A good mental exercise for CEOs is to periodically ask: if I were building a company that served the same customer needs, would I rebuild the exact company I run today or would I build something totally different? There's usually some level of path-dependency in how companies eventually come to operate, which means that they need to get pruned from time to time. But occasionally the result of that pruning is to conclude that the business is actually in terminal decline and it needs to be demolished and rebuilt as quickly as possible.

The better the CEO is, the worse this choice will look to shareholders. The last time the company now known as Meta Platforms decided to go all-in on a new category, it was growing revenue 88%, with a 42% incremental operating margins. At the time of its IPO, Facebook, monetized primarily through desktop ads and games, looked like one of the best businesses of all time. It was counterintuitive, to say the least, to start cannibalizing that business by shifting more usage to mobile, which at the time had a worse user experience and no monetization, and where ads typically earned far lower profits. Shifting to mobile was, in a sense, a way to bet the company's $80bn market cap on the idea that their desktop business was a local maximum and that mobile was a bigger opportunity.

Which worked out fine for them. Shares returned 25% annualized from the IPO through their next bet-the-company move, the Meta rebrand, which took place in October 2021.

There's a long and sometimes glorious history of this kind of decision. As IBM expanded in the computer business in the 50s and 60s, they'd generally design new systems from the ground up based on what different categories of customers needed. This was a decidedly low-risk approach for IBM: the more influence your customer has on the spec, the more likely they are to buy and continue to use the product. But it was also creating two kinds of waste: internally, it meant that some features were built repeatedly for slight variations on the same hardware. And externally, it meant that upgrading from one IBM system to a more powerful one entailed rewriting most of the code—so IBM's only sales advantage was its customer relationship.

The company responded to this with the System/360 line of computers, which almost all used the same instruction set, meaning that customers could upgrade hardware without rewriting much software. (This was in the very early days of the collective realization that developing software takes longer than people think. Douglas Hofstadter codified this a decade later: "Hofstadter's Law: It always takes longer than you expect, even when you take into account Hofstadter's Law.")

This was not a small project. IBM ended up spending $5B in mid-1960s dollars, over a period of four years. That total commitment was roughly 0.6% of GDP, from a single private company. The equivalent budget today would be $164bn, or $41B annually.

This could have turned out poorly. IBM might have launched, all at once, a series of computers that each didn't quite measure up to what more focused competitors offered at the same price point. Or the schedule could have slipped to the point that the whole project had to be abandoned, which was also a reasonably close call — one of the classic books on how hard it is to manage software projects was basically a case study on the System/360. Fortunately for IBM, two unknown-unknowns roughly canceled each other out: the incredibly favorable experience curve of integrated circuits and the difficulty of managing large software projects managed to stay roughly in balance, and IBM's dominance of the computer industry was solidified until someone else figured out a higher-margin variant on standardizing everything.

There are many case studies besides IBM, of course:

• Microsoft made an aggressive pivot towards the Internet in the mid-90s. (Sample: "Amazingly, it is easier to find information on the Web than it is to find information on the Microsoft Corporate Network. This inversion where a public network solves a problem better than a private network is quite stunning."[1] This move eventually got the company into trouble, but it was trouble that they only could have encountered if it had been the right move. It was also a daring one, since there were only 44M Internet users globally at the time and there were no known business models. You can describe many of Microsoft's best strategic decisions in the last twenty years by asking what the CEO of the company would do if they'd just reread that 1995 memo again.
• Netflix had long intended to be a streaming media provider, with the DVD-by-mail business as a nice wedge to get customers, cash flow, and data. But the only two times to go all-in on streaming were too early or too late. Their stock had an 80%+ drawdown over six months in late 2011 when they launched and then killed a separate DVD-only brand, but retained a separate price for streaming and DVDs. Before that switch, the business was growing 30% annually, and profits were growing faster; in 2012, revenue growth slowed to 12% and profits dropped more than 90%. But this turns out to be the signature of a well-timed bet-the-company choice—new models are purely accretive only when they've been perfected somewhere else, so a company whose big bets aren't costly is a company that's catching up to somebody else's earlier bet that's already paying off.
• OpenAI tried a few different approaches to building AI before settling on large language models as the most promising approach. This Sam Altman profile ($, WSJ) cites earlier projects like "teaching robots how to perform tasks like solving Rubik’s Cubes," and the company both pivoted to focusing on transformers and switched to being more of a for-profit entity around the same time, in late 2018 and early 2019. In this case, there was both a technological and organizational pivot: once they were working on a project that was capital-intensive, a non-profit model wasn't going to work.[2]

You can think of all of these moves as late-stage pivots. When a three-month-old company decides to switch from selling to consumers to selling to businesses, or when it decides to pivot from NFTs to LLMs, it's technically a bet-the-company decision, but there's not a very big opportunity-cost bankroll to bet. When a more mature company does it, though, they're facing a measurable opportunity cost against a harder-to-estimate upside.

As it turns out, there's a sense in which you can't meaningfully bet the entire company on a new model. Some parts will be preserved: IBM used its sales relationships and existing technical expertise in a new way, as did Microsoft three decades later; your Metaverse account, if you ever actually have one, will be seeded with data from whenever you signed up for Facebook; Netflix (like Facebook in its pivot to mobile) initially got customers to sign up by giving them a new way to consume the same content and not directly monetizing it at all; and OpenAI had aggregated lots of talent already, and knew that with the workforce they'd selected for, any bets that clearly weren't paying off would prompt their best people to leave.

Most businesses will eventually face a bet-the-company choice; business models change over time, and sometimes these shifts require a step-function change in what a company does and how it sells its products. The big late-stage pivots tend to look too early from the outside and to feel too late from the inside. They're partly about adapting, and partly about being acutely aware of the risk of leaving the technological/economic slipstream. Even if you never consciously bet the company, you're still gambling, you’re just staying deliberately blind to the odds.

This piece was originally featured on Byrne Hobart's The Diff.