What should you automate first? (Not the fun thing)

When a business decides it’s time to automate something, the first candidate is almost always wrong. It’s usually the flashy thing (the AI chatbot on the website, the customer-facing whatever) because that’s what automation looks like in ads.

The right first candidate is boring. It’s the thing someone does every single morning while mildly resenting it. Here’s how to find yours.

Score it: frequency × tedium × rules

Three questions about any task:

How often? A daily task saves roughly twenty times more than a monthly one. Frequency is where the payback hides: automating the annual report is a hobby, automating the daily one is a raise.

How tedious? Copy-paste between systems, re-typing what’s already in an email, assembling the same status update from the same three places. If the person doing it describes it with a sigh, score it high. (If it lives in a spreadsheet, you may be holding one of these.)

How clear are the rules? This is the filter that kills bad projects. “Copy the order number into the tracker” has perfect rules, so automate it. “Decide if the customer sounds upset” used to be impossible, and modern AI handles judgment-adjacent work surprisingly well now. Still, start with tasks where a human quickly checks the output, not ones where nobody looks.

Multiply, roughly, in your head. The winner is almost never the chatbot. It’s the invoice data entry, the quote follow-up email, the Monday report.

The shortlist that shows up everywhere

The same first projects show up in most small businesses: getting data out of emails and PDFs into whatever tracks the work; drafting the routine messages someone writes fifteen times a week; the weekly summary assembled by hand from three systems; and the double entry between two tools that don’t talk. That last one is sometimes an automation, sometimes a small custom connector, and the line between those is blurrier than either label suggests.

Why not the customer-facing thing first

Because your first project is also your learning project. When an internal automation makes a weird mistake, an employee catches it and you adjust. When the public-facing one does, a customer catches it. Earn confidence internally first; the chatbot can wait until the boring wins have paid for it.

And be suspicious of anyone who skips this reasoning to sell you the flashy thing. The same instinct I flagged in AI won’t build your software for you applies to buying automation, not just building it.

Start smaller than feels ambitious

Pick the winner from your scoring, automate that one thing, and measure the week before against the week after. A first project that saves forty minutes a day, every day, beats a moonshot that demos well and dies in a drawer.

Finding these is most of what the AI consulting side of my work is. If you want a second opinion on your shortlist, send it over.