What to automate first, and what to leave alone

By the end of this you'll have a simple test for what to automate first, and the discipline to leave the wrong tasks alone.

The mess

When someone asks me to 'automate the business,' the project usually stalls, because it starts with the flashy thing. Meanwhile the real cost is hiding in plain sight: a dozen small daily handoffs nobody logs. Copy this into that. Chase a status. Re-key the same data into a third tool. No single one feels worth fixing, and together they quietly eat the week.

The wrong way people solve it

They automate what is exciting instead of what is frequent: a chatbot, a dashboard, a clever demo. Or they try to automate everything at once, and the project collapses under its own weight. Worst of all, they automate a process that is already broken, which just makes the mess arrive faster and more reliably.

The system view

Every task worth automating has the same shape, and if you cannot draw it, it is not ready. Something triggers it. There is a decision, which is either a clear rule or a judgment call. An action follows. A human reviews only when the cost of being wrong is high. It alerts on failure, and it records what happened so it can be audited later.

Trigger (something happens) → Decision (clear rule, or real judgment?) → Action (the machine does it) → Human review (only if a mistake is costly) → Alert (on failure) → Record (so it is auditable).

What I would build

Start with the test: high-frequency, rule-clear, low-judgment. Pick the single most repeated task that passes it and remove it completely, including the part where someone checks that it worked. Good first candidates are data moving between tools, the same follow-up after the same trigger, routing and tagging incoming requests, and chasing a status someone has to remember to check.

What can break

Automating a messy process, so fix the steps by hand first. Automating judgment work that quietly needs a person, like pricing a non-standard deal or handling an upset customer. Bad input data that the automation faithfully copies everywhere at speed. And no alert, so a silent failure runs for a week before anyone notices the numbers are wrong.

What the business gets

Hours back every week, fewer dropped balls, and faster response because a machine never forgets a step. You also get visibility: a clear view of what is actually happening, instead of a process that only lives in one person's head.

Automate the work that is boring and certain. Keep the work that is rare and consequential. Most teams do the opposite and wonder why it all feels fragile.

Bring me the workflow that eats your week. I'll tell you what I'd automate first, and what I'd leave human.