The thing that slows down a large American company is almost never the idea. I have sat in enough of those rooms to know. Someone notices something true about a customer, and the noticing is correct, and everyone in the room agrees it is correct. Then it goes into the machine. Legal has a question and brand has a preference, and there is a governance forum that meets monthly and a data-sharing agreement written for a different purpose. Nothing in that sequence is unreasonable on its own, but by the time an answer comes back the thing that was noticed is no longer true, and the company starts over on a customer who has since moved.
I could not tell you which project taught me this, and that is what stays with me. It was never a single one, just a shape that repeated: an observation that was true on the day someone made it, and a process that treated it as though it would keep. Customer knowledge does not keep. None of it failed, exactly. It arrived correct and expired.
I did not have a word for that pattern until I went to China, which is a strange thing to say about a problem I had been living inside for years. I went with a Yale delegation, two weeks of arranged company visits, sitting with the people who run these businesses and asking them questions. Walking around the same cities on my own would have told me almost nothing. I went expecting to find a technology gap, and I did not find one.
I should start with what I got wrong, because it sets the terms: I could not work out Meituan on my own. Meituan is how most of China orders food, and also how they buy groceries and book restaurants and hotels, all the things we keep in three or four separate apps. I stood on a street in Beijing with the app open and no English anywhere I could find, and someone on the trip had to walk me through what I was looking at. Then the numbers had to be explained to me too. What I was told, once I understood what I was seeing, was that the numbers on the screen were the restaurant’s: a rating, how many of each dish had sold that month, which ones the kitchen recommended and which ones other customers had made popular, what an average person spends there. In the US that kind of data stays inside the company, and the only version of it a customer ever hears about is the collection they are supposed to worry about. Here it was just there, on the way to dinner. The food was almost incidental. What the platform ran was a continuous read on a few hundred million people, and the read was valuable only because it reached the person deciding. Filed in a quarterly deck, it would have been trivia.
I kept meeting that pattern for two weeks before I understood I was looking at something closer to a stance than a series of impressive companies. In March 2020 the Central Committee and the State Council issued a document naming data a factor of production. Land, labor, capital, technology, and now data, the fifth thing an economy is built out of. Xi had used the framing a couple of years earlier, but 2020 is when it became doctrine. I want to be careful with how much weight that carries, because a document is not a cause and plenty of countries publish industrial policy that changes nothing. What I can say is narrower: the companies I was shown were behaving as though data were fresh food rather than a stored asset, something you use at the speed it spoils, and that posture turns out to matter more than capability.
Luckin Coffee opened its first stores in Beijing in 2017 and ended that year with nine of them. It passed 35,000 in the spring of 2026, having opened its thirty-thousandth only that February. This is, worth saying, the same company that got caught inventing a year of sales in 2020 and was delisted for it. The machine I toured is the one they rebuilt from that wreckage, which is its own data point about speed. In 2025 it launched more than 140 new drinks, and what they told us was that the point is to have something for everyone, then let the stores show you what people actually reach for and keep what sticks. You cannot plan 140 drinks, only discover them, and the store network is dense enough that results come back while they are still true. In the US coffee is sold as ritual and identity, and the boutique cafe is the aspiration. Luckin went the other way and rebuilt coffee as a fast, digital-first, personalized product, and the country moved. China drinks something like 40 percent of the world's tea, and coffee consumption per person has more than doubled in five years, up roughly 150 percent across the decade. Changing what a country drinks is not a demo. It is the hardest kind of adoption there is.
At NIO the competition had moved inside the car, because with more than a hundred EV brands fighting over the same buyers, performance and price stop being enough to tell you apart, and what is left is how people spend the time they are already sitting there. I had been watching that happen all trip without registering it. DiDi is what people there use instead of Uber, and in ride after ride the driver held an entire conversation with the car, sending texts and once translating something for me from the front seat. I came in thinking I was unusual for treating my car as a second home. NIO had built the whole product on the assumption that everyone does.
I want to be careful about what a two-week trip can prove, because we were shown things: companies that host a Yale delegation are companies that are winning, and they present accordingly. I did not see the write-offs, or the hundred EV brands that will not exist in five years, and most of them will not. But the caveat cuts in a useful direction, because the argument was never that China is winning. Nothing I was shown was built out of anything we cannot build. The difference ran on the clock, in how much of what they learned was still true by the time they acted on it. You can see the same gap opening at home right now, as the latest wave of enterprise technology arrives on schedule and then sits in the queue behind the same reviews that aged everything before it.
Some of our friction is deliberate and worth keeping. The reason a read on a few hundred million people is possible there is that the people being read have very little say in it, and I would not trade our protections for that speed. But there is a test that separates the two, and it is a simple one: can you name the failure this rule prevents? Good process can. It exists because something went wrong once and somebody wrote it down, and it gets revisited when it stops paying. Most of what I have watched slow companies down cannot answer the question at all; it is sediment, a forum that exists because it has always existed, a policy written for a risk that no longer applies. We tell ourselves we are being careful, when often we are just being slow and calling it care.
Insight has a half-life. Every true thing a company learns about a customer starts decaying the moment it is learned, and every review it waits in is time off the clock. So the question I ask about a company now is rarely about what it knows. Most of the good ones know plenty. I want to know how much of it is still true by the time it reaches the customer, and whether anyone inside has ever thought to measure the decay.