Start with what wakes you up at 3am

Ask an owner what AI could do for their business and you’ll get a shrug. Ask them what they lose sleep over and you’ll get a list with dates on it. That list is the project plan.

The easiest AI wins are specific

Nobody lies awake worrying about “digital transformation.” You lie awake because two guests once showed up to the same condo, and you still don’t fully trust the calendars. Because a five-figure payment went missing for three weeks and you found it by accident. Because the calendar sync broke once and nobody noticed for four days, and you don’t actually know it’s working right now.

Notice the shape of those fears. Every one is specific. Every one is repeatable — it’s not a freak event, it’s a category of event. And every one has a checkable condition: do any reservations overlap? did every invoice get a matching payment? did the sync run in the last hour?

Specific, repeatable, checkable. That is precisely — precisely — what software with AI on top is best at. Not strategy. Not vision. Checking things. Tirelessly, every few minutes, forever, without getting bored or going on vacation.

What this looks like in practice

In my own operation, the double-booking fear became a watchdog: every calendar, every property, checked daily for overlaps, with a second watchdog making sure the first one’s data is fresh — because a monitor reading stale data is worse than no monitor at all. Conflicts now get caught while they’re still an email, not a guest in a doorway.

The missing-payment fear became a reconciliation engine: every payment matched to an invoice, every unmatched dollar surfaced with a reason. The “is it even running?” fear became a heartbeat: every automated system reports in, and silence raises an alarm.

None of these took months. Watchdogs are days of work. That’s the absurd economics of the 3am list: the emotional value is enormous and the engineering cost is small. There is no better ratio anywhere in AI.

The one requirement

You have to know what you’re looking for. That’s the catch, and it’s why this works for me and fails for the grand-vision crowd. The fears are only easy to solve if you can translate them into checkable conditions, and you can only do that if you understand the operation deeply — ideally because you’ve run one yourself.

MIT’s researchers found the same thing from the other direction. Their 2025 State of AI in Business report concluded that 95% of enterprise AI pilots deliver no measurable return — not because the models are weak, but because the tools never learn the actual workflow. Specific beats general. It’s true for fears, and it’s true for the software that watches them.

So before anyone sells you an AI strategy: write the 3am list. The things that went wrong. The things that almost went wrong. The things you check manually “just in case.” Each line is a small, cheap, high-relief project — and collectively they buy you something no strategy deck ever will: sleeping with both eyes closed.

This is part two of a series on applying AI to a real business. Part three covers the other kind of low-hanging fruit: the time-tax.