Software, Full-Stack, and Sumo Startups

Plus! Google Paying Publishers, Quant Crowding, Compare and Contrast, Stimulus for the Dead, Social Credit, more...

This is the usually-once-a-week free edition of The Diff. Welcome to the 1,212 subscribers who signed up since last week’s issue. A record! Issues this week:

Coming up next week for paid subscribers: two company deep-dives, one on how Microsoft defined software economics before flipping them, another on Disney as a meatpacking company for culture, and a piece on the 1.5th wave of Covid-19.

In this issue:

Software, Full-Stack, and Sumo Startups

What a company does well is not necessarily what it gets paid for. Nike and Coca-Cola are in the business of producing and marketing emotionally-resonant 30-second films, as well as running a global logistics system. They get paid when, as a side effect of this, somebody buys a very marked-up pair of shoes or can of carbonated high-fructose corn syrup with a splash of water and some food coloring.

The challenge, sometimes, is not creating a valuable product, but creating a valuable product and capturing the value. Government contractors run into this issue all the time: building a better website for filing unemployment claims would make millions of people better-off, but building a website that’s barely functional but technically meets the spec pays just as well and is a lot easier.

Or, to take another example: the economies of Singapore, Dubai, and Atlanta can only function because of air conditioning. The industry has created trillions of dollars in value, but that’s mostly captured by the buyers, not the sellers. Sometimes not even the buyers: when department stores started adding air conditioning, the first department store to do so benefited from incremental customers and sales. But once every store had them, market share returned to normal; it was just a cost of doing business. A benefit to the customers, but not to the owners.

When capital is expensive, that’s a fact of life. When it’s cheap, it’s an opportunity for a company to expand from a narrow intermediate product—something bought by companies that sell to end users—to a full-stack company whose competitive advantage stems from that intermediate product, but whose actual business directly interacts with the end user.

There’s a set of hypothetical software companies where the product can be built, but the business is hard. Not all of these are good businesses, but trying them out is viable when capital is abundant. WeWork is the classic example of this dynamic; the company’s existence showcases the possibilities, the company’s results highlight the risks. In my WeWork writeup, I argued that they’re trying to be a branding, marketing, and software company that works with landlords—but that it’s easier to sign leases than to sell vague services to managers. (One rule of enterprise sales: never pitch a decisionmaker on software that makes their job obsolete.)

Building a product is an exploratory process: it might be valuable, but not valuable enough to create a sustainable business, or it might be at that agonizing midpoint where it’s worth building, worth using—but too expensive to market. (This happens a lot in fintech, HR software, and other categories: “We replaced the expensive and inefficient process of spending $500 in sales-related overhead to close a deal with a clean and scalable model where we spend $1,000 on AdWords instead.”) In WeWork’s case, the only way to test it was to build it, and once it was built the only way to break even was to scale absurdly fast and hope that the long-term unit economics worked out.

Other companies fit into this model. Oyo, for example, is theoretically rebuilding the hotel industry in an asset-light way: they have an operating model and a demand harvesting system. In practice, they guaranteed booking levels to add hotels to their system. This is economically pretty close to owning the hotel outright. And a levered buyer of hotels is in a tough position in 2020.

There’s a happier version of this model, though: most major hotel chains in the US don’t have much of a real estate footprint at all; there may be thousands of hotels in their system, but the vast majority are independent. Hyatt is somewhat of an exception; they do keep hotels on their balance sheet, though not the majority. It’s a very useful tool for a smaller chain: if Hyatt thinks one of their licensed locations is going to switch to another chain, but that it’s underperforming for temporary reasons, Hyatt can swoop in, buy the property, revamp it, and sell it. In pure return on equity terms, this is not a great deal: it’s hard to beat the ROE of owning a logo, a website, and a loyalty program, and letting somebody else handle the annoying grunt work of operating hotels. But strategically, it’s a great idea; hotel chains' loyalty programs need network density. They want to have several locations in every major city, so they can always find an appropriate room for their most loyal customers, and being able to buy properties to keep that network strong is valuable.

Uber and Lyft operate in a similar way. They make nice apps, and they can choose how much margin risk they offload to drivers. In practice, they keep a lot of that risk themselves, even though it’s a disaster for margins. The expensive part of every ride-sharing business is keeping drivers in the network, and one of the best ways to do that is to make drivers' short-term compensation more predictable than their fares. When the industry is brutally competitive, as it was up until roughly last year, that means that margins suffer. But when the industry decides to play nice, the upside accrues to the same companies that were taking margin risk before.

This is a very old model: you could view Walmart as an IT and logistics company that found it easier to sell toothpaste and toys to consumers than to sell margin-accretive services to thousands of store owners. But it’s getting more common, for a simple reason: building a prototype of a software product is incredibly cheap in most domains. One of the theses Y Combinator was founded on was that pre-launch, being part of a startup was roughly as expensive as being unemployed. And there’s an ecosystem for funding tech companies as they scale. The amount of VC funding has been rising steadily, and returns are skewed by a few positive outliers, so any fund that doesn’t have a specific size mandate is actively looking for companies that can absorb a lot of capital as they grow. The best way to get more capital is to move from a capital-efficient business to a capital-inefficient one, so there’s a strong incentive to pivot in this direction.

The incentive is sometimes too strong. Some companies go beyond the “full-stack” model to what I think of as the “sumo” model: raising an intimidating amount of money just to scare off everyone else. The sumo model does prevent one failure mode for startups: the situation where every time Company A raises a round, it validates the model and lets Company B raise more, which forces Company A to burn through their marketing budget faster and raise an even bigger round, and so on until the entire space is over-capitalized and everyone’s assumptions about long-term unit economics are implausibly optimistic. It’s an easier strategy to try when capital is abundant, but it’s a harder strategy to pull off; the bar for “an absurd amount to invest in a company that just does X” keeps going up.

The surplus of capital that fuels this will probably last a while. The argument for why is beyond the scope of this post, but the short version is that the world is aging, which increases the demand for savings, but productivity growth is declining, which reduces the need for investment. Since the cost of capital is determined by the supply of savings relative to the demand for investment, they’re lower in a slow-growth world.

All this is a symptom of more capital and fewer great ideas, but it’s not necessarily a problem. Societies run into problems when all the most ambitious people want to do exactly the same thing, whether that’s banking, consulting, working for the government, or theorizing about physics. In startups, there’s an exit valve: a founder thinks “For someone with big dreams like me, there’s no choice but to found a tech company in Silicon Valley,” and a decade later that same founder is running a hotel company, a cab company, or a grocery store. Overfunded startups and VC strategy drift are just society’s way to launder talented people into the other 90% of the economy.

A Word From Our Sponsors

Covid-19 has accelerated cultural and business changes around work-from-home by a decade or more. Huge numbers of employers are letting office leases expire. If you haven’t already gotten the “you work from home now, permanently” memo, you likely soon will. This is a good news / perplexing news situation. The good news: you can live anywhere you want - including the suburbs or the country. You can trade your current postage-stamp home for a much larger one, at a similar price. You can have a half acre… or 50 acres, your kids can play in the backyard (or forest), and your spouse can have a garden.

More space, low crime, and low cost of living are great…but what are unknowns?

How do you weigh your options?

How do you get the most bang for the buck, and avoid expensive mistakes ?

Buy “Escape the City” (5 days left in Kickstarter), a book on the hows and whys of relocating from the city to the country, by a white collar professional who did it. Get Travis Corcoran’s seven years of experience on everything from house hunting and high speed internet, to getting along with neighbors, small town services, gardening, hiring contractors, raising animals, and heating with firewood.

Elsewhere

Google Paying Publishers

Olive branch, or just another example of commoditizing the complement? Google is planning to pay publishers for high-quality content. Google has previously been critized by online media outlets for the sin of providing them with free traffic. In Europe, there have been successful efforts to make them pay, so this is close to inevitable. What’s not inevitable, though, is the terms on which Google pays. If they can find a way to subsidize quality and penalize clickbait, more news-reading sessions will start with Google searches rather than Twitter or Facebook.

In semi-related non-news news: Facebook has cleverly started warning users that they’re sharing dated headlines. A very popular Facebook prank is to post a story like “Biden withdraws from Presidential race” without noting that the story came out in 1987. This is mostly a prank, but the problem sometimes has real-world consequences.

National Champions

European courts give US Internet companies a hard time for many reasons, but one of those reasons is protectionist: there are a lot more US-based Internet companies with high European markets share than the other way around. It’s an uphill battle: it’s very hard to compete with companies whose home market is the world’s biggest economy, and whose default language is the rest of the world’s default second language. But it’s not uncommon to try.

Two recent headlines to that effect:

Quant Crowding

Bloomberg highlights the risk that everyone is using the same datasets to trade the same stocks, and showcases some of the “market-for-lemons” dynamics in selling datasets. If you don’t disclose how a given data product works, buyers have a hard time knowing if it’s priced-in. But if you do disclose how it works, they have an easy time copying it.

One thing Bloomberg doesn’t emphasize, but that raises the stakes in an entertaining way: when a given signal gets more crowded, returns go up: people who’ve been using the data for a while are faster to react to each new datapoint, but the market catches up with them quickly, giving them a rapid profit. But the net result of this is that every data-driven trade ends up being crowded, and since the historical sample size is small, data users tend to learn a signal’s hit-rate the hard way.

I wrote about this in more detail in my Alternative Data Primer, especially part two.

Compare and Contrast

China’s is the world’s top exporter, and gained share during the crisis. “At the start of February, [China] made about half the world’s supply [of N95 masks], 10m a day. Within a month, output had increased to nearly 120m.” Meanwhile, in the US, an entrepreneur imported defective masks and paid gig workers to remove the “Medical Use Prohibited” labels so he could sell them to hospitals. Fortunately, one area where the US is at parity with China is its tech panopticon, though ours is voluntary: the ProPublica investigation of the mask situation researched its subjects on Facebook and LinkedIn, and tracked the flow of funds on Venmo(!). Taking notes on a criminal conspiracy is one thing, but broadcasting it to the entire Internet is another level.

Working Capital

Pipe, a company that essentially securitizes SaaS contracts, has raised another $60m. This is a very exciting business, because SaaS companies have ridiculous balance sheets: a long-term contract is basically equivalent to lending the buyer money with somewhat complicated payment terms. Bizarrely, SaaS companies raise equity (very expensive capital) but have most of their economic balance sheet invested in credit (very cheap capital). Pipe is helping to fix this.

They’re not the only ones. Google is planning to lend merchants money through Google Pay. This is a notable move because Facebook’s Jio investment was predicated on helping Jio expand in online payments. India’s banking system is a mess, and the entry of two well-capitalized US tech companies will be a significant benefit.

Stimulus Checks for the Dead

A few days ago I highlighted the Good News Pitched As Bad News story that the US has a mostly worthless stockpile of hydroxychloroquine—if your country isn’t wasting money on potential cures that don’t pan out, it’s not spending enough on cures. Similarly, if your emergency stimulus program doesn’t send money to the deceased, it’s not aggressive enough. While the US’s response to the pandemic itself leaves much to be desired, the response to macro side effects has been pretty decent—more so on the monetary than fiscal side, but ok marks all around. One reason the Fed can move faster is exactly this kind of story: it’s a lot harder to have adverse selection in buying tradable assets than sending 160 million small checks. (If nothing else: the Fed’s aggressive actions kept many companies out of bankruptcy by allowing them to roll their debts; a metaphorically dead company that gets a cash injection can stay alive, while the non-metaphorical analogue doesn’t apply.)

Social Credit

Wired profiles a new specialty lender: a company that offers credit cards to high-income “influencers” who don’t otherwise qualify for much credit. This is a very niche product, but a good place to start: there’s a tiny number of YouTube stars who a) can’t get huge credit lines, but b) can pay huge credit card bills. But these users have an audience, so they’re a decent distribution platform for a more conventional credit card.

More Browser Commoditization

A few days ago, Apple quietly announced that they’d allow iOS users to select a new default browser and email app. Now, Apple is helping developers move extensions from Chrome to Safari. Together, these stories imply that Apple views the browser as a much less strategic property than they previously did. This year’s WWDC seems to have the general theme of switching from apps to functions; perhaps it’s a sign that Apple can monetize those functions better than it can monetize the apps themselves.

QVC Returns

TikTok is experimenting with ‘live commerce,’ TikTok-sized infomercials. In true TikTok fashion, the purported scale is absolutely enormous: it’s “projected to become a 961 billion yuan ($135 billion) industry this year, according to Chinese data provider iResearch.” Home-shopping has been a surprisingly durable industry in the US; Qurate Retail, which mostly consists of QVC and the Home Shopping Network, did $13.5bn in revenue last year, most of which is now e-commerce. TikTok’s product is almost perfectly designed for this model: an algorithmic feed can slowly guide viewers from entertainment to demand-generation to the infomercial itself, then close the sale.