A VP of operations I worked with kept a spreadsheet of every process her team had handed to software. Ticket routing, inventory reorder points, which invoices got flagged for a second look, the order the delivery vans ran their stops. Forty-some processes, all running on models, most of them humming along without anyone watching closely. She was proud of it, and she had earned the right to be.
Then, in the same quarter, her company debated whether to raise prices on their largest product line. That decision did not go on the spreadsheet. It went into a room. Eleven people, a full afternoon, printed decks, a whiteboard, and at the end of it a call made mostly on the gut of the two most senior people there. When I asked her later why the pricing question never touched any of the tools her team trusted forty times a day, she looked at me the way you’d look at someone who asked why you don’t let the intern fly the plane.
That reaction is the whole problem, and almost everyone has it.
The instinct is not stupid, so give it its due. When a decision is big, the argument goes, it carries nuance a model has never seen. Relationships. Timing. A read on how a competitor will feel, a board’s mood, a founder’s appetite for risk. Those are human things. A model trained on what already happened cannot know the one detail that makes this moment different from every moment before it. So you protect the important call. You keep it in the room, in human hands, where the judgment lives. It feels like the responsible thing. It feels like exactly what a serious leader is supposed to do.
Look at how the sorting actually happens, though. The decisions we automate without a second thought are the small ones, the frequent ones, the ones we make and remake every day. The decisions we shield are the big ones, the rare ones, the ones that come around once a year if that. We have quietly decided that frequency earns trust and rarity earns suspicion. We built our confidence on how often a decision repeats, when we should have built it on what it costs to get wrong.
Years ago Jeff Bezos put a plain line under this. Some decisions are two-way doors and some are one-way doors. Walk through a two-way door, look around, and if you don’t like the room you walk back out and try another. Those are the reversible ones, and he argued you should make them fast and cheap and push them down to people closer to the work, because a wrong one costs little and does not last. The one-way doors are different. Once you are through, the door locks behind you. Those earn the slow, careful, senior attention.
He was mostly talking about who should decide and how quickly. The same split tells you something he was not arguing but that falls right out of it: where the help belongs.
Take one of her reversible doors. The delivery routes. Every morning the system sequences a few hundred stops across a dozen vans, weighing traffic, package weight, time windows. Say it gets Tuesday wrong. A van doubles back, a driver loses forty minutes, a handful of packages land an hour late. It stings a little. By Wednesday the model has the day’s data and reroutes, and nobody remembers Tuesday by the weekend. The wrongness has a short half-life. You can be wrong here all week and the company barely feels it.
Now her one-way door. The pricing move. Say they raise prices eight percent on that product line. The number is not the cost. The cost is what the market learns about you the moment you move. The biggest account reopens its whole contract. A competitor you had been ignoring reprices to sit just under you. Two reps quit because the calls got harder. Six months later, even if you roll the increase back, you cannot roll back what the market learned. The door is shut. You are living in the room you walked into.
Line those two up and the sorting we do looks strange. She poured decision support into the van routes, the place where being wrong costs a driver forty minutes and evaporates by Friday. And she walked into the pricing room with a whiteboard and eleven guts, at the one door in the building that locks behind you.
Said plainly, the logic is hard to argue with. If a decision is small and reversible, a wrong call costs a little and you fix it fast. That makes it the safest place in the whole company to be wrong, which makes it the safest place to lean on a tool. If a decision is large and hard to take back, then nudging it even slightly toward right is worth an enormous amount, because you will live inside the outcome for years. That makes it the most valuable place to bring better information, not the place to bring the least.
I sorted this wrong myself, and not long ago. A client was choosing the core platform to build the next five years of their product on. A one-way door if there ever was one. Migrating off it later would have cost more than the company was worth at the time. I did what felt responsible. I closed my laptop, cleared the tooling and the models off the table, and told the founder this was a judgment call, that we should think it through as people. Two floors down, that same client had a model scoring inbound leads all day, tuning which ones got a call within the hour. A decision they could revise every single afternoon.
Months later it landed on me what I had done. I had given the fully instrumented, second-guessed, data-rich treatment to the choice we could undo by Friday. And I had walked the irreversible one, the eight-figure one, into a quiet room and asked everyone to trust their feelings. I had the whole thing backwards, and I am supposed to know better.
Here is the part I did not see coming until I watched it play out across a few companies. When you keep every structured input away from the big decisions, you do not get more human judgment. You get less examined judgment. The room fills with the loudest voice, the freshest anecdote, whatever the senior person happened to read that week. The absence of a tool does not summon wisdom into the empty space. It just removes the one thing that might have made someone slow down and check the story they were about to bet the company on.
None of this means you hand the pricing call to a machine and read the output like a verdict. The leader still decides. The leader should decide. But the leader making the eight-million-dollar call that cannot be walked back deserves at least the support we already lavish on the analyst making the eighty-thousand-dollar one we can undo by Friday. The gut still gets the last word. It just should not be the only word in the most expensive room in the building.
The reflex to keep AI away from the big decisions feels like prudence. Look closely at where it points, and it is the one place the prudence is misfiring.
At Kennesys, most of the fractional CIO/CTO work comes down to this: building the strategic layer that puts the best information in front of the doors that lock behind you.