Refactoring Restaurants

Plus! 401(k)s With Chinese Characteristics; Middlemen and Leverage; Concentration; A Mixed Trucking Recession; Deglobalization in Autos

Programming note: I'm switching up the schedule slightly, and will be doing free posts on Monday instead of Friday. Friday is apparently the single worst weekday to send a newsletter, so this has probably been bad for Diff growth the entire time.1 Coming attractions next week: we're talking about the car industry, as a force for economic development, as a financial engine in two different ways, and as an industry going through a massive transition that's hardly its first big struggle for relevance.

Refactoring Restaurants

Yesterday's subscribers-only post on the flavorings business ($) talked a bit about how food in the developed world is mostly a manufacturing business, especially if you count it in terms of calories affected rather than revenue earned (there are some high-margin service elements in there, but a lot of them start with more mass-produced ingredients). The most factory-like food experiences most people interact with are fast-food restaurants, which rigorously manage themselves around revenue per foot and revenue per worker-hour. The big franchises have, over the decades, gotten very good at getting the maximum revenue out of both—next time you're in a fast-food restaurant, think of all the worker-hours that went into every detail of the menu, layout, pricing, and consider how it's amortized ~$10 at a time. Entire lifetimes worth of effort have been devoted to the question of how to increase your probability of adding McNuggets to an order.

One of the perks of the newsletter/recruiting/consulting/investing business is having chats with interesting founders, and I recently caught up with one of the founders of Local Kitchens, which led to a fun rabbithole on how food economics work and how they're changing.

The general trend for the restaurant industry has been a push and pull between decentralization and recentralization, with the latter usually winning over time. Chain restaurants are 77% of total restaurant visits in the US, and have grown over time. And within the independent restaurant space, there's been consolidation at the backend, where the food distribution industry has consolidated through both public company and private equity rollups and on the frontend, where more demand gets funneled through review sites, or directly fulfilled by delivery companies.

Consolidating the industry through apps and delivery makes intuitive sense to people in the tech sector. On the app side, it's natural for someone who codes for a living, or who is familiar with software margins, to aim for that layer of the business.

And delivery has appeal for more mundane reasons: white-collar workers who either put in long hours or get to the office fairly late believe that delivery is the way of the world. As Matt Levine has pointed out, there are even league tables of which banks have the most generous meal allowances, and, inevitably, that these league tables are being gamed by savvier banks.

But as it turns out, that's not the standard pattern: takeout is roughly twice the size of delivery in the US, and the restaurant industry has evolved around serving commuters on their way home, not just people dining out or staying in. Online ordering platforms work fine for takeout, too, and that's creating a problem for restaurants: their rent and labor force is based on serving more of an in-store dining market, but if in-home dining is gaining share, then the kitchen is constrained by square footage while the rest of the restaurant is a big sunk cost that no longer produces a great return.

Some restaurants will be able to survive that squeeze, with a sort of barbell distribution: Eleven Madison Park is basically a live theatrical experience with food included, where the waiters tell jokes and stories in addition to putting on a few prop-based skits that result in appetizers. At the other end, fast food restaurants with playgrounds partly act as a bit like Breather for daycare: half an hour in the ball pit for the cost of a few Happy Meals is a great trade.

In the middle, where the restaurant has atmosphere but you wouldn't go out of your way to go there, and where the food is the main attraction, there will be a slow bleed. It will be slower in cases where the food doesn't travel well, but consumption patterns can drive cuisine rather than the other way around.

Which is where the Local Kitchens opportunity comes in: they license brands from and cook food for multiple restaurants from one kitchen for takeout and delivery orders. By getting rid of most of the seating, they can pack a lot more kitchen space into a given location, and the marginal cost of that kitchen space can be quite low. At one level, this looks a bit like the MrBeast Burger model, or the many, many delivery-only wings brands, where a few extra menu items are produced by an existing kitchen at the cost of some extra square footage and labor and a bit more operational complexity.

But doing this as a franchise model runs into problems; the successful franchise restaurant models are religious about quality, because they're trying to sell a consistent experience. (Raising quality levels matters less than reducing the standard deviation of quality, as long as the price is right.) If what they're offering is just a brand name and some instructions, the restaurants that select in are the ones with spare capacity in the kitchen, i.e. the ones that are already having trouble doing enough business with their existing menu items. Local Kitchens is able to extend the reach of local restaurants that are already popular, so it's avoiding the adverse selection of the virtual kitchen model—in fact, it's reversing it, , instead of owning the brand and outsourcing everything else, they own every part of the process except the brand.

One way this model gets interesting is when you think about the catchment area for brand recognition versus convenient access. National chains make sure both of these are, basically, "everywhere" and "everywhere": they market nationally, and they have locations in every place people are likely to drive by.2 To a first approximation, the good restaurants in a city are either located "downtown" or "in the middle of nowhere," depending partly on circumstance. And for a restaurant in an inconvenient area, this means that a lot of their local brand recognition is wasted. Satellite locations for pickup mean that a restaurant can serve the entire area where people know about it.

This has another benefit: coverage for ethnic cuisine is largely a function of population density, which means that suburbs tend to have it based on luck. But the name recognition for more obscure cuisines is rising, partly thanks to Instagram. Since good-looking food is ‘grammable, and obscure food even more so, there's a much larger population than there used to be that a) really wants to try Ethiopian, or Filipino, or Turkish food, but b) doesn't live within an hour of a place where they can get some. If a given neighborhood has 0.5 restaurant's worth of demand for that specific cuisine, then a restaurant won't thrive there—but if it just needs a few square feet of food prep space, it can.

And that also means that the long tail of ethnic food can get even longer, since the minimum viable name recognition is going down over time. Within my lifetime, Thai and Vietnamese food went from somewhat unusual to being pretty standard in any mid-sized city, and one element of early-80s yuppie stereotyping was that they had a then-bizarre preference for vegetarian fare and sushi. The share shift from shrinking the fixed cost for restaurants is not just between different kinds of known foods, but between currently popular ones and obscure ones. And this means that the model is a way to bet that the American palate has not reached its limit.

There are other fun aspects of the shared-kitchen model. Pickup and delivery are both cases where time matters, and not just the absolute length of time: variance and predictability are important, too. Doordash has written about the lengths they go to to reduce surprisingly long delivery times, without affecting the average. But they're limited by the restaurants they work with. If a restaurant is packed one night, and the kitchen has to choose between a low gross margin order for Doordash or keeping a table full of customers who are on their second round of drinks happy, the high-margin option is the one they'll take. A full-stack restaurant operator can log and monitor every step of the process, so estimates are clean and so shared resources get shifted around as needed. A delivery company can do some of this—rearranging drivers so the next available one goes to the meal that's already ten minutes behind schedule even if the cost-optimal approach is to have another driver take it a few minutes later—but any company that's trying to deliver a predictable outcome on top of an unpredictable process is eating lots of deadweight loss in the process.3

Consolidating kitchens has yet another side benefit: aside from increasing the sample size of restaurants and thus lowering the variance, it has higher employee utilization and less food waste. Food and labor costs are both roughly a third of restaurant operating costs, and Toast says the average profit margin from restaurants is usually 3-5%; a three-point improvement in either food waste or labor waste adds a point of net margin, and getting both of those makes a restaurant 40% more profitable. Low-margin businesses are tough, but if there's a way to improve them it creates a lot of operating leverage.

A shift to more dining in home would normally mean that the restaurant business gets more consolidated, as bigger companies get more effective at maxing out kitchen margins, hammering down delivery costs, and sprinkling locations optimally to get takeout business. But it doesn't have to work that way—it can also mean that the minimum viable restaurant location is a lot smaller than it used to be, and that small, beloved establishments that were too undercapitalized to expand can start reaching more customers. Industries usually get more concentrated over time, but as businesses from Stripe to Substack to Shopify have demonstrated, there's also a lot of money in enabling smaller companies to scale and thrive by giving them access to tools that only the largest could previously afford.

Thanks to Jon Goldsmith at Local Kitchens for talking me through the model. For earlier thoughts on the restaurant industry, see this Diff writeup of Doordash.

Diff Jobs


401(k)s With Chinese Characteristics

Global asset managers are excited by the prospect of China's tax-advantaged defined-contribution pension system, which is now rolling out $, *FT). Current contribution limits are just under $2,000 per person per year, but multiply that by China's sizable and growing (and aging) middle class, and the potential market is huge. As The Economist points out in its coverage of the same plan ($), China has long aimed for a three-part system, with government, company-provided, and individual pensions channeling people's savings. Individual savings have been a small fraction of the total, and corporate pension funds haven't seen much uptake. Waves of speculation in real estate demonstrate that China's savers are looking for somewhere to stash their funds, and tax-advantaged accounts that can't be accessed until retirement would be one way to stabilize the country's stock market.

In the US, the 401(k) took a while to get started, but eventually grew into a multi-trillion dollar industry, and probably contributes to the US's fairly high rate of equity ownership. And that had second-order effects: when the stock market became a vehicle for middle-class savings rather than upper-class speculation, it also became more politically tenable to regulate accounting abuses. Once the claim shifts from "speculators made too much money" to "everyday people lost too much money," regulations tend to focus on transparency rather than punitive measures. And that would be a healthy development for China's market, too.

Middlemen and Leverage

It's fair to describe the commodities trading business as financially-engineering an oil major in real time: buying oil and shipping it in cases where the natural buyer of crude, the best shipper, and the ideal refiner are not part of the same organization. This business can do well, and it's put a few people on the Forbes global billionaires risk, but it has one of the same problems the oil industry does. As one energy executive once put it:

The good Lord didn't see fit to put oil and gas only where there are democratically elected regimes friendly to the United States. Occasionally we have to operate in places where, all considered, one would not normally choose to go. But we go where the business is.

Trafigura, an oil trader that came to Russia late and then expanded aggressively in 2014 when other companies pulled back, is in exactly the sort of situation that can ensue when companies do business where "one would not normally choose to go" ($, WSJ): they're busily trying to unwind long-term agreements to purchase Russian oil. That's both driven by regulatory concerns (Trafigura can comply with sanctions on paper by using offshore units, but could risk secondary sanctions) and for economic ones (buyers of Russian oil are getting massive discounts). Asset-light companies are usually more nimble than the ones that own physical energy assets, but a long contract with an international pariah is economically pretty close to owning their oil directly, at least in the worst case scenario.


The Economist highlights how much big tech companies are dependent on a small number of business lines for most of their revenue and growth ($). The usual pattern is that a company will experiment with a few lines of business, find one that's wildly more valuable than all the others, and bet everything on that business for a while. The more disproportionate it is, the better the underlying business—how likely is it that one founder will discover two or three unicorn-level opportunities while running the same company? But after a while, things spread out a bit more as companies expand into adjacent businesses. There's a little double-counting here: yes, Apple derives a lot of profit from its Google deal, but Google also derives a lot of expense from its Apple deal, and for the deal to go bad on both sides you'd have to assume something terrible and unforeseen happening to the search business generally. These companies also benefit from the venture business:

According to a report by Bridgewater Associates, the world’s largest hedge fund, of all the money invested in early-stage companies about a fifth is spent on cloud services, a market dominated by Alphabet, Amazon and Microsoft. Another two-fifths goes on marketing, which in the digital realm is dominated by Alphabet, Meta and, increasingly, Amazon. Bridgewater estimates that, all told, around 10% of the total revenue of Alphabet, Amazon and Meta is derived from the startup ecosystem. That is the equivalent of $84bn each year.

It's a good reminder that these companies are not just generalized bets on growth, which will automatically add revenue at ~5x GDP growth or some other multiple. They're a set of discrete bets on specific trends, and buying them means underwriting the risk that those trends will slow or reverse.

A Mixed Trucking Recession

Freightwaves has been warning about a slowdown in the trucking industry recently, but this earnings season has been good for trucking companies so far: J. B. Hunt, Knight-Swift, and Old Dominion Freight Line all beat estimates. One theory is that larger companies with long-term contracts are doing better, while the smaller ones that do short term work are absorbing almost all the drop in demand. Better for investors, but much worse for the economy: these smaller companies are undercapitalized relative to the large players, and there are 3.5 million truck drivers in the US, so a hit to their earnings would have a meaningful impact on unemployment numbers.

Deglobalization in Autos

In yesterday's post ($), I alluded to the possibility that Russia might be able to restart its sanctions-hit auto industry by importing parts from China. That could happen, but not any time soon. In fact, Volvo is moving its sourcing away from China, as Covid lockdowns once again shut down factories ($, FT). A world where more manufacturers have multiple sources for the inputs they need, in multiple countries, is a more expensive one for consumers. But it's also one that's a little less likely to get disrupted by natural disasters and political events, not to mention one where global manufacturing and especially manufacturing growth is a little less overweighted to China. Trading cost for lower variance and using that lower variance to earn brand value is not a strategy restricted to fast food.

  1. The free Friday post was, for what it's worth, a completely arbitrary decision that resulted from me choosing a specific week in which to go paid. I've thought for a while that this was a good idea, but there is incremental work for a free newsletter versus a subscribers-only one, and I never got around to it. But I've become increasingly obsessed with the idea of one-off changes that lead to permanent improvements. The go-to example here is that the difference in effort between a lifetime of having a sink full of dirty dishes every morning, and a lifetime of literally never having that problem, is washing exactly one sink's worth of dishes one time. Most other examples are not so trivial, but outside of insisting on writing a fully general solution instead of an ad hoc one, my bet is that most of the time these opportunities for upfront work followed by zero marginal cost benefits are worth taking.

  2. Stay tuned for a future post on the Interstate Highway System as a precursor to the Internet: a Federally-funded project that created a better distribution platform for privately-owned brands, which then offered complementary services. Fast food chains are, in part, a sort of long-running conspiracy to ensure that there are conveniently-spaced and frequently-cleaned restrooms all along the interstates.

  3. "Solve for maximum utilization at all times" and "solve for minimum difference between order estimate and order delivery time" are two different problems, and since solving the first is best for margins assuming some fixed costs, the latter must be worse. Many companies do prosper by making unpredictable processes more predictable—Apple, for example, has done a great job of keeping iPhones and MacBooks available even when various suppliers are shut down, ports are backlogged, etc. (Not always, as this quarter shows.) But it's expensive, and even high-margin companies feel some hurt when they commit to it. For a low-margin company, this can easily mean wiping out all of their profits.