Michael Ewens and Xavier Giroud wanted to know what American companies actually look like on the inside, so they went and got nearly seven million resumes. No surveys involved, and no org charts handed over by a corporate development office. Just resumes, posted online by ordinary employees describing the jobs they had held and the ones they moved into next, covering more than 3,100 U.S. public firms between 2016 and 2023. They ran a network estimation across all those transitions to sort the jobs into ranks, since promotions, demotions and lateral moves leave different fingerprints in a career history. What came out is a map of the rungs inside corporate America that nobody in those companies drew. The average firm has ten layers.1

An executive once described their company to me as flat. Two hundred and sixty people, and they meant it as a claim about how decisions got made. Later in the same conversation they walked down their own reporting lines, from themselves to the people running production: team leaders at the front, plant managers and operations directors above them, four or five stops in all. They had the number the whole time. What they did not have was any sense that the two descriptions sat oddly next to each other.

What ten layers buys you

The Ewens and Giroud data lets you price the thing, which beats another round of argument about whether hierarchy is good for people. Firms with more layers earn a higher return on assets and a higher net profit margin. Their employees stay longer, and more of their internal moves are promotions instead of outside hires, which is what a career ladder looks like when one genuinely exists. Stock volatility is lower, and so is operating asset volatility, which fits the theory that extra managers work as problem solvers who absorb trouble before it reaches the reported numbers.1

The bill is itemized too. Selling, general and administrative expense runs substantially higher, and so does the wage bill, because the people you add at each level cost more than the people below them. Revenue per employee drops, which follows mechanically once you staff up with people who do not touch the product.1 None of that is a scandal. It is the price of the thing, and it is a defensible price if you know you are paying it. The trouble is that the cost of a layer sits in a line item your CFO reviews every quarter, while what that layer teaches the people underneath it has no line at all.

More patents, no better patents

Here is where the paper stops being reassuring. Firms with more layers file more patents, and the result survives two different estimators. When the authors turn to citations per patent, which is the standard rough proxy for whether anybody else found the work useful, the association goes away. They also tested novelty and right-tail outcomes and found nothing there either.1

Ewens and Giroud read this charitably, and they are entitled to. Their stated conclusion is that multi-layer hierarchies are not detrimental to innovation production. That is accurate. My read is less comfortable than theirs. If you add layers and your patent count climbs while your citation profile holds steady, what you purchased was throughput. The apparatus got bigger and produced proportionally more of the same grade of work, and you covered the expansion out of overhead and payroll.

Adding a layer is a procurement decision with a known bill attached. You get more output and steadier numbers, while the quality of the thinking stays exactly where you left it before the reorganization started.

Eighty-two percent of you will never touch it

Then comes the finding that reframes everything above it. Across the whole 2016 to 2023 window, only 18 percent of firms experienced any change in their number of layers.1 Eight years of strategy offsites, consulting engagements, new chief executives, a pandemic, and the arrival of generative AI, and better than four in five companies came out with the same number of rungs they went in with. That says less about whether restructuring works than about how seldom anyone actually does it. The conversation is continuous and the structure is close to inert.

The direction of travel is old news. Raghuram Rajan and Julie Wulf documented the flattening decades ago, using job descriptions, reporting relationships and pay data from more than 300 large U.S. firms between 1986 and 1998, and found more managers reporting directly to the chief executive as the levels between division heads and the top thinned out.2 Ewens and Giroud pick the story up much later and watch the average layer count keep drifting down through 2023. They also find firms cutting layers after adopting AI technologies, though those estimates hold only at the 10 percent level, thin enough that I would not build a plan on it.1 Theirs is a working paper and has not been through peer review.

I wrote in June about what disappears when a company deletes its middle management layer, and I am not going to relitigate that here. The narrower point is this. Most senior leaders talk about their structure as a lever within reach, something a determined executive could pull next quarter. The evidence says it is closer to sediment, laid down by a decade of hiring decisions, retention accidents, acquisitions nobody fully integrated, and the ordinary human habit of adding a manager whenever a team gets uncomfortable.

Count it before you argue about it

The useful exercise here is far smaller than a reorganization, and most executives I meet have never done it. Count your layers. Skip the boxes on the chart in the board deck and count the stops a decision makes between the person who first sees a problem and the person who can authorize a fix. Then ask what that count has been buying, and check the answer against your administrative expense line and against the work your organization has produced since 2023. If the count is high while the margins look fine and your best output has not moved, you have a fairly precise picture of what you bought.

So how many layers deep is your company, and when did anyone last sit down and count them?