Essay · AI & Capital Discipline · Part 07 of 12
I once watched a leadership team spend seven weeks choosing between two tools that cost about the same, did about the same thing, and could each be swapped out in a fortnight.
Seven weeks. Four meetings, a bake-off, a spreadsheet with weighted criteria. The switching cost they were protecting against was smaller than the cost of the deliberation.
The asymmetry nobody prices
Executive attention is the scarcest thing in a company and the only one with no budget line. Nobody submits a request for eleven hours of the leadership team. It just gets consumed, and because it's invisible it gets consumed by whatever is loudest rather than whatever matters.
A reversible decision that takes seven weeks doesn't cost you the tool. It costs you the seven weeks, applied to something with no consequence, while the decisions that actually can't be undone get made quickly because they're less fun to argue about.
The sorting question
Before a decision goes on an agenda, ask what it costs to reverse in six months.
Cheap to reverse means decide fast, at the lowest level that has the context, and move. Getting it wrong costs you a switch. Deliberating costs you the quarter.
Expensive to reverse means slow down, get more people in the room, and write down what you believed at the time so you can audit the reasoning later rather than the outcome.
The failure isn't caring too much about decisions. It's spending the same care on both kinds, which in practice means the reversible ones get more attention, because they're safer to argue about. Nobody's career is on the line in a tool bake-off.
Why this gets worse with AI spend
This category is unusually confusing right now because the surface decision and the underlying decision have different reversibility.
Which model you call is reversible. Swapping providers is a configuration change and, increasingly, an afternoon.
What you build around it often isn't. If you've redesigned a workflow, retrained a team, restructured a support function, or committed data into a shape that only makes sense for one provider, you've made an expensive-to-reverse decision while telling yourself you made a cheap one.
So the question isn't whether the vendor is swappable. It's whether the thing you built on top survives the swap. Ask that before the build, not during the migration.
What I use
Two lists. Decisions that can be undone for less than a month of work, and decisions that can't.
The first list gets delegated with a deadline and no meeting. The second list gets my actual attention, and I write a short note on what I expect to happen and why, so that when it goes wrong I can tell whether the reasoning was bad or the world moved.
This sounds like a small administrative habit. It's the single change that has bought me back the most time, because it kills the meeting that shouldn't exist before it gets scheduled.
On Monday
Look at your calendar for the past month. For each decision meeting, ask what reversing that decision would have cost.
Anything you could have undone in under a month should not have been on your calendar. Push it down and take the switching cost when you're wrong. You'll be wrong less often than the deliberation implies, and it'll cost less than the deliberation did.
The operating principle. Sort decisions by what reversal costs, not by what they cost. Cheap to undo means decide today and delegate it.
Juan Vegarra is the author of An Outsider’s Playbook. The views here are his own. More essays · Advisory · Write me