The highest-leverage use of AI for an operator is not writing content or drafting outreach. It is taking a business apart on a screen, to see how it actually makes money, before you put real capital behind the bet.
For a long time I used conversational AI the way most people first do. Draft this, rewrite that, summarize the other thing. Useful, but it was helping me produce faster, not decide better. The shift happened the first time I stopped asking it to write something and started asking it to take a business apart with me.
I had a deal in front of me. Real money, real downside, the kind of thing you talk yourself into because the story is good. Instead of building the spreadsheet I did not want to build, I put the whole thing into a working session and asked it to be the skeptic. Rebuild the model. Question the revenue line. Show me where this only works if three optimistic things all happen at once.
Twenty minutes later I had a cleaner picture of that business than I had after two conversations with people trying to sell it to me. The bet did not survive its own math. That is the point. It was cheaper to kill it on the screen than in the market.
Every operator knows they should model a bet before they make it. Most do not, and the reason is honest: it felt like a chore. Building a proforma from scratch is slow, it surfaces numbers you would rather not look at, and by the time it is done the excitement that started the whole thing has cooled. So we substitute a gut read for the work, and we call it conviction.
What changed is the cost of the chore. The modeling that used to take an afternoon and a spreadsheet you half trusted now takes a conversation. When the price of checking your assumptions drops far enough, there is no excuse left for not checking them. The bottleneck was never the math. It was the friction, and the friction is mostly gone.
The goal is not a prettier forecast. It is understanding the machine. You are trying to see, in plain terms, how a dollar enters this business, what it has to pass through, and how much of it survives to the bottom. A few lenses do most of the work.
Take the proforma apart, do not accept it. Any franchise or acquisition packet comes with a model built to make the sale. Rebuild it from the unit up. What has to be true for these numbers to hold, and how many of those things are inside your control versus the market's.
Separate what you own from what you rent. On anything financed, the headline payment hides the split between principal and interest. Early on you are mostly renting the money, not buying the asset. Knowing that ratio changes how a deal looks in year one versus year five.
Normalize the competition. Raw competitor counts lie. Ten competitors in a metro of two million is a different world than three in a town of forty thousand. Density per population tells you whether you are entering a crowded room or an open one.
Picture a model that projects a healthy margin in year one. You do not argue with the conclusion. You ask it three things.
1. What ramp does this assume? Most models quietly treat month one like month twelve. Push the revenue curve out to a realistic ramp and watch year one turn.
2. Whose labor is free here? Owner hours are often costed at zero. Put a real wage on them and see if the business is profitable or if it is just buying you a below-market job.
3. What breaks at the trough, not the average? Averages hide the month you cannot make payroll. Model the low season, not the annual mean.
The numbers here are illustrative, not a claim about any real business. The value is the interrogation, not the figures.
None of these questions are clever. They are the boring ones that a motivated buyer skips because the answers are inconvenient. The work is having the discipline to ask them every time, and AI makes asking them cost almost nothing.
It would be a mistake to read any of this as handing the decision to a machine. A model is only as honest as the inputs, and the inputs are judgment. What ramp is realistic in this category. What the owner's time is worth. Whether that market is genuinely open or just looks open from the outside. Those are your calls, and they are where bets are won or lost.
The role of the tool is narrow and powerful. It removes the excuse. It turns a day of dreaded spreadsheet work into an hour of hard questions, so the only thing standing between you and a clear-eyed view of the bet is your willingness to look. This is the Intelligence layer of a revenue engine doing its actual job: taking raw inputs and turning them into a decision you can defend.
Before capital goes anywhere, run the bet through five questions. If you cannot answer them, you are not ready to write the check.
Answer those with real inputs and most bad bets disqualify themselves. The good ones get stronger, because now you can see exactly why they work.