Four things to do before you spend a dollar on "AI."

Somewhere in your business right now there's a tool you bought in a panic. Signed up, paid for a year, used it twice. Now it's a line item nobody wants to admit to.

A lot of those are about to be AI tools, if they aren't already.

I build AI systems. Actually build them, with my own hands, not slides about them. So when I tell you to slow down before you spend, it's not because I think the technology is overhyped. It's because I've watched too many owners buy the demo and skip the work that makes the demo pay off.

Here's the thing nobody selling you software will say out loud. The tool is almost never the hard part anymore. Access to AI is cheap and getting cheaper. What's scarce is knowing what to point it at. That's judgment, and judgment is the part you can't buy in a subscription.

So before you spend a dollar, do these four things. None of them cost anything.

1. Find the repetitive, rules-based work first

AI is not magic. It's leverage. And leverage only pays where there's something repetitive to lean on.

So go find it. Walk your week and look for the work that happens over and over, runs on clear inputs, and produces a predictable output. Same thing in, same kind of thing out, again and again. That's where AI actually earns its keep.

The stuff that needs real judgment, reads a person, or handles the case nobody wrote a rule for, leave that alone for now. That's still human work.

If you've ever sorted your work into "a person should do this" versus "a system could do this," you already know the move. AI doesn't change the sort. It just makes the system side a lot cheaper to build than it used to be.

2. Make it measurable before you automate it

Here's the mistake that burns the most money. Automating something you can't measure.

If you don't know how long a task takes today, how often it goes wrong, or what it costs you, then you have no way to know whether the tool helped. You'll feel busy and modern and have no idea if you came out ahead.

So before you automate it, put a number on it. How many times a week. How long each time. How often it has to get redone. Now you've got a baseline. Now, when the tool's in place, you can answer the only question that matters: did this actually pay off, or did I just buy a more expensive way to do the same thing.

You can't improve what you can't see. That's true for your people, and it's twice as true for a machine you've handed work to and stopped watching.

3. Start where a mistake is cheap

Owners love to point AI at the scariest, highest-stakes problem first. The big one. The one keeping them up at night.

Wrong order.

Start where a mistake is cheap. High-frequency, low-risk. Somewhere a wrong answer costs you a few minutes and an eye-roll, not a customer or a reputation. You want a place where the thing can stumble while it's learning your business and the only casualty is a little time.

Get a win there. A real one you can measure. Then move up. The owners who blow themselves up with AI almost always started at the top of the risk ladder instead of the bottom.

4. Use it yourself before you roll it out

Last one, and it's the one everybody skips.

Use the thing yourself first. Hands on the keyboard, on real work, for a couple of weeks, before you put it in front of your team.

Two reasons. First, you'll find out fast whether it's actually useful or just impressive, and those are not the same thing. Second, you can't lead your team into something you don't understand. If you hand them a tool you've never touched and tell them to figure it out, half of them will quietly decide it's not worth the trouble, and they'll be right.

Judgment comes before delegation. You earn the right to roll something out by knowing exactly what it does and where it breaks. Then you teach it, then you hand it off. In that order.

How to tell real from hype

While you're doing those four things, you'll be getting pitched. Hard. Here's how to sort the real from the noise without being a technologist.

Ask the vendor what specific task it replaces and how you'll measure the difference. Watch whether they answer in your language or theirs. Real value sounds like "this cuts your follow-up time from a day to an hour, here's how you'd check." Hype sounds like "transform your operations with the power of AI." One of those you can verify. The other is a feeling with a price tag.

And ask what happens when it gets something wrong. Anyone who tells you it won't is selling you the demo, not the tool. Everything that runs on real data gets something wrong eventually. What separates a useful tool from a dangerous one is whether you find out when it does. Ask how you'd catch a bad answer before it reaches a customer. If they don't have a clean answer, you're not buying a system, you're buying a liability with a nice interface.

What to do this week

The lesson? AI pays off for the operator who did the unglamorous work first. Found the right task. Measured it. Started small. Learned it personally. Skip that and the smartest tool on the market becomes another line item you don't want to talk about.

This week, do one thing.

Name one task you do constantly that bores you to tears. The repetitive one, the one you could do in your sleep.

Then run it through the four. Is it rules-based? Can you measure it today? Is it low-risk enough to test? Would you use the tool yourself first?

If it clears all four, you just found where to start. Notice you haven't spent a dollar yet. That's the point.