Owners pick their first automation project the way they pick a vacation: by what sounds most impressive to describe.
The AI receptionist. The predictive lead scoring. The thing that will "transform how we operate." Big, visible, strategic-sounding, and — this is the part that gets missed — almost always the wrong place to start.
Here's a better selection criterion, and it will feel too simple: build the thing your best employee hates most.
Not the biggest number. Not the most strategic. The one that makes a person you cannot afford to lose sigh audibly when it lands on their desk.
Why morale is a better filter than ambition
Three reasons, and each one is about how first projects actually succeed or fail.
You'll get an honest process map. Automation projects live or die on whether you correctly captured what actually happens, including the eleven exceptions nobody documented. Getting that out of a person requires them to want you to succeed. Someone who hates the task will tell you everything — every workaround, every edge case, every reason the official process is a lie — because they desperately want it gone. Someone who's ambivalent gives you the sanitized version, and you'll discover the truth in production.
You'll get adoption for free. The hardest part of any internal system is the part where people use it. If you build the thing everyone dreads, nobody has to be sold on it. The office manager who's been assembling the same packet every Wednesday for three years does not need a change-management workshop. She needs the packet to stop.
You'll bank credibility for the hard one. Your first project is not really about the first project. It's about whether the next four get funded and believed in. Ship something that visibly makes a beloved employee's life better, and you'll have the whole company's cooperation for the next thing. Ship an impressive-sounding failure and you'll spend a year hearing "is this going to be like the chatbot?"
The thirty-person firm test
Picture an operations manager at a thirty-person firm. She's been there six years. She's the one who knows everything, who trains everyone, who catches the mistake before it goes out the door.
Ask her one question, privately, and mean it: what part of your week would you pay money to never do again?
You'll get a real answer inside ten seconds, because she's been carrying it around. It won't be glamorous. It'll be something like: reconciling the two systems that don't talk, or building the Wednesday report, or chasing signatures.
Now price it — hours, loaded rate, fifty-two weeks. Nine times out of ten it's a number between $15,000 and $50,000, which is exactly the range where a well-scoped automation clears a 2x return in year one without any heroics.
You have just done, in one conversation, the entire diligence process that most companies spend a quarter on.
The objection, and the answer
The objection is: that's not strategic. That's just a chore. I want AI to do something transformative.
Here's the thing about transformation. It doesn't arrive as a project. It arrives as an accumulation.
The firms that have genuinely changed what they're capable of over the last two years did not do it by building one visionary system. They did it by removing eleven chores, one at a time, over eighteen months — and then waking up to find that their operations team had thirty percent more capacity, that their people were happier, that nobody had quit, and that they could take on a class of work they previously had to turn down.
That's transformation. It just looks like maintenance while it's happening.
The visionary project, meanwhile, tends to arrive at a company that hasn't built the muscle yet — no clean data, no integration experience, no track record, no internal belief — and it collapses under its own weight, and everyone learns the wrong lesson.
What to do this week
Have three conversations. Your operations manager, your most senior producer, and whoever handles the money. Same question to each: what would you pay to never do again?
Write down what they say. Price all three. Pick the one with the best ratio of annual cost to build difficulty — and if two are close, pick the one belonging to the person you'd be most devastated to lose.
Then tell them you're fixing it. Watch what that does to the room.
There's a version of AI adoption that starts at the top, in a strategy document, and never quite reaches the floor. And there's a version that starts at somebody's desk, with something they hate, and works its way up — building trust, capability, and credibility as it goes.
Only one of those actually ends up transforming anything. It's the unglamorous one, and it starts on Wednesday, with the report nobody wants to build.