What to automate first (and what to leave alone).

The first thing most people want to automate is the thing that annoys them most. That's almost never the thing worth automating first.

AI made automation cheap enough that everyone's racing to apply it. But automating the wrong task is worse than doing nothing — you've now spent money to make a bad process run faster. Speed is only a virtue when the thing you're speeding up is worth doing.

Here's the framework I use with clients to decide what goes first.

One. Score the work, not the annoyance.

For every repetitive task, ask two questions: how many hours a week does it eat, and how much judgment does it really need? Plot those against each other and the priority order writes itself.

• High hours, low judgment — automate first. This is the gold: data entry, lead enrichment, report assembly, invoice matching, scheduling.
• High hours, high judgment — augment, don't replace. AI drafts, a human decides. Think proposals, hiring screens, content.
• Low hours, low judgment — automate later, if ever. Small wins, not worth the setup yet.
• Low hours, high judgment — leave it alone. This is your actual work.

Two. Never automate a broken process.

Automation is an amplifier. Point it at a clean process and you get leverage; point it at a messy one and you get a faster mess — plus a system nobody understands well enough to fix.

If you can't describe a process in clear steps, you're not ready to automate it. You're ready to fix it.

The fix is usually the hard part, and it's where most of the value hides. Half the "automation" projects I run end up being process-cleanup projects that made automation unnecessary.

Three. Keep a human at the edges.

The best automations aren't fully autonomous — they're a clean middle with a human at the start (deciding what matters) and a human at the end (catching what's wrong). Full autonomy sounds impressive and quietly erodes trust the first time it does something dumb at scale.

Design for the failure case from day one: what happens when the input is weird? Where does a human get pulled in? An automation you can't trust is one you'll end up checking manually anyway — which means you've automated nothing.

The honest test

Before you automate anything, run it through one question: if this ran perfectly and invisibly, would the business actually feel it? If the answer is "not really," you've found a task that's annoying, not important — and there's a more valuable hour somewhere else.

Start there. One clean, high-leverage automation that the business genuinely feels beats ten clever ones that just make the noise faster.