How I think about AI in a real company

1. The cost-cutting story is true. It’s just not the interesting part.

My own first pitch led with a line a contractor gave me over lunch: “I hate paying someone a hundred grand to do a $20-an-hour task.” It’s a great line. The math behind it is real, and I’ll keep doing that math for as long as owners keep paying six-figure salaries for work a machine can carry.

But run the math all the way out. Say you run an $8M business at a 25% net margin, and a sweeping AI effort lifts you to 40%, which is the rosiest case anyone would dare put on a slide. You just found $1.2M a year. That’s real money and you should take it. Now try to do it again next year. You can’t. Cost-cutting runs until it hits a wall, and the wall is the business itself: same customers, same offer, same model. There are only so many costs to take out of a company. Past a point you’re not improving the business anymore, you’re just wringing it tighter.

2. The contractor already knew it.

The contractor who gave me the $20-an-hour line said something else in the same conversation, about what AI could actually do in his company: “The payroll one isn’t even the bigger one. That’s 10% of what it could add.” He bids one job a week and watches eight more cross his desk unbid. “If I can estimate 10x more jobs, that could be a 7-figure value... I really think I’m missing out on a million dollars.”

Nobody in those sentences is trying to save money.

3. Every AI project in an established company sits on a graph.

Ask two questions about any AI project: how hard is it, and how much does it change what the business can do? Plot them and you get the Leverage Graph, with three zones that matter.

Personal leverage (bottom-left): the owner gets AI working in their own day. This part is easy, and it genuinely changes that person’s week. It doesn’t change the company.
Margin moves (the middle): AI does what the business already does, cheaper and faster and with fewer errors. The $20-an-hour math lives here. It changes how the company runs, and the dollars are real and provable.
Capacity moves (top-right): AI makes the business able to do something it couldn’t do before. Bid eight jobs a week instead of one. Settle cases faster where every saved week is pure margin. Get in front of buyers you couldn’t physically reach. Once in a while, open a line of business that didn’t exist last year.

The Leverage Graph. AI projects plotted on how hard they are against how much they change what the business can do, with three zones: personal leverage at the bottom left, margin moves in the middle, capacity moves at the top right.

4. Margin moves are capped by your own cost structure. Capacity moves are capped by the market.

That asymmetry is the point of the whole graph. The best margin year you will ever have is bounded by the business you already built. You can’t save your way past your own cost structure, and the returns go thin long before you get near it. A capacity move’s ceiling isn’t inside your four walls at all. It’s whatever the market will give you. Take the same $8M business: optimized hard, it hands you $1.2M more. Grown to $20M at the same old margin, it hands you $3M more, and nothing about next year says stop.

The same $8M business under two ceilings. A margin move adds $1.2M once, capped inside the company’s own cost structure. A capacity move adds $3M every year, capped only by the market.

There’s a second asymmetry, and it’s coming whether anyone likes it or not. Efficiency is an edge that expires. Any tool you can buy off the shelf, your competitor orders next quarter, and the savings get competed away into prices. The work you won while they were still optimizing stays won. The jobs, the clients, the reputation, the data: those you keep.

5. The top right is not reinvention. It’s a lot more of what you already do.

When a consultant says AI will turn your company into something else, reach for your wallet, because someone else already is. Established companies don’t become AI companies, and they don’t need to. For a real business the top right almost always looks like more of the thing you’re already good at: more volume through the same door, more speed where speed is money, more reach for the same offer. Once in a while, when a client sees it themselves, a genuinely new line. It sounds less ambitious until you price it.

6. Ceiling numbers are scenarios. Floor numbers are arithmetic. I won’t pretend otherwise.

The payroll math is checkable. The salary exists, the hours exist, the multiplication is honest. A growth number is a scenario: it depends on win rates, demand, crews, working capital, and your own follow-through. So I hold myself to two rules. The ceiling is always your number, built from your bids, your intake, and your turned-away work, stated together with whatever constraint binds next. If you can’t audit it, I won’t say it. And I guarantee floors, never ceilings. You should fire anyone who does it the other way around. The map’s dollar guarantee counts conservative, checkable dollars only; the ceiling is the reason to do the work, not the guaranteed part.

Under both rules sits one more honesty line. AI can capture demand you can already see being missed. It does not manufacture customers. Selling that is the agency business, and you’ve already paid that tuition once.

The floor is arithmetic: the salary exists, the hours exist, and it is guaranteed. The ceiling is a scenario that depends on win rates, demand, crews and working capital, and it is not guaranteed. The gap between them is the reason to do the work, not the guaranteed part.

7. You earn the top right.

The horizontal axis of the graph is real. The top right is harder, and nobody starts there by accident. The climb has an order. Fluency comes first, because an owner who has personally felt what AI does is the one who recognizes the capacity move when it shows up in his own shop. That’s what personal leverage is for. It’s the on-ramp, not the destination. Margin moves come next. They’re fast, they’re provable, they train your team on real work, and the savings usually pay for the climb. Then the capacity move, built on a team that has done this before and numbers everyone trusts. Sometimes the map says skip the line and go straight at it; the contractor’s bid engine was that kind of rock. But when the sequence holds, it isn’t stalling. It’s how a fifty-person company reaches the top right without betting the company.

The climb in order. Fluency is the on-ramp, not the destination. Margin moves come next, because they are fast, provable, and pay for the climb. The capacity move is built on a trained team and numbers everyone trusts. A dashed arc shows the case where the map says skip the line.

8. You’ll know it worked when your problem changes into one you chose.

The owners I work with all start with the same problem: a hundred possibilities, no first five hours, and a nagging feeling the meter is running. Done right, that problem gets replaced. For the contractor, the problem I’m building toward is winning more work than he can staff. That’s a hard, expensive, genuinely enviable problem, and solving it becomes the next project. For the attorney who told me he wanted to “make the same amount of money, do 30% less work, and not functionally increase my risk profile,” the destination is exactly what he asked for: the floor he described, protected, with his risk profile intact. Either one counts as winning. Pick yours on purpose, with your own numbers on the table, instead of defaulting into whichever one the person selling to you happened to stock.

That last distinction is the whole practice. The floor (same revenue, less work, no new risk) is real, and I protect it on every engagement. But nobody lies awake at night excited about the floor.

The savings pay for it. The growth is the point.

James Green, Dominion AI

Start with the Audit