Byrne Hobart
I’ve written about lots of technology and financial trends in my life, but there’s one I’ve covered more thoroughly than the rest: I’m charging for The Diff.
Readers will be delighted to know that today marks the last day of nonstop promotion, but also the last day of mostly-free posts. Starting on 3/2, I’ll be writing at least three per week, only one of which will be available to non-paying subscribers.
If there’s one way to sum up who reads The Diff and why, it’s this: The Diff is for people who want to understand the future, and want to act on that understanding. Finance provides metaphors and live data; technology provides the underlying trend.
It’s a commonplace observation in finance that when markets go down, all correlations go to one. This makes perfect sense from a Minskian perspective: investors feel safe levering up when they expect economic fundamentals to stay healthy, but the more they lever up, the more any one fundamental change can break the entire system. But it’s also a broader truth: “Black Swans”—extreme events that blow up the assumption of a normal distribution—really only happen if a lot of seemingly-unrelated things are serially correlated. The reason models of the 2016 election underrated Trump was that they underrated the chance that the polling error could go in the same direction in every swing state. The reason credit default swaps on real estate-backed structured products were cheap in 2006 was that most investors didn’t realize that cheap credit had raised the correlation between housing markets, and that asset selection raised the correlation further within each structured product.
The outcome of this is that every technology entrepreneur and investor needs to care about the global economy. The trends you’re counting on—free flow of capital, goods, information, and people—are dependent on a set of conditions that might not hold. And they’re correlated. Most useful macro discussions revolve around China, since it’s the axis around which the world economy revolves. But it’s also the lynchpin of the global electronics supply chain. Any plan that presupposes continuous improvements in smartphones and continually cheaper components assumes that China keeps on growing at the same pace, and remains tightly-coupled to the US, Europe, and emerging markets.
Manufacturers are realizing, and consumers are about to realize, that supply chains offer their own sort of leverage, with their own potential for a “Minksy Moment” in which a disruption in one place causes cascading chaos everywhere else. Coronavirus might be a minor speedbump (editor’s note, early 2021: it was not), but it, or something like it, will eventually force a wholesale change in the pace and nature of globalization.
The meta debate in American politics is a question of terminology: should we say “Except in 2016” or “Until 2016?” The meta debate on globalization is harder to call: is it “Until Trump,” “Until Coronavirus,” or something else? But it’s an important question. Long supply chains produce operating leverage, and like any other kind of leverage it can’t go up forever.
Every Platform Business Looks Like Local Politics
I’m reading Steven Levy’s Facebook: The Inside Story—recommended—and one of the things it elucidates is exactly what kind of politician a tech CEO has to be. The easy analogy is to say that Mark Zuckerberg is a head of state: he affects the lives of billions of people, has regular chats with senior political and religious leaders, gives speeches that get livestreamed around the world and written up by major media outlets. But Zuck is basically required to be nonpartisan. (Like any important person who doesn’t choose a side, this means most people tend to think he secretly supports their enemies.)
Moreover, the problems Facebook has to solve are not the cosmic, inspiring ones. They’re pretty trivial, technocratic issues, mostly dealing with competing interest groups rather than competing ideologies. Zuckerberg isn’t an emperor, or even a prime minister; he’s the world’s most competent and most overworked mayor. The question of how much data an advertiser should be able to collect and use, and how they should be able to use it, isn’t a question with the same scope as a treaty or a labor law; it’s a lot more like deciding where a sewage treatment plant goes or choosing which bus route to cut.
This doesn’t mean that the company lacks ideology. Clearly, Facebook has always had strong views on the importance of openness, connection, and communication. But that ideology get less important over time as more people depend on FB for fairly prosaic needs, like entertainment, keeping up with friends, and selling products. Salt Lake City was also founded by people with very strong, very distinctive beliefs, but the mayor of SLC probably thinks about the same basic issues and tradeoffs as the mayors of Modesto or Yonkers.
This is not unique to Facebook:
- Google wants people to search, so they need an accurate search engine. But they also want revenue growth, so they need to populate some searches with ads. As it turns out, they discovered a revenue-and-utility-maximzing solution for pricing ads, with auction results partially determined by quality rather than price. But there isn’t such a solution available to the question of ad load. Once they’re dominant, they know they can increase ad load while keeping quality constant to get continued growth, but it makes companies reluctant to build a business on organic traffic. And that slowly chips away at the utility of search as a business.
- Twitter wants to be a mass-communication platform, but as it turns out, some people want to communicate certain things (conspiracy theories, invective, death threats) that other people really don’t want to hear. At one time, Twitter looked like a communications protocol—the default way to send a device-agnostic, human-readable status update. They eventually realized that if you build a protocol and monetize with ads, every other app that uses the protocol is strictly superior because it doesn’t have ads. So they made that impossible, gradually and then suddenly.
- Netflix wants to offer a bundle of shows, some of which are appealing in theory and some of which people actually love to watch. They want to accumulate data they can use to recommend shows that drive usage (or, more precisely, that predict low churn rates—but the simplest way to keep someone from unsubscribing in a given month is to get them in the habit of watching every day). Netflix definitely doesn’t want anyone else to have access to their data, or they lose their edge in bidding for talent. They’re finally offering limited popularity data, but a top-ten list doesn’t tell producers and actors the two things that really matter: the shape of the popularity distribution and the marketability of the show.
- Amazon retail wants as many products as possible, as cheaply as possible, and they want to capture as much of each transaction as they can. This selects for low-cost merchants, and tempts merchants to skimp on quality. The decision to start with books was smart in many ways—books have searchable text and pre-formatted metadata, perfect for bringing a product online—but one other reason it worked well for Amazon was that counterfeit books are hardly worth the bother. That’s not so with electronics. And a bootleg book is less likely to catch on fire when you plug it in.
- LinkedIn needs to serve the needs of recruiters (who provide revenue) and regular employees, who motivate the recruiters. In particular, they need to avoid the adverse-selection problem: if the quantity of bad inbound recruiter contacts is a function of how desirable the employee is, LinkedIn’s userbase will tend towards mediocrity. They can tamp down on this by making it costly for recruiters to spam, or by giving people non-jobseeking reasons to use LinkedIn, like the news feed. Dating sites face a problem with exactly the same shape, and they’ve learned to use some of the same solutions.
But while it’s something tech companies have in common, it’s not something other industries have to deal with. Coca-Cola, P&G, GE, GM—big companies have some political tradeoffs to make, but they’re more narrowly-defined, and a lot more of them are simply monetary. GM needs to balance the interests of drivers, dealers, employees, suppliers, and shareholders, but basically all of those groups either want to get more money or spend less money, so GM has the comparatively simple option of building a really good product. But if Facebook builds a really good product for spreading news, they’ve built an exceptional product for sharing fake news; if Google has a good way to surface information, it’s also a good way to surface misinformation.
This is the problem of platforms: they build a place, rather than a business, so they can’t enapsulate complexity by making everything transactional. The upside to this model is that it leads to long-term, high-margin growth. Building a platform means homesteading a new economic frontier, and then running it as an idealized government that taxes at the Laffer maximum and most lets participants alone. (This might explain the phenomenon Mike Gibson recently observed: big tech companies are much more influential in national politics than they are locally. It’s because their local-politics skill is completely absorbed by their business.) Unfortunately, new property rights require an immense and tedious investment in codification. When you’ve solved product, sales, marketing, and operations, the only thing left is politics, and by its nature, politics doesn’t get solved.
Elsewhere
I have a new piece in Marker on how the IPO transformed from a coming-of-age event to the onset of corporate middle age.
You shouldn’t trust every organization to have the correct amount of paranoia about Coronavirus, but you should pay close attention to the ones that have good real-time data. Facebook is cancelling the in-person part of their F8 conference in May. While FB doesn’t have much coverage in China, other than via VPN, they can track behavior in other countries, which means they can see clusters of people who suddenly stop going to work. FB and Google might be the non-Chinese companies best-equipped to identify where Coronavirus is spreading, and how bad it gets once it hits.
Fidelity is spinning off a company that helps consumers decide which companies get their financial data, and how it’s used. Consumer financial data has become an extremely important data source for investors, particularly in retail, restaurants, and e-commerce, and this company is at least trying to eliminate this data source. A product that lets consumers choose who gets their data is, in practice, a product that lets consumers choose not to share their data: if you ask them to share in the fine print, they shrug; if you give them a yes/no option, they tend to say no.
While the first-order effect is just less price efficiency and more volatility, it has a bigger long-term impact: traders who have daily transaction data can afford to take on more risk, since they can track how their thesis plays out intra-quarter. So losing this data wouldn’t just exaggerate short-term swings in price; it will tend to compress valuations for the fastest-growing companies because it’s harder to predict the bear case. Right now, it’s only available to Fidelity customers, but it will be more widely available later this year.
Eric Schmidt has an NYT editorial on the need for the US government to support national champions in tech in order to beat China. It’s one of those problems that’s too late to solve when it’s obviously a problem: very few Chinese Internet brands have made meaningful inroads in the US, with the recent (and huge) exception of TikTok. But they’re starting to expand outside of China in other places.
What’s really interesting about Schmidt’s piece is that big tech is an issue of uncertain partisan valence. Republicans like big businesses more than Democrats, but they don’t like industries that donate almost exclusively to Democrats. Democrats are predisposed to like people who thumb their nose at the establishment, but Internet companies did that so well that they became the establishment. Now CBS (est. 1929) is a plucky underdog compared to YouTube (founded 2005). Republicans think social networks tried to rig the election for Hillary; Democrats think they let Trump run away with it.
Big Tech seems like an unassailably progressive player in politics. But union members and free-trade skeptics were heavily Democratic through the 90s and at least leaned that way until 2016. Ideally, tech companies would like the national-champions argument to be obvious and nonpartisan, but in practice if it’s not something one party is jazzed up about, it’s something neither party cares about.
Here’s an article from 2008 on why Zappos, a low-margin e-comm company, offers new employees a $1,000 bonus if they quit. It’s 2020, capital is cheap, and SaaS businesses have higher margins (in theory!) than online shoe stores. So the pay-to-quit offer is a bit better:
This is a clever way to amortize the cost of hiring good employees over more transactions, and it’s also a way to reduce one of the drags on startup performance: that employees who have decided to quit, but haven’t pulled the trigger, aren’t working as hard because they’ve already mentally forfeited their unvested options.
The biggest hotel chains are basically online travel agencies: they send leads to hotels, and collect a piece of the topline. They borrowed that last bit from Microsoft, which had some infamous royalty deals that required a fixed payment per computer from PC manufacturers, whether or not that computer had Windows. It’s a good way to get the most out of a dominant market position. The beauty of this model is that it requires little cash but generates a lot, so hotel chains can tactically deploy their cash to close deals, either buy buying hotels or by lending money to hotel developers.
Then, there’s Oyo. Oyo is trying to build a hotel chain in the US, following a model that has grown fast in India. Unfortunately for them, they’re a few decades behind the US chains, so the remaining indepdent hotels are either a) pretty bad businesses, or b) not interested in joining a big chain. Oyo’s solution is to offer them revenue guarantees, which has the perverse effect of making Oyo a company whose fundamental business is low-risk and asset-light, but whose operating results actually match that of an asset-heavy hotel owner with operating risk.
Another company doing the same thing—taking operating risk in an asset-light model—is Etsy, which is now running ads for high-volume sellers and taking a mandatory commission when those ads lead to sales.
Sam Altman channels the Peter Thiel argument for zero-to-one monopolies (and indirectly another Thiel-ism: that it’s easy to get early employees with big option grants, and easy to get late-stage employees with stability, but starting at about employee #20 you have to be the only place in the world to work on solving a particular problem). Most interesting in this piece is the footnote: that the right headcount for a startup may shrink drastically as a result of rising cash comp at bigger companies. Something to meditate on, especially as more service providers eat up the ops layer and replace in-house products with pay-per-use API calls.