I was working on a project this week that started with a pretty simple question: Where should this company find its next customers?

We had a list of current and past customers. We knew the general type of business that bought from them. We knew there were probably more companies like those customers in markets they had not really pursued.

This seemed like exactly the kind of project AI should be good at.

I could have uploaded the customer list and asked ChatGPT to analyze it, research the competition, find promising markets, and give me 100 companies to call.

And it probably would have.

That was exactly what I didn't want it to do.

There is a temptation with AI to see how much we can get it to accomplish with one prompt.

Analyze this. Research that. Develop a strategy. Find prospects. Write the emails. Give me the finished plan.

It feels efficient because we are getting a lot of output quickly.

But there is a problem. Every step depends on the step before it being right.

In this project, I wanted AI to start by looking at the company's existing customers and identifying patterns. What types of companies bought from them? Were there geographic patterns? Did certain industries show up more often? Were their customers concentrated around metro areas or Main Street manufacturing communities?

That first analysis would determine what we researched next.

If AI misunderstood the customer base, everything downstream could still look impressive. We could get a beautifully organized competitor analysis and a spreadsheet full of prospects.

They would just be the wrong prospects.

So instead of asking AI to finish the project, I asked it to work through the project with me.

I broke the project into stages.

First, analyze the customer list.

Then stop.

I reviewed what it found against what I already knew about the business. Some patterns were obvious. Others were interesting enough to explore further.

Once we agreed on what a good customer actually looked like, we moved to competitors.

Then we stopped again.

We looked at who else was serving that market, how they positioned themselves, where they appeared to be selling, and whether that changed our thinking about the opportunity.

Only after we had worked through those pieces did we start talking about where to look for new prospects.

This took a little more interaction than typing one giant prompt and waiting for the answer.

It also gave me a lot more confidence in the result.

And that is an important distinction in how we use AI.

Speed isn't very valuable if you're moving quickly in the wrong direction.

We've talked a lot here about using AI as a thinking partner instead of simply a writing tool. This is what that looks like in practice.

You don't have to know the answer before you start. You don't even have to know exactly where the project is going.

But you should know where you want to stop and check the work.

Think about how you would handle a project with another person. You probably wouldn't hand someone a complicated assignment involving customer research, competitive analysis, market selection, and prospecting and tell them to come back when the whole thing was finished.

You would talk along the way.

You would look at the first round of research and say, "Yes, that's what I'm seeing too," or, "No, you're missing something important."

Then you would move forward.

AI should work the same way.

The checkpoints are not slowing the process down. They are what keep the process pointed in the right direction.

The biggest difference was not some magic wording inside the prompt. It was the checkpoints.

Instead of asking AI to analyze my customers, research competitors, identify new markets, and build a prospect list all in one shot, I made it stop after every step.

It had to show me what it found. I could react, correct the assumptions, and then decide whether we were ready to keep going.

That matters more than it sounds.

If its read on my best customers had been wrong in Step 1, I would rather catch that immediately than let it spend the next ten minutes researching competitors and generating leads based on a bad assumption.

This is where AI becomes much more useful as a thinking partner. You are not just giving it a task. You are building a process that gives you places to steer.

Here is the actual prompt, word for word, so you can adapt it to your own business:

Swap in your own business description and customer list. The exact industry does not matter nearly as much as the structure.

You can use the same approach for customer research, competitive research, a new market, pricing, messaging, or almost any project where a bad assumption early in the process could send everything that follows in the wrong direction.

The goal is not to write one giant prompt.

It is to build a better conversation.

There is a bigger business lesson hiding in this than just how to use ChatGPT.

A good process makes it easier to see when something is going wrong.

That is true in marketing. It is true in sales. It is true when you are looking at your numbers. And it is definitely true when you are using AI.

If the only thing you ever look at is the finished product, you have no idea where the thinking went sideways.

Build checkpoints into the work, and suddenly AI becomes less of a black box that spits out answers and more of a tool you can actually manage.

That's a much more useful way to work with it.

Reply and tell me what project you would want to run this way. I'm happy to help you think through the checkpoints.