Stop Buying AI Tools. Start Buying Outcomes.

Every January, gym membership sales spike. Every February, the treadmills sit empty. The purchase felt like the change. It wasn't.

AI procurement runs the same con on smart people. A budget opens up, a mandate comes down from the board, and somebody buys 200 Copilot seats. The invoice clears. The Slack announcement goes out. And for one warm quarter, it feels like the company did something about AI.

A license is not an outcome. It's a permission slip.

The Tool-Shaped Hole

Here's the pattern, and I'll name the mechanism in the same breath: procurement theater is when the act of buying substitutes for the change the buying was supposed to cause. You feel the tool-shaped hole in your operation, you buy a tool shaped like it, and the buying itself scratches the itch. The anxiety goes down. The work doesn't move.

I've watched this in Fortune 500s and federal agencies, which are two very different places that fail in exactly the same way. Someone senior wants to be able to say the org is "doing AI." A seat count is legible. It fits in a board slide. Twelve thousand seats deployed reads like progress, and nobody in the room can argue with a big number going up.

The tell is always the same. Ask the person who signed the PO one question: what specific number is this supposed to move?

Watch them reach. If the answer is "productivity" or "efficiency" or "we'll be more competitive," you've bought a gym membership. Those aren't outcomes. They're the feeling of an outcome. Real outcomes have units. Days. Dollars. Tickets closed. Percentage points of margin. If you can't put a unit on it, you can't tell whether you got it, which means you'll renew the license next year on faith.

The Friction Is Gone. The Cost Isn't.

There's a reason this is worse with AI than with the CRM you overbought in 2015.

AI First Principle 12 says to justify resource consumption, because AI makes wasting resources trivially easy. The friction of spending is gone. You can stand up a pilot in an afternoon, expense a hundred seats before lunch, and spin up a "center of excellence" by Friday. None of it hits a wall the way a data-center build or a six-month SAP rollout used to.

But the cost is still real. Every one of those seats is a monthly line item forever. Every pilot is a slice of your best people's attention, which is the one resource you can't buy more of. Cheap to start is not the same as cheap. When the friction disappears, the discipline has to come from you, because nothing else will supply it.

So the question flips. Not "what can this tool do." Tools can do nearly anything now; capability is no longer scarce. The scarce thing is knowing which single capability, pointed at which specific problem, produces a result you can measure.

Buy the Number, Not the Seat

Rewrite the purchase. Same money, different sentence.

Before: "Buy 200 Copilot seats for the sales org."

After: "Cut quote turnaround from three days to three hours for the mid-market team by end of Q3. Maria owns the number."

Look at what the second version forces into the open:

• The outcome. Three days to three hours. A unit, a target, a direction.
• The baseline. Three days. You can't claim a win if you never wrote down where you started. Most orgs skip this, then declare victory against a number they invented after the fact.
• The scope. The mid-market team, not "sales." One team you can actually watch.
• The owner. Maria. A name, not a committee. Someone whose quarter gets worse if the number doesn't move.

Notice the second sentence never mentions Copilot. Maybe the answer is Copilot. Maybe it's a scripted workflow, a better template, and no new license at all. That's the point of AI First Principle 10: validate the problem before you commit to the solution. Iterate toward what works. You wrote down the problem in a way that lets a boring answer win if a boring answer is right, and boring answers are cheaper to run and easier to keep.

This is the whole Solve idea in one move. Deliver one working thing tied to a real outcome, not a pile of capabilities you hope somebody assembles into value later. A seat is a capability. Three-hour quotes are a solution.

The Monday Rule

Here's something you can enforce at the next approval meeting without a strategy offsite.

Don't approve an AI purchase until someone can state the outcome, the baseline, and the owner in one sentence.

If they can't, the purchase isn't ready. It's not that the tool is wrong. It's that nobody has done the thinking the tool is supposed to finish, and the seat will paper over that gap for exactly one renewal cycle before someone asks what you got for the money.

Make people say the sentence out loud. The ones who can have already done the work. The ones who can't are about to buy a treadmill.