What should a growing business automate first?

The short answer: automate the boring, repetitive, high-frequency task that already slows your team down, the one that happens the same way several times a week, follows a clear rule, and needs no judgement. Not the impressive thing. The tedious thing.

Most businesses get this backwards. They start with whatever sounds most futuristic, a chatbot, a fancy AI report, and end up with a demo that impresses in a meeting and gets quietly abandoned two weeks later. The automations that actually stick are unglamorous. That is the point.

Find the task, not the tool

Before looking at any software, spend a week noticing where your people do the same thing over and over. The strongest candidates share three traits: they happen often, they follow a predictable pattern, and they are already a source of low-grade friction. If a task is rare, or changes shape every time, or genuinely needs a human decision, leave it alone for now.

A useful test: if you can write down the rule for a task in a few if-then sentences, and it happens most days, it is a strong first automation candidate.

Common first automations that earn their keep are the plumbing between systems: moving an enquiry from a form into your CRM, sending a follow-up when a step is reached, generating a routine document from data you already hold, or reconciling two lists that should always match. None of these will impress anyone. All of them return time.

Go narrow and deep, then stop

Pick one process. Automate it end-to-end. Let it run for a couple of weeks and watch it in real conditions before you touch anything else. This discipline is what separates automation that compounds from automation that becomes a fragile mess nobody trusts. A second layer added on top of an unproven first layer is how businesses end up with systems more brittle than the manual work they replaced.

The goal in the first month is not scale. It is one workflow that runs reliably whether or not anyone remembers it exists.

Where AI actually helps, and where it doesn't

AI is genuinely useful for a narrow set of jobs inside a workflow: reading messy input and turning it into structure, sorting or classifying things, and drafting text a human then approves. It is not a reason to automate. The friction is the reason; AI is occasionally the right tool for one step of the fix.

Be wary of anything that puts an AI in charge of an irreversible decision, or that depends on the model being right every single time. Automate the stable, rules-based scaffolding; keep human judgement where judgement is actually needed.

A simple order of operations

1. 1Watch a normal week and list the tasks repeated most often.
2. 2Cross out anything rare, changeable, or judgement-heavy.
3. 3From what's left, pick the one that costs the most time or causes the most dropped balls.
4. 4Automate that one thing end-to-end, and only that.
5. 5Run it for two weeks. Measure the time saved and the errors avoided.
6. 6Only then consider the next one.

If you do this honestly, the first automation is usually smaller and duller than you expected, and worth more than the ambitious one you had in mind. That's the job.