I watched a clip from Alex Hormozi this morning. He asked a room who uses AI every day, and a lot of hands went up. Then he asked who was making significantly more money because of it. Most of those people weren't.
His explanation was simple. AI multiplies whatever effort you put into it, and multiplying effort on something that doesn't matter still gets you nothing. If you're still making bad calls, AI won't save you. And deciding not to do something at all can be worth more than automating it.
That matches what I see from the other side. People hire OpSpring to build AI into their business, so I spend a lot of my time looking at what owners want automated. If AI isn't making you more money, the usual reason is that it's pointed at work that was never costing you money in the first place.
The bottleneck is almost never where you think
Most of my job is figuring out what's actually worth building, because the bottleneck is almost never where clients tell you it is. That isn't a knock on anyone. When you're inside a process every day, you notice the tasks that are annoying, and annoying isn't the same as expensive.
Writing emails is annoying. So is summarizing meetings. Those are also the first things most people point AI at, because they're visible and the tools make them easy. Saving time on an email you were going to send anyway feels productive. It rarely shows up in revenue.
The expensive problems tend to be quieter. A quote goes out and nobody follows up on it. Someone spends part of every day typing the same data into a second system. People outside tech often don't know what AI can actually do, so they aim it at what they can see instead of what's costing them. I think a lot of projects are misdiagnosed, not badly managed.
Some of it doesn't need AI at all
We built a pre-assessment system for MindCare Health, an ADHD assessment practice. They wanted their intake automated: several long forms with symptom scales that patients fill out before an appointment. It came to us as an AI project.
When I went through the whole workflow instead of just the task we were handed, about 80% of the report turned out to be scoring standardized tests. Those tests have published scoring rules. That's arithmetic, and plain code does arithmetic better than a model does.
We used the model only for the part that needed judgment, which was turning the scores and the patient's answers into a summary a provider could read before the appointment. Everything else runs in code, gets the same answer every time, and can be checked by anyone. If we'd sent all of it through a model because it was "an AI project," it would have been harder to check and occasionally wrong in a way that looks right. In a clinical setting that's the scariest kind of wrong.
Sometimes the answer is to buy something, or stop
Not every problem needs something built. Before quoting a project this year, I pointed the client to an existing twelve dollar app that might cover a good part of what they wanted. I'd rather tell someone to buy something for twelve dollars than build them something they didn't need.
And sometimes the right move is the one Hormozi described: stop doing the thing. If a report gets made every week and nobody acts on it, automating it just means nobody acts on it faster.
How to tell if your AI is pointed at the right thing
You can check this yourself in an afternoon.
Write down what you used AI for last week. For each one, ask what would happen to revenue or costs if that task got done twice as fast. Be honest about it. I'd guess most of the list comes back as nothing, because those tasks were never what was holding the business back.
Then go the other direction. Write down where money actually leaks, like quotes that never get a follow-up or hours someone spends copying data between systems. For each one, look for the step that's slow and the same every time. That's usually where automation pays, and a lot of the time the fix is plain code or software you already own, not a model.
Last, look for anything you'd be better off dropping: a meeting, a report, a step in a process that exists because it always has. Those just need someone to make the call.
If you want a second set of eyes on where your time and money actually go, that's what our AI audit is for.