In 2015, taking what I thought was the advice of the investor Warren Buffett, I bought shares in Wells Fargo, one of the four largest banks in the United States. Oops. Over the next five years the bank admitted to opening millions of accounts its customers had not asked for, paid fines, lost its chief executive, operated under an asset cap imposed by its regulator, and watched its share price fall while the rest of the sector rose. I sold (and the share price recovered after I was gone). Yet Wells Fargo has operated continuously since 1852 and is still there; it seems pretty nearly impossible to kill this organization. Few companies last as long as Wells Fargo, but incumbents in general are remarkably resilient. A company that survives infancy is very likely to live for decades, either as an independent firm or as a division of whoever buys it.
An organization is a machine for doing a job. It exists because it can do the job for its customers better or more cheaply than they could do it for themselves. For a company, economic viability is a pretty simple formula: revenue above cost, with revenue rising and cost falling over time relative to rivals. But what allows incumbents to keep that formula running through scandals, recessions and their own mistakes? I call the mechanisms that sustain them flywheels. In a machine, a flywheel is a heavy wheel that accumulates energy from an uneven power source and smooths the output. In an organization, the same word fits the mechanisms that store the results of past effort with customers, cost, quality, reputation and capability, and release them as needed.
Most organizations, I believe, have five flywheels. Each is a stock that grows as the business runs and a candidate alpha asset, since each improves results and each takes years to build. Which of the five counts for most depends on the industry. As chapter 4 put it at its close, the flywheels do not ask where the first push came from.
Customer Network
Very few businesses serve each customer just once; cataract surgeons and MCAT tutors come to mind. Nearly everyone else acquires a customer and then earns from the relationship over years. This is why marketing teaches customer acquisition cost and lifetime value as a pair.
When we were just starting out at MakerStock, a company I co-founded that sells pre-cut acrylic and plywood to people who fabricate things with laser cutters, we bought search advertising. Two or three of every hundred people who clicked placed a first order, and about one in seven of those placed a second and became a repeat customer. So for every three hundred clicks or so we acquired one repeat customer, and most weeks we added a handful. By the second year we had a few hundred of them, and we worked hard to keep them: we'd special-order materials, send samples of new materials, and offer cutting services. There is really no shortcut to a customer base, since customers are added one at a time. Once accumulated, though, the stock persists even when the effort stops; satellite television companies still bill customers who installed a dish on the roof twenty years ago.
In some categories the customer network does something the other flywheels do not. It accelerates. When the value of a product to a new customer rises with the number of customers who already use it, the product has a positive network effect, and each customer acquired makes the next one cheaper to acquire. A payment app is worth little when only your children use it and a great deal when the man who plows your driveway does. Platforms that assemble many sellers for many buyers, such as an app store or an online marketplace, have network effects on both sides. In a category with strong network effects the market tends to tip toward one provider. Once it has tipped, the installed base is among the most powerful alpha assets a firm can hold. A rival cannot buy an installed base, nor can it build one against a tipped market. Good luck launching a mobile operating system to rival Apple's iOS and Google's Android.
I have two cautions. First, network effects are mostly a property of the product, and a product that lacks them usually cannot be given them. I have tried to think of a way for MakerStock to benefit from one, and I can't, because the value of the right sheet of acrylic on the right day does not depend on how many other people are buying acrylic. Second, switching costs, often listed beside network effects as a source of lock-in, are a weaker asset than they look. Customers resent them, competition tends to push markets toward interoperability, and a customer who stays only because leaving is painful is already looking for the exit.
Brand Equity
My education is in mechanical engineering, my father is a chemical engineer, and my brothers are engineers. We worship at the church of science and reason, and at first glance brands are ridiculous. In one of my first papers as a professor, Taylor Randall and I examined prices in the mountain bike industry. At the time nearly every frame was welded in a handful of Asian factories, and models from different brands often differed in nothing but the decals. Even so, we found that the brand name alone accounted for hundreds of dollars of price. The same thing can be observed in a Costco every day. Shoppers pay half again as much for Tito's vodka as for Kirkland, Costco's own brand, and even the Kirkland bottle commands a premium over the unbranded one. (My preferred cocktail is vodka mixed with a potent home-made kombucha, and I am 100 percent certain I could not taste the difference in that use.)
There are three logical reasons people pay for brands. First, a brand is a credible promise about qualities the buyer cannot inspect, since the owner of a valuable name has more to lose from an impure batch than an anonymous producer does. Second, a brand reduces search cost. Yes, the shopper could read reviews and run tastings, or they could spend fifteen seconds in the aisle and grab the Tito's. Third, a brand is a signal to other people; the bottle on the shelf where guests can see it is chosen differently from the bottle in the cupboard. Put the three together, and even for goods of identical intrinsic quality, the stronger brand commands some combination of higher price and higher share.
Brand equity is the accumulated value of the name, and for many companies it exceeds the value of everything physical they own. It is a flywheel because it grows with time and volume. Awareness rises with every unit sold and every year in the market, longevity itself becomes a signal (count the firms that print their founding year on the label), and improved products feed the name. Bad conduct drains it, as Wells Fargo shows, but the drain is slow enough that an established brand can absorb a scandal that would end a newcomer.
Cost Efficiency: Learning and Scale
Have you ever hung a sheet of drywall? Drywall is a paper-faced gypsum board, about twenty dollars a sheet, heavy, fragile and dusty to cut. A novice takes about an hour to hang the first sheet, while a crew of three professionals hangs twenty an hour. It's crazy how they do it. I once watched an installer with short stilts strapped to his boots balance a sheet on his head, screw it to the ceiling, find the electrical box hidden behind it from the bulge in the sheet, and cut the opening with a hand router, all in about three minutes.
The learning curve is the empirical regularity behind that crew: with each doubling of cumulative units, the time per unit falls by a constant percentage. On an 80 percent curve (and such curves are common), if the first sheet takes sixty minutes, the second takes 80 percent of that, or forty-eight minutes. The fourth takes thirty-eight, the eighth thirty-one, and the sixteen-thousandth, fourteen doublings later, about two and a half minutes.
I use the fancy academic term empirical regularity because the learning curve is not a law derived from anything; it just works out that drywall, coronary bypass surgery, airframe assembly and insurance claims processing all follow it. The individual improvements are lumpy and unpredictable: the stilts and the router are big steps, positioning the sheet with two nails before driving the code-mandated screws is a small one, and a mechanical pencil gives a barely detectable edge. Blur your eyes, though, and the steps merge into a smooth curve.
Two features of the curve bear on advantage. First, because each percentage improvement needs a doubling of cumulative volume, the curve has diminishing returns, and a firm with a million units behind it improves slowly. Second, the rate of descent depends on deliberate effort. Frederick Taylor's time studies, the statistical quality methods that Walter Shewhart developed and W. Edwards Deming carried to Japan, and the production system that Taiichi Ohno and Eiji Toyoda built at Toyota are all techniques for descending the curve faster than repetition alone would. A firm that combines a head start in volume with a better improvement process has a cost advantage that compounds.
Scale economies are distinct from learning and often confused with it. For twenty-four years I have owned a small business, Xootr, which makes adult kick scooters. When we first started, my brother machined the aluminum hubs for our wheels in a barn in New Hampshire. Those hubs cost 6.14 dollars each when he was making ninety-six a week. At our peak we used 3,200 hubs a week, and at that rate they cost just 2.14 dollars. Some of the reduction was learning; for instance, we figured out that sandblasting the aluminum gave a better finish than polishing it, for less work. But most of the reduction came from switching from machining to die casting, which requires a custom mold (at the time about forty thousand dollars) that makes no sense at ninety-six hubs a week and every sense at 3,200. Scale economies arise when a higher rate of output justifies a different process, usually one with more capital and less labor per unit. They also extend upstream, since a large buyer gets better prices from its suppliers, who enjoy scale economies of their own. Services have weaker scale economies than factories, but specialization still produces some, until coordination costs overtake them.
Both learning and scale are flywheels: lower cost permits lower prices, lower prices bring volume, and volume brings more learning and more scale. Both are hard to acquire, because the only route to volume is volume. They are also stickier than they look. Can a way of hanging drywall really be hard for a rival to acquire? The strategy researcher Gabriel Szulanski studied the transfer of proven practices between units of the same company, where everyone wanted the transfer to succeed. He found that even simple practices moved slowly or not at all. Now imagine a rival trying to copy the same know-how from the outside, against its owner's resistance; that takes years.
Product Quality
When I was in college I owned a series of beat-up used cars, none of which cost me more than a thousand dollars: a 1969 Plymouth Fury, a 1970 Plymouth Valiant, a 1975 Toyota Corolla and a 1979 Chevrolet Nova. Each was a little over ten years old when I bought it, and I spent a dozen Saturdays a year under the hood of whichever one I owned. Today a ten-year-old car runs like a new one, and I'm pretty sure most of my students do not own the deep socket needed to change a spark plug.
Some of that improvement came from Toyota, Honda and Nissan making reliability a dimension of competition in the 1970s and 1980s. Most of it came from the ordinary process that every car company and every supplier has run for half a century: collect failure data, find root causes, rank the fixes by cost and impact, and implement them in order. Product quality is a flywheel because every unit delivered produces information about the next one, so the firm with the most units and the best process reaches the highest quality first. Like learning, quality improvement has diminishing returns: in 1970 reliability was a decisive advantage, whereas today new cars vary too little in reliability for it to be one. So the product-quality flywheel confers its largest advantage while a category is growing, and the advantage erodes as the category matures.
Organizational Capability
Of the five flywheels, this one is the most exposed to the halo of chapter 3, in which a successful firm's culture gets described by its success. So what would make an organizational capability more than a halo?
The Ritz-Carlton hotels, a unit within Marriott, have for years been at or near the top of the luxury segment in independent guest-satisfaction surveys. Part of the capability behind that is a system, in which any employee can record a guest's preferences and the record follows the guest to the next property. Part is a culture of service, built over decades, that attracts and keeps people who like delighting strangers. Marriott operates several other luxury brands, none of which has matched the Ritz-Carlton's results, even though the sister brands have the same parent, the same capital and access to the same system. Because that failure to match is the pattern Szulanski found inside companies, an organizational capability, though intangible, can pass the hard-to-acquire test.
The capability is a flywheel for two reasons. Like the other four, it improves with repetition; a service organization learns from every complaint. And it has a positive feedback loop of its own, in which a stronger culture of service attracts and retains employees who want to work in one and who in turn strengthen it. The same loop can run on other capabilities; for instance, a company known as a good place to work attracts the engineers who make it a better product company, and a company known for environmental seriousness attracts people with expertise in it. Diversity, process discipline, speed and innovation can each run the loop.
Phil Rosenzweig, author of The Halo Effect, would raise the obvious objection, and I raised it against myself in chapter 3: a capability named after the fact explains anything. A firm succeeds and we discover its culture of customer focus; a firm fails and we discover its culture of complacency. My defense is procedural: name the capability as a set of observable practices, not an adjective. Point to an observation that does not depend on the result, such as the sister brand that could not copy it, the survey conducted before the financial result was known, or the practice the firm invested in at a visible cost before it paid. And include the counter-case.
Sierra Trading Post, a retailer of discounted outdoor gear, was founded by a man who printed Bible verses on its order forms and declared its ethics to be those of its owners' faith. The practice was real, costly and sustained. It produced a loop of its own, attracting employees who shared the faith and who in turn reinforced the practice. But it was not an alpha asset. Its customers were backcountry skiers and climbers who, judging by the online forums of the day, found it somewhere between irrelevant and off-putting, and when the company was acquired, the verses disappeared. Cotopaxi, a gear company founded a generation later around poverty reduction in Latin America, runs the same kind of loop with values its customers share. The fifth flywheel is an alpha asset only when the capability feeds one of the other four.
How many such loops can one organization run? One or two, not five. The psychologist Anders Ericsson's research on expertise (the real version, not the ten-thousand-hour slogan) makes the point for people: being the best in the world at one thing consumes everything, and being the best at two is nearly unheard of. Organizations face the same limit, with finite management attention and finite cash. A single capability is also easier to communicate; the sign James Carville hung in Bill Clinton's 1992 campaign headquarters read "the economy, stupid," and almost nobody remembers the other two lines. And a single objective is easier to act on. A Ritz-Carlton bellhop asked to store twelve bags knows what to do when the organization's value is guest satisfaction, and does not when it is guest satisfaction balanced against cost.
The best evidence that organizational capability is causal rather than a halo comes from the experiment chapter 3 cited. When the economist Nicholas Bloom and his colleagues randomly assigned months of management consulting to textile plants in India, the treated plants adopted practices in quality control, inventory and efficiency. They became about a sixth more productive in the first year. When the researchers returned nine years later, some of the practices had eroded and some of the gain remained, so the intervention acted more like a single injection than a flywheel. A practice is an asset. A capability that keeps adopting and refining practices is a flywheel.
Flywheel Physics
Whether a later entrant can catch an incumbent depends on how each flywheel responds to speed.
Brand equity shows roughly constant returns. It is built brick by brick, with no reliable shortcut, and the firm that started earlier and invested continuously has more of it (think of Starbucks in coffee, Adidas in sportswear, Sony in electronics). A later entrant can match it only by investing for as long.
Cost efficiency and product quality show diminishing returns, by the arithmetic of the learning curve: the twenty percent that comes easily between one hundred units and two hundred comes glacially between one million and two million. So a newcomer with some initial boost, usually a better product, stands a chance of reaching competitive cost and quality against an incumbent whose own improvement has slowed.
Customer networks show constant returns in most industries, where the ten-thousandth customer takes about as much effort to win as the hundredth. In industries with network effects they show increasing returns, since each customer makes the next one cheaper. Increasing returns are the exception and also the most important case. A firm that gets a toehold early in a category with strong network effects and keeps its base growing becomes nearly impossible to dislodge, for as long as the category lasts.
Organizational capability can show increasing returns as well, through the talent loop: the better the culture, the better the people it attracts, the better the culture. Systems thinkers call loops of this kind snowballs. Chinese has a four-character phrase for the general idea, which my first teachers used to encourage me: the more, the better it gets, and the better it gets, the more.
The single word flywheel covers three different things. Some of what I have described is accumulation of a stock: more customers, more cumulative units, more years of brand. Some is learning, in which later units are made better than earlier ones, and some is self-reinforcement, in which the existing stock makes the next addition cheaper. Although the three can coexist, a customer base can be large without network effects, in which case it is a stock that protects but does not accelerate. Stocks also depreciate: learning that is not used decays, as the management scholar Linda Argote and her colleagues found in shipyard production data; a brand that is not fed fades. So for each flywheel in a particular business, I would ask what grows, why that makes the next round easier, what has to be invested to keep it growing, and what eventually limits it.
Not every alpha asset is a flywheel. If a famous singer founded a fragrance line, the singer would be an alpha asset. But the boost they gave the business would rise and fall with their celebrity rather than grow with the business. A patent is an alpha asset for as long as the patent runs, and a regulatory license is one until the regulation changes. These are endowments, one-shot injections of advantage of the kind chapter 4 described. Who wouldn't love to receive one? But I know of no process for acquiring geese that lay golden eggs. Flywheels, in contrast, can be built from nothing by any firm willing to spend the years, and once spun up they store energy that an endowment cannot.
Then there is the tool that everyone now has. A frontier language model is performance enhancing, but because it is not hard to acquire, it fails chapter 2's second test, as electricity does. What it changes is friction. A product that can be imitated in a week has a shorter life as an alpha asset than one that took a rival a year to copy. So the product-quality flywheel protects less than it did, while the flywheels that a model cannot shorten are worth relatively more: the customer base, the brand, the relationships, and the organization's capability to deploy the tool well. Is a firm's use of the model a fifth flywheel or a halo? That is settled the same way as for any other capability, by naming the practices, pointing to the sister division that could not copy them, and waiting for the result.
Notes
Taylor Randall and I examined prices in the mountain bike industry Taylor Randall, Karl Ulrich, and David Reibstein, "Brand Equity and Vertical Product Line Extent," Marketing Science 17, no. 4 (1998): 356–379. The study uses mountain bicycle price data to estimate brand premiums.
the production system that Taiichi Ohno and Eiji Toyoda built at Toyota Jeffrey K. Liker, The Toyota Way: 14 Management Principles from the World's Greatest Manufacturer (New York: McGraw-Hill, 2004), for the system's history and principles.
Gabriel Szulanski studied the transfer of proven practices Gabriel Szulanski, "Exploring Internal Stickiness: Impediments to the Transfer of Best Practice Within the Firm," Strategic Management Journal 17, Winter Special Issue (1996): 27–43. Causal ambiguity, the recipient's absorptive capacity, and arduous relationships between source and recipient were the main predictors of stickiness.
Nicholas Bloom and his colleagues randomly assigned months of management consulting Nicholas Bloom, Benn Eifert, Aprajit Mahajan, David McKenzie, and John Roberts, "Does Management Matter? Evidence from India," Quarterly Journal of Economics 128, no. 1 (2013): 1–51 (about 17 percent productivity gain in the first year). The nine-year follow-up is Nicholas Bloom, Aprajit Mahajan, David McKenzie, and John Roberts, "Do Management Interventions Last? Evidence from India," American Economic Journal: Applied Economics 12, no. 2 (2020): 198–219.
Linda Argote and her colleagues found in shipyard production data Linda Argote, Sara L. Beckman, and Dennis Epple, "The Persistence and Transfer of Learning in Industrial Settings," Management Science 36, no. 2 (1990): 140–154. In Liberty ship production, knowledge depreciated rapidly enough that cumulative output overstated the persistence of learning.