Why All Big Companies Eventually Die

 . . . because they’re full of average people doing average things. 

Executive Summary

• Clay Christensen was right--big companies often die because they were trying to do the right thing (listen to their customers.) But he may have been too generous--big companies also die because they are big.
• As companies grow, they accumulate pains of scale that offset the better-understood benefits of scale.
• One of the most important (and counterintuitive!) pains of scale is that average employee quality invariably declines as firms grow.
• Employee quality declines because of the Law of Large Numbers, the same phenomenon that makes casinos rich and frequent bettors poor.

Introduction

Have you ever wondered why big companies, with all their resources and advantages, manage to so consistently get sick and die? Of course you have. We all have, because we know there’s no reason why it has to happen. Companies aren’t living beings. Their hair doesn’t thin, their eyesight doesn’t fail, and their arteries don’t harden. They are repositories of knowledge that can be refreshed every time new people join them.

So why does it happen? Why do companies rot as if they are biological beings? Every business book published tries to answer this question, but they never seem to succeed. That’s why I hate business books. 

One of the few business books that makes sense is the late Clay Christensen’s Innovator’s Dilemma, a much-admired and much-misunderstood work that starts with the assumption that companies are full of competent, well-intentioned people who are listening very carefully to their customers. This was a brilliantly original approach, defying the conventional wisdom that claimed that big companies get “disrupted” because they become fat and lazy.   

Christensen’s hypothesis still rings true, but with more than twenty years of experience inside large firms, I now believe that he may have been too generous. Big companies, as they grow, experience pains of scale (the opposite of economies of scale.) These pains accumulate and create a friction that offsets the advantages of being big. Pains of scale are the corporate equivalent of creaky joints and arterial plaques.  

Many of us have good intuitions about what these pains are, often because we’ve seen them firsthand. But we don’t all have a common model for exactly what is happening, which is surprising because we do have a shared understanding of the benefits of scale (e.g. lower fixed costs, employee specialization, leverage over suppliers, etc.) 

This article is the first in a series where I will explore these pains of scale in detail. 

Details of the Argument

First, let’s agree that great corporate accomplishments (or accomplishments of any kind, really) require great individual deeds, performed by people who are extraordinary in some way. Perhaps these people are brilliant, or maybe they’re incredibly creative, energetic, or brave. Whatever they are, they are not average. Average people doing average things is important for keeping a firm functioning, but it doesn’t create value in the same way as the invention of brilliant new products, or the discovery of incredible new insights, or the delivery of fantastic customer service.

For some reason, this is a difficult concept for people to accept. The same people who know that they’ll never replace Patrick Mahomes as the Kansas City Chiefs’ quarterback often believe they are one lucky break away from becoming the next Bill Gates. Business is commonly viewed as somehow different from basketball or sculpture or music composition. We don’t want to believe that natural endowments or thousands of hours of intensive practice are required to do amazing things. 

Businesses often promote this myth themselves, in large part to build employee engagement. But the danger of telling everyone that they can accomplish anything is that they start to believe it. This breeds overconfidence and creates the expectation that every team member gets to have an equally loud voice in steering the company. For reasons I will explore in later posts, this democratic spirit can be very bad for a firm’s health, leading to a tyranny of the mediocre majority.     

Convincing you that corporate democracy is a bad idea will take some time, so let’s set that aside for now and just agree that great companies need a high density of extraordinary people. Our intuitions suggest this should be easy for large firms to accomplish. After all, they typically have lots of benefits they can use to attract the best candidates--high salaries, big titles, large scopes of responsibility, and widely recognized brands.

But it doesn’t work out that way. To understand why, we have to dredge up some math you may have tried to leave behind long ago. 

The Law of Large Numbers

The Law of large numbers (LLN) is something nearly all of us have heard of, many of us have cited, but few of us understand. If most of us really understood the LLN, we would behave far differently than we do in many dimensions of our lives.

First proved by Jacob Bernoulli in a proof published in 1713, the LLN is a mathematical theorem that states that the outcome of a series of trials is more likely to approach the expected value as the number of trials increases. That’s quite a mouthful, so let’s try to pull that apart. 

The “expected value” is the average outcome we would expect to happen before an event occurs. This average is calculated by summing up all the possible outcomes multiplied by their associated probabilities. A common example that’s used to illustrate expected value is rolling a die. Assuming the die is not “loaded”, each side will have an equal probability, one out of six, of turning up. The math for expected value, then, will simply be:

Expected Value = (⅙ x 1) + (⅙ x 2) + (⅙ x 3) + (⅙ x 4) + (⅙ x 5) (+ (⅙ x 6)

E(x) = 1/6 + 2/6 + 3/6 + 4/6 + 5/6 + 6/6 

E(x) = 21/6 = 3.5

Obviously a single roll of a die can’t turn up a 3.5, but averages calculated for situations with “discrete” outcomes can sometimes fall between the real world possibilities. It’s like the old joke about a family having “2.2 children.”  

With our die roll example, the LLN simply predicts that the average of all the rolls we make will settle at or very near 3.5, and that this outcome becomes more certain as you make more attempts. Figure 1 below illustrates what this looked like in one experiment. As you can see, the average started as high as 5.5 (perhaps the person first rolled a 5 and then a 6) before quickly plummeting to 4 (their next roll may have been a 1). After a few more rolls, the average fell further until it dropped below 3.5. This gambler continued to experience “bad luck” for quite a while, with his or her average persisting below 3.5 for a few hundred rolls. But by the 400-roll mark, the roller’s average was very close to 3.5, with the two lines becoming almost indistinguishable after 650 tosses of the die. 

Figure 1 (Source)

The LLN is what makes casinos such an attractive business. The games of chance they offer their customers involve no chance at all for the casinos (you may insert your own joke about Trump here.) With millions of trials happening within their walls, the outcomes for their businesses are never in doubt--they will generate margins at a completely predictable rate. Gamblers who visit only occasionally may get lucky and win some money, but the hard-core bettors who are playing very frequently will have outcomes as certain as the casinos'--except they will be on the losing side. So, the next time your Aunt Betty describes to you all the times she has won during her weekly trips to the local casino, just remember that she is conveniently forgetting all her larger, off-setting losses. 

Most things in the world are not limited to a few discrete outcomes. Human height, weight, or intelligence, for example, can fall anywhere within a range. The distribution of how frequently a random person will fall in a particular spot within that range is commonly described as a normal distribution. Figure 2 below shows a normal distribution’s bell-shaped curve: 

Figure 2

A distribution of height for American adult males might look something like what’s shown in Figure 3. The average height, according to this distribution, is 5’10”. On the right side of the curve, you can see that few men are taller than 6’8”. On the left side, you will notice that it’s relatively rare for men to be shorter than five feet tall. 

Figure 3

The shape of this curve and the LLN have two implications for an experiment where we randomly choose American men and check their height. First, the shape of the curve suggests that the most likely height we will record for any random man is 5’10”. “Most likely” doesn’t mean that it’s terribly likely. As you can see, there’s a wide range of possible heights we might observe, from less than five feet to more than seven. Second, the LLN tells us that the more men we randomly select and measure, the more certain it is that the average height of those men will be 5’10”. Like our rolling of the die example, we might have a very short or very tall group early in our experiment, but as we continue on, the average will inevitably converge on the distribution’s peak of 5’10”. 

It’s safe to say, because of these statistical phenomena, that the average height of the male employees of any large, American firm in this world we’re describing would be 5’10” or very close to it. It may be slightly different if there’s a regional or ethnic bias to the company’s employee base (perhaps the firm is based in Minneapolis where there are a lot of tall Scandinavian Americans, or in Los Angeles where there is a higher density of Latin Americans), but as long as the firm is big enough, has several offices across the country, and doesn’t systematically discriminate against short people, it’s a near mathematical certainty. 

Do you know what other human characteristics are distributed normally in bell-shaped curves? Nearly everything we might value in an employee: intelligence, curiosity, integrity, bravery, creativity, and persistence to name just a few. Ordinary employees have some amount of all these things, but it is rare to find people who have a large dose of any of these qualities. 

How difficult is it to hire a high ratio of extraordinary employees? As the LLN predicted in our height example and our die experiment, it will be increasingly challenging as a company grows larger. Early in a firm’s life, when it has only a few employees, it may be lucky enough to have a disproportionately high share of amazing people. The opposite can be true too, of course--a small company could be stricken with too many below-average employees, which would probably cause it to fail early in its life. Figure 4 tries to illustrate how this will work. The two red lines outline the outer bounds for the quality of a firm’s average employee as it grows larger. The bigger the company, the more average its average employee becomes. 

 Figure 4

The statistically fluent reader will point out here that Figure 4 is only an accurate view of the world if new employees are randomly selected from the population. Surely companies aren’t doing that! They have very sophisticated hiring practices that are designed to accurately identify the very best candidates, right? 

While hiring processes at large companies are not purely random, the outputs of such processes become more random as companies grow larger. Instead of a few founders scrutinizing applicants, big firms have hundreds of independent hiring managers making their own decisions about who to bring into the company. They may have to pay lip service to hiring guidelines, involve other managers in a review process, or even subject their candidates to psychometric testing, but as we all know, in the end, each hiring decision will mostly reflect the hiring manager’s personal tastes. 

The Finite Supply of Talent

If you’re not convinced that the LLN will drive down the quality of the average employee in a growing company, consider the effects of the finite supply of talent. 

Let’s imagine that in 1982, Bill Gates and Paul Allen were the two most talented software engineers in Washington state. Furthermore, let’s assume that they had perfect information about who the third-best engineer was in the state, and that when the time came for them to grow, they were able to convince that person to join Microsoft. At that point, Microsoft’s worst engineer was better than all but two other engineers in Washington.

Now let’s imagine that every time Microsoft needed to hire its next engineer, the same conditions applied--the company knew exactly who the best available engineer was and was able to successfully hire that person. Let’s assume further that the company managed to keep all its best engineers, that it never lost them to its competition. 

Even with these ridiculous assumptions, Microsoft would have experienced a relentless, inevitable decline in the quality of its employee base. When the company hired its 5000th engineer, that person would only be the 5000th-best engineer in the state of Washington. The average engineer on staff would be the 2500th-best. 

This would be fine if the best engineers at Microsoft consistently had the most influence and the most decision-making discretion. But as we will see in later articles, this would probably not be the case--power and decision-making rights get distributed as firms get larger.

But what about all the other small software firms in Washington? They didn’t have Bill Gates and Paul Allen working for them. Wouldn’t they have a chance to acquire better talent as they grow? Yes, but first they had to grow. Growth requires a great proposition, and a great proposition is often created and fueled by extraordinary people. The hundreds of small software companies in 1980s Washington that didn’t have Bill and Paul all died, probably at an early stage. It’s not true that all small companies have better people than large firms, just that the best ones do. And all large companies, at one time, were the best small companies. 

Putting the LLN and the finite supply of talent together, we can generate . . . 

The First Law of Large Companies: Large companies will always have lower quality employees, on average, than the best small companies.