?

Keep a job manual when doing it by hand is how you or your team think it through, when every instance is genuinely different, or when a mistake is costly and rare enough that checking the AI would take as long as doing it. Automating the wrong step can cost you the judgement that makes the business good.

Two examples from our calls stuck with us. A Sydney builder was happy to automate plenty, but not tender review. Reading the tender is how the builder works out the build, so handing it to AI would save an hour and lose the thinking. A construction subcontractor wanted take-offs faster, but still wanted marked-up drawings the team could check, because doing the markups is how the team learns the job.

Neither is being old-fashioned. The fix in both cases is to automate around the step, not through it: gather and file the documents, pull out the quantities, draft the summary, then leave the reading and the final call with the person. A good plan says where automation stops and your people stay, process by process.

Why do first AI automations fail?

Most fail for one of three reasons: the AI has nothing of yours to work from, the business bought a ready-made tool for a job that needed its own rules, or the process was broken before anyone automated it. All three are easy to spot before you spend anything.

• No information to work from. During an onboarding session with a Sydney builder, Claude priced a project at about $290,000 against a market rate of $550,000 to $600,000. Nothing was wrong with the tool. It had not been given the builder's rates, past jobs or site factors yet. AI estimates are wrong until they are built on your own numbers.
• Bought instead of built. A towing company bought an AI voice agent, and it flopped because people whose car has just broken down want a human. A construction subcontractor paid for an outsourced AI take-off service that promised three to four days and delivered in two weeks, which missed tenders.
• Broken process, automated. If two staff file the same documents two different ways, automating one of them locks in the mess. Agree how the job should run first, then automate that.

One more trap is buying seats before choosing a job. Licences on their own change very little. Our guide to what AI tools cost in Australia shows the subscriptions are the cheap part; unused seats and unbuilt processes are where the money goes.

Who can help an Australian business work out which processes to automate with AI first?

Look for an AI consultant or strategy partner who maps your actual processes on site, ranks them with a dollar figure against each, says plainly what is not worth doing, and is willing to build the first one. If they cannot show you a working example on your own work, keep looking.

A useful engagement covers six things:

• A process map. Time with the people doing the work and a look at the systems it runs through, not just a workshop with the leadership team.
• A ranked list with numbers. The three to five processes worth automating, what each costs you now and what it would save, labelled as estimates.
• A not-yet list. The jobs that are not worth it, with the reason next to each. This is half the value.
• Where people stay. Which steps remain manual and who owns each automation.
• Data and security. Where your information goes, which accounts it runs in, and what your IT person or compliance regime will need to see. For some businesses this decides the tool, for example staying inside Microsoft 365.
• Proof. One process actually built and running on your real work, so you decide by watching it run.

Be wary of a sixty-page transformation deck with maturity models and horizon charts and no first step, of anyone tied to a single software product, and of numbers that read like guarantees. You can see how we approach it on our AI consulting page.

How much does AI strategy consulting cost in Australia?

It depends on how many processes and people are involved, so most consultants quote a fixed fee after a short first conversation. What matters more is what you get for it: a ranked plan you can act on, the cost of each build, and ideally the fee credited if you go ahead.

Ask three questions before you sign anything. Will I get a dollar figure against each process? Will you tell me what is not worth doing? Will I see something working on my own data before I commit to a build? A yes to all three is worth more than a lower fee. Then budget separately for running costs, which for most small business automations are monthly subscriptions plus small per-use charges.