The owner had forty-seven AI demos saved on his phone. Screenshots, clipped videos, links he had sent to himself at eleven at night. He started collecting them in January. By July he was at forty-seven.
In those same six months, not one thing about how his business actually ran had changed.
He raised his hand in a room full of other owners and asked the question he had been asking since January.
“What can AI do?”
Fair question. It is also the exact reason he has forty-seven demos and zero systems.
The question has a thousand right answers
Here’s the thing about “what can AI do.” It has a giant answer. It can do a staggering amount, and every answer you get back is true, and not one of them is about your business.
Ask a question with a thousand right answers and you get a thousand right answers. That feels like learning. Watching a good demo genuinely feels like progress.
Right? It isn’t.
It is shopping. You are standing in the tool aisle with a cart, and you have not yet decided what you are cooking.
And the reason this habit is so sticky is not that owners are lazy about it. It is that tools are easy to compare and workflows are not. A tool has a price, a logo, a feature list, a free trial. You can line three of them up on a Tuesday afternoon and feel like you did work. The dispatcher workflow that eats twelve hours a week in your business has no logo. Nobody sends you a link about it at eleven at night.
So the comparable thing wins. Not the important thing.
The tool is a station
Here’s the thing that reorders the whole problem.
The tool is a station. The workflow is the shape the station has to serve.
Think about a kitchen for a second. Nobody walks into an empty room and says “what can a combi oven do?” That is not how a kitchen gets built. You start with the dish you are putting on the pass, then you work backward to the stations you need, then you buy the equipment that fits the station. The oven is real and it matters. It is also the last decision, not the first one.
Every owner already knows this in the physical world. You did not buy the second van and then go looking for routes. You had routes. The routes were painful. The van served the routes.
Somehow AI arrived and everyone forgot the order.
Yeah.
The better question
Swap one question and the whole conversation moves off the tool and onto the business.
Not “what can AI do?”
“Which workflow in my business needs to get twice as good?”
That question has maybe five answers, not a thousand. You can hold five in your head. You can rank five. And the answers are things you already know, which is the part owners are always surprised by. You do not need research to answer it. You need ten quiet minutes.
Watch what happens to the same five conversations when the question changes.
Customer emails. The tool question is “which chatbot?” The workflow question is which replies do we send over and over, and what does it cost us when one goes out wrong.
Renewal follow-up at an insurance agency. The tool question is “can AI write the emails?” The workflow question is that follow-up sequence is already sixty percent templated and nobody has ever written it down.
Front-desk intake at a medical practice. The tool question is “is there an AI scheduler?” The workflow question is the bottleneck the front desk complains about every single week and the owner has literally never sat and watched it happen.
Proposals at a professional-services shop. The tool question is “which writing tool?” The workflow question is every proposal starts from the last three proposals, and no one has ever named the standard for a finished one.
Customer support at a retail operation. The tool question is “which helpdesk has AI now?” The workflow question is the manager runs the whole response pattern from memory, and when she is out, it stops.
Same five businesses. Same five weeks. One question keeps you in the demo. The other one puts you in front of the actual work.
Does that make sense?
Four columns before you go shopping
Once you have named the work, there is a short scoring pass that tells you which one to hand over first. We call it the Prep List. In plain terms, it is the four columns you score a workflow on before you go shopping for a tool.
How often does it happen. How much time or money it burns. Whether you can describe what a good result looks like. Whether a bad result can be caught and fixed before it costs you.
Score each one out of five. Add them up. The highest total is your first candidate.
Notice what is being scored. Not the app. The work. You are not asking whether a tool is good. You are asking whether a piece of your business is shaped in a way that can be handed off at all. Some work is not, and that is a real answer. Work that happens twice a year, that you cannot describe, and that nobody can check, should not be your first AI project no matter how good the demo was.
The column that catches the most owners is the third one. Describing what a good result looks like. If you cannot say what done looks like in one sentence, you do not have a workflow yet. You have a vibe. And a tool pointed at a vibe produces output nobody trusts and nobody uses.
Then, and only then, the tool
Name the workflow. Score it. Now go find the tool.
Now the tool conversation is short, which is the tell that you did the first part right. You are not comparing everything against everything. You are asking one narrow question: does this thing serve that station. Most options fall away in about four minutes, because most of them were never built for the shape you just described.
That is the whole reorder. The demo does not get smarter. Your question does.
A quick note on rooms, since this comes up. Getting your team in a room to help name the workflow is a good use of a room. Just remember what a room is for. It can surface which workflows hurt. It cannot pick the one you commit to. That is a separate problem, and I wrote about it in The Room Can Participate, But It Cannot Drive.
The Monday Move
Write down three workflows in your business that are working right now and could be twice as good.
Not the broken ones. That is the instinct and it is wrong. Broken work is unpredictable, and unpredictable work is the hardest thing to hand to anyone, human or otherwise. You want work that is already producing. Work that runs, that repeats, that somebody owns today.
Score each of the three out of five on the four columns. How often. How costly. How definable. How recoverable. Add them up. Highest total wins.
That workflow is your first AI candidate. Only now do you open a browser tab and look at tools.
The guardrail: if you cannot name the workflow in one sentence with a real number attached, do not go tool shopping yet. Not “our customer service is slow.” That is a feeling. “We answer about ninety support emails a week and roughly forty of them are the same six questions.” That is a workflow. The number is what makes it one.
The check: one workflow named, one candidate picked, before Friday.
Next step after that: run the workflow for one week with no AI change at all and write down what it actually costs you. That is your baseline. Then run one AI candidate against it for one week and compare. Without the baseline week you will never know whether the thing worked, and you will end up back on your phone at eleven at night, saving demo number forty-eight.
The close
You are not behind because you have not found the right tool. You are behind because you have been asking a question that cannot be answered in your favor.
Change the question. The tool is a station. The workflow is the shape it has to serve.
Original framework. Distilled from client work.
~ source material · Original framework. Distilled from client work.
