Start with the job that repeats daily

There is a moment in almost every discovery session where I ask which job is eating the week. Nobody needs time to think about it. They know. It is usually the one that gets done at six o'clock, after the phones stop.

Writing up site notes. Chasing a quote that went quiet. Answering the same six questions that come through the website. Setting up a new client from scratch for the eleventh time this quarter.

That job is almost always the right place to start. Here is how to choose it, test it before spending anything, and put enough control around it to make the improvement last.

The projects that work are often the boring ones

The work worth automating first usually has three features.

• It has a shape you can draw. Four or five boxes, arrows between them, and general agreement that this is how the job moves.
• It has an owner. One person can describe it in a sentence, knows where it breaks and feels the cost when it does.
• It produces an observable result. An email goes out, a quote lands, a report gets filed or a client receives the right next step.

Work that fails those tests may still be worth improving. It is simply a poor first project because the team spends its time arguing about scope instead of proving that the change works.

Three questions to test any candidate

Before choosing a tool, run the job past three questions.

1. How often does it happen?

Weekly is a useful minimum. Daily is better. Automation creates value each time the workflow runs, so a job that happens twice a year rarely returns much, regardless of how frustrating it is.

2. Does it follow roughly the same steps?

"Mostly" is a good answer. You are looking for a stable spine with a manageable set of exceptions. If every instance is genuinely different, you may be dealing with a judgement task rather than a repeatable process.

3. Does everyone agree on a good outcome?

This is the question people skip. If two people would describe success differently, you cannot test whether the new workflow works. Settle the definition before you build.

What this looks like in practice

Across the businesses we support in Western Australia and South Australia, the same kinds of recurring work appear:

None of these projects needs to be technically novel. They happen frequently, follow a recognisable sequence and have a clear definition of done.

Why not start with the biggest opportunity?

The interesting problems are usually the messy ones. There is always a case for pursuing the biggest number first, often supported by a business case that makes the small recurring job look unimportant.

Those projects generally take longer and contain more failure points. They can also use up the goodwill needed for everything that follows, because your team forms an opinion of AI during the first project and carries that opinion into the next one.

One workflow that reliably returns four hours a week gives you stronger evidence than a projected saving in a business case. It also creates a practical foundation for funding and designing the harder project.

Ask what the problem costs before asking where AI goes

The useful question is not, "Where can AI go in this business?" It is, "What is this problem costing us, and what is the simplest reliable way to fix it?"

Sometimes the answer involves AI. Often it is a form connected to a spreadsheet, a template that has never been documented, or two existing systems that need to exchange information. When those fixes are cheaper, faster and easier to maintain, use them.

Starting with the cost and workflow keeps the project grounded. Starting with the technology can produce a clever solution to a problem that was never worth solving.

Keep a person at the end of the first version

For drafting and communication workflows, a sensible first version prepares the work and a person reviews or sends it.

This supports quality, but it also builds trust. The people who do the work can inspect the output for a few weeks, identify exceptions and decide which steps are safe to automate further. High-impact or consequential decisions should retain clear human accountability and escalation.

Measure the workflow before you build

Write down two numbers this week:

1. How many times does this job run in a normal week?
2. How long does one run take from start to finish, including the thinking and handoffs around it?

Multiply them to establish the baseline. Measure the same numbers one month after launch. That comparison tells you whether the change returned time and gives you real evidence for the next project.

What a good first month looks like

For Motivate Property Group, website enquiries had to move through several different service pathways. Elevion AI built a controlled website inquiry agent that asks relevant questions and passes useful context into the existing CRM.

The verified early result was three qualified leads in the first eight days. The point is not that every business should install the same agent. It is that the workflow had a narrow scope, an observable outcome and a result the team could review.

That is the shape to aim for: practical, controlled and measurable.

Common questions

What makes a good first AI project for a small business?

A job that runs at least weekly, follows roughly the same steps and has an outcome everyone would describe in the same way. Enquiry response, report drafting, quote follow-up and client onboarding often qualify.

How long should a first project take?

Think in weeks rather than months. If the first release cannot be tested within about six weeks, look for a smaller workflow or split the project into controlled stages.

Should we buy an off-the-shelf tool or build something custom?

Buy where the process is standard across many businesses, such as scheduling or invoicing. Consider a tailored workflow where the process depends on your terminology, controls, systems or professional judgement.

What if the process is messy?

Map it first. Some apparently messy processes are simply undocumented. A whiteboard session can reveal a stable sequence with a small number of exceptions; genuinely variable work may need a different approach.

How do we know it worked?

Compare the same baseline measures one month after launch: runs per week and minutes per run. Add an outcome measure where relevant, such as completed reports, qualified enquiries or quotes followed up.

Where to start

If you already know which job is eating your week, write down its steps, frequency, current time and definition of done. That is enough to begin a useful conversation.

Explore Elevion AI's business process automation, or Chat with Ethan about the recurring workflow you want to improve.

Ethan Sandery

Founder, Elevion AI — AI and automation for growing Australian businesses.