Ask a leadership team to draw how goals move through their company and they can do it without thinking. The company sets a direction. Finance turns it into a budget. Each function turns the budget into targets. Each team turns targets into a quarter of work. Somebody drew it on a whiteboard once and now it lives in everyone’s head. One decision at the top, and if the machine is working, you get aligned movement three and four layers down.
Ask that same team to draw how intelligence moves through the company, and the marker stops.
That gap has been sitting in front of me in a lot of rooms lately, and for a while I could not name what bothered me about it. We spent the better part of twenty years learning to cascade objectives. We built rituals around it. Planning cycles, OKRs, the quarterly ritual of translating one layer’s goal into the next layer’s work. We hired people whose entire job was to run that translation and keep it honest. It is one of the few genuinely hard organizational skills that most companies have actually gotten decent at.
Then AI showed up, and we quietly threw all of it out.
Watch how AI actually enters a company. A marketer tries a writing tool because a peer mentioned it. An engineering team adopts an assistant because the developers asked. Support buys a deflection product because a vendor ran a good demo. Every one of these is a real, defensible local win. And not one of them points anywhere in particular. There is no line back to the company’s direction, because nobody drew one. The tools arrived from the bottom, scattered, in whatever order the enthusiasm happened to strike.
Now imagine running your goals that way. Every team picking its own objectives, in its own quarter, with no thread connecting any of it to where the company is trying to go. You would not call that a strategy. You would call it a mess, and you would put someone in charge of fixing it by Friday. Run AI adoption on exactly that model and we call it a digital transformation and put it in the board deck.
I want to be fair to the bottom-up instinct, because it is not stupid. It comes from a good place. The people closest to the work know where the friction is. Push tools down to them, let a thousand experiments bloom, and the best ones survive. That is how a lot of genuinely good technology has spread through organizations, and there is real wisdom in not waiting for a committee to bless every useful thing. If you had asked me a decade ago how new tools should get in, I would have argued for exactly this. Get it into the hands of the people doing the work and stay out of their way.
Here is what that argument misses, and it took me too long to see it. A bottom-up rollout makes each team faster at whatever it was already doing. That is the whole ceiling. Faster. Not better-aimed. A team that was quietly working on the wrong thing is now working on the wrong thing more efficiently, and it feels like acceleration the entire time. The tool did its job. It just had no way of knowing the job was pointed in a stale direction, and neither did the people using it, because the direction was somebody else’s to set and nobody had refreshed it.
A cascade behaves differently for one reason. Direction sets context. When the layer that makes the real decisions gets sharper, every layer beneath it inherits a clearer sense of what to optimize for. The marketer is not just producing more copy. She is producing copy aimed at the segment the company actually decided to chase this year. The engineers are not just shipping features faster. They are building toward the decision that got made last month instead of the one a nine-month-old roadmap still implies. Same tools. The thing that changed is what they point at.
I watched a COO live the whole arc of this. Her org had done everything the bottom-up playbook asked. Tools in every function, a real budget behind them, a genuine and constant sense of acceleration in every status update. And underneath it, a quiet unease she could not source. Everything was moving. Nothing was converging. When we mapped where the intelligence actually sat, it was all pooled at the bottom, in a dozen local speed-ups that had never been asked to agree with each other.
So we started over at the top. Not with the tools. With the handful of strategic decisions the company was actually running on, most of which were being made on stale or thin information while every team below them had better tooling than the executives did. We got those few decisions onto better footing. Then we let that clarity flow down the same channels the company already used for goals.
The stack did not change. I want to be clear about that, because it is the part people expect to be different. Nobody bought anything. Nobody ripped anything out. What changed was what the tools were aimed at, and within a couple of quarters the same dozen tools that had been producing a dozen separate accelerations started producing coordinated movement instead.
The effect I did not see coming showed up in her leadership meetings. Once the top few decisions were built on real evidence, the standard for evidence everywhere else quietly rose to match. People stopped bringing gut calls to the table, because a gut call now looked underdressed next to the decisions around it. Nobody announced a new policy. The bar just moved on its own, the way it does when the person at the head of the table starts modeling it. That second-order shift, a leadership team raising its own standard of proof, ended up mattering more than any single tool in the building.
Why does the bottom-up version feel like progress the whole way through, even when it is not converging? Because every individual piece of it is real. The marketer genuinely is faster. The support team genuinely is deflecting more tickets. Every local metric is up and to the right, and each one is honestly earned. Progress you can feel in a dozen places at once is very hard to argue with, right up until someone asks what all of it added up to and the room goes quiet. Motion is easy to see. Direction is the thing you have to stop and check for, and nobody schedules the meeting to check.
None of this is exotic. The mechanism is the one your company already runs every planning cycle. Set intelligence at the decision layer first. Let each layer below translate that clarity into what it optimizes. The bottom-up tools still matter. They finally have a direction worth serving.
You already know how to do this. You do it every quarter with goals, and you would never dream of setting those one team at a time, in random order, and hoping the total pointed somewhere. The strange part is that with the most capable tool any of us has been handed in a generation, we decided to start at the bottom and hope it added up.
The strategic layer is the part almost nobody is building on purpose, and it is the part we build first.