What's Happening in Logistics

Plus! Infohazards, How The Grinchbots Steal Inflation, The Hack Expands, AR and the Physical/Digital Switch, Antitrust and Real Interest Rates

Byrne Hobart Byrne Hobart

Welcome back to The Diff. Here are the subscribers-only posts you missed this week:

Subscribe now to read these and access the full back catalogue. The Diff will be taking a break starting December 24th, and returning on Monday, January 4th, but in the meantime, here are some of the most popular subscribers-only posts from this year:

Subscribe now

This is the once-a-week free edition of The Diff, the newsletter about inflections in finance and technology. The free edition goes out to 17,868 subscribers, up 229 week-over-week.

In this issue:

Programming note: The Diff will be taking a holiday starting 12/24 and extending through the end of the next week. Back on Monday, January 4th.

What's Happening in Logistics

Roughly one eighth of the world economy is dedicated to moving goods from where they're produced to where they're consumed, and storing them along the way. This is a not-insignificant share of all human activity, and it's designed to be as invisible as possible. The whole point of retail, from Amazon, Walmart, and Target down to indie thrift shops and farmer's markets, is to encapsulate the complexity of getting products from A to B (closer to A to Z, given all the intermediate steps, middlemen, and constraints), so consumers just have near-instant access to nearly everything they want to buy.

The limitations of modern logistics are occasionally visible—Amazon temporarily halting the shipment of non-essential items in March, PPE stuck in warehouses in April ($, WSJ), and much less disastrously e-commerce companies use passenger boats on the other Amazon to deliver goods to remote customers ($, Economist). But for the most part, the system works incredibly well. Consumers have access to abundant goods, with a broad enough selection that entire businesses are built on curation, and those goods are shipped to retailers so quickly that with a handful of exceptions, even a global pandemic only caused shortages for a few weeks.

Meanwhile, logistics seems to be one of the most fundable spaces out there now: in the last month: Loadsmart, a freight broker raised $90m ($, WSJ), Gatik raised $25m for autonomous trucks, Flock Freight has raised $113.5m for its software to consolidate partial truckloads into single shipments, Cargo.one raised $42m six months after its last round, Stord raised $31m for on-demand warehouses, and self-driving truck company TuSimple got $350m. At the same time, Blackstone and KKR are acquiring warehouses, and Amazon alone is responsible for a quarter of all warehouse space leased so far this year ($, WSJ).

Clearly, investors are excited about many facets of logistics. At the same time, shouldn't they be worried? The last investor I mentioned, Amazon, has a reputed tolerance for losing buckets of money to achieve economies of scale, and logistics is a major focus for them. They own warehouses, trucks, and planes, and they're not exactly slouches at software—including autonomous vehicles. What has changed the industry to make it so compelling that even head-to-head competition with Amazon doesn't offset the potential profits?

Regulation and Legibility

One of the proximate causes of the boom is a shift in the logistics industry is the Federal Motor Carrier Safety Administration's Electronic Logging Device mandate. US truckers have limits on how many hours they can work, either consecutively or in a week. Prior to the mandate, they filled out paper forms listing how long they worked. Truckers are paid by the mile, not the hour. So when they were an hour away from making a delivery and ten minutes away from exceeding their allotted hours, they had an incentive to round in order to keep their reported hours within the limits.

That works with paper forms. It doesn't work with electronic logging, which tracks time and location automatically. This had three effects:

  1. It reduced the supply of trucker-hours, since working overtime now reliably led to escalating fines and the risk of carriers getting shut down entirely.
  2. It also increased the uncertainty for each trip, since the driver may not be allowed to complete it.
  3. It meant that drivers were being tracked everywhere, at all times.

This has had all sorts of side effects. There's now much more data on who's in transit, where they're heading, how fast they're going, etc., and whenever there's a data dividend, some of it accrues to software companies who can use the data to make better decisions. The increased probability that a driver would be unable to finish a trip as scheduled created another sort of demand—a startup called OLIMP, for example, specializes in finding temporary warehouse space for deliveries that didn't make it to their destination by the cutoff time. Often, the recipient will ask the driver to come back—in a few weeks. So either the driver can hang around, not working, or they can drop off their cargo and drive somewhere else. But this presents a problem: they have to find a warehouse, negotiate a price, and actually pay. OLIMP is able to create on-demand spot instances of warehouse space, without having to own the underlying property.

The cost of connectivity is also dropping; one reason the ELD mandate was possible was the steady drop in data costs. But once you have to collect some data, and transmit it in real time, why stop there? Trucking companies are using more detailed telematics to predict mechanical problems. Because human drivers' behavior is more regulated, spotting mechanical problems earlier is more beneficial.

Ubiquitous ELD has another benefit: it's a regulation-driven version of a commoditized complement: suddenly, the market for any software that analyzes trucks' location data in real time is every single semi truck in the US. The AV operators have another tailwind: stricter regulations on truckers' work have led to fewer available trucker-hours per worker, so wages have risen. Relative to just a few years ago, the marginal cost of autonomous trucking has gone down, while the cost of not having it has gone up.

The 4D/2D Factor

A simple way to understand why warehouse prices can be more volatile than other forms of real estate, and can potentially rise more, is geometric. They're priced in terms of square feet, but they contain a certain number of cubic feet, and those cubic feet can be used more efficiently over time. It's a dimensional arbitrage, where buyers pay for square feet of space and earn based on cubic feet of throughput per minute.

Warehouses are getting more technological to take advantage of this. Newer warehouses tend to have higher clearance (extending a spatial dimension), and they have more outlets than they used to, which accelerates the time piece. Stories abound about how fast-paced and dehumanizing Amazon's warehouse work is, which raises interesting practical and philosophical points. The reason there are humans in the warehouse at all is that automating the job of picker has proven difficult. Grabbing a box off a known space on a specific shelf is something robots can do, but identifying the contents of a bin, reaching in, selecting the right item, and putting it in a box is still hard. It is, as it turns out, one of the processes that can't technically be dehumanized. (Amazon, for what it’s worth, has been able to staff these warehouses, indicating that however demanding these jobs are, the alternatives are often worse.)

Amazon, which has hired over 200,000 net employees year-to-date, is certainly aware of the high cost of all this labor. But they're also aware of "labor" in the sense of organized labor, and that the efficiency and reliability of their business could be disrupted by a strike. Amazon is taking a sort of Otto von Bismarck approach, where if you provide a safety net to make poor people middle class they're less likely to cause problems for the powers that be.

So technological change explains some of the demand for warehouse square footage. Traditionally, in conditions of high demand, supply rises to meet it, but that hasn't happened as quickly as econ 101 would suggest.

This is, as it turns out, for econ 101 reasons: warehouses can only exist in places where land use regulations permit them, and those regulations are set by local governments. These governments care, approximately, about jobs, but in the short term they're more interested in sales taxes. A retail outlet or, ideally, a car dealership will generate lots of sales taxes, and tends to get the zoning it needs. A warehouse, though, doesn't generate as many purchases that are directly taxable by the city. They're a big business, but an abstract one.

Covid's Acceleration of Time-Shifted Demand

In one sense, the logistics system is one big linear algebra problem. Unfortunately, it's hard to solve because some of the entries in the matrix only update when you send them a fax and they fax you back something they copied off a literal clipboard. External forces are changing this. Amazon has always been a competitor with traditional retail, but increasingly retailers are competing on exactly the same dimensions Amazon is. Two-day shipping is only possible with a vast and well-run logistics system, and big box stores have invested in approximating it.

The odd competitive dynamic is that shipping speed advantages vary at different points: Amazon is the best place to buy many products if you need them in two days, but its competitors are often better when you need the product either a) some time this week, or b) in the next hour.

Covid threw all sorts of wrenches into this system, by causing a one-time shift to e-commerce, which will partially reverse, but also by changing the form of e-commerce. Now, some of the last mile is managed by the customer, rather than the seller, which means that the store is a sort of delivery station that offers same-day availability.

Curbside pickup leads to some interesting inventory-management possibilities. Historically, stores had to have enough goods on hand to sell to whoever walked in the door. Now, in many cases they know exactly what those customers will leave with, a few hours in advance. This lets them run with a bit less inventory on hand, or with a bit broader of a selection, as long as they can have high confidence that the rest of their logistics backend can keep products in-stock.

Early in the pandemic, many people (including me) thought that one result would be the decline of just-in-time inventory management. If a key link in the supply chain can shut down, everyone had better have a few more months of inventory, just in case. While that's true at the production end of the supply chain, it turns out that the opposite is true closer to the consumer: changes in their behavior, from a shift to e-commerce to a blending of online and in-store, give stores a slightly earlier look at what they'll need when. But that only pays off if their entire fulfillment system has high throughput and reliability.

Software is eating every industry, but it's growing in logistics faster by advancing on three fronts: regulation makes everything more trackable, and makes tracking more important; Amazon sets a standard that other retailers have learned they can occasionally beat, although they need to pay up for scarce warehouse space to do it; and Covid convinced consumers to give retailers just a little bit more warning about what they'd need in stock soon.

Subscribe now

A Word From Our Sponsors

As economic activity comes online, new data trails are left behind. Over 200 innovative investment firms leverage this data for alpha-generating insights. Thinknum’s alternative datasets power investment decisions across more than 500,000 public and private companies around the world.

For example, track job listings to identify which companies are expanding in real time. Or track the individual tenants by commercial mall to analyze churn on a daily basis.

Interested? Come learn about all of our 35 datasets here.

Classifieds

Stripe is hiring a head of strategy. That’s a woolly title. What it really means is: manage a small team of people who spend their days thinking about the ecosystem around Stripe, developing a perspective on how it’ll change, and then writing thought-provoking and tightly-reasoned memos about how Stripe is doing things wrong. (Do memos matter? A particular document of this type was significantly responsible for Microsoft’s Azure business.) If you’re interested in this role, email John Collison (john@stripe.com) with a short memo about an important but non-obvious opportunity that some business is overlooking. (It can be Stripe or any other company.)

Elsewhere

Infohazards

Jeff Lonsdale has an important essay on the concept of infohazards, and its application to life. An infohazard is any piece of knowledge that makes you worse-off, from the trivial (you are now conscious of your breathing) to the serious (consider your mortality). A useful view of life is that social technologies are built on load-bearing misconceptions; if you and your neighbors are all somewhat patriotic and civic-minded, for example, you'll probably have a more pleasant life, with more prosocial behavior, even though the specific people and place you have those feelings towards is an accident of birth. Learning about your your neighbors’ moral flaws or gaps in your country’s collective myth makes you better-informed, but it also makes you less likely to engage in cooperative, positive-sum behavior. (Patriotism, ironically enough, is more socially valuable in a globalized, neoliberal system, since it encourages exactly the sort of economic specialization that produces gains from trade given low tariff barriers.)

When I was younger, I assumed that shared delusions were always harmful, and that everyone had a moral obligation to tear them down. This did not just make me somewhat unpleasant to be around, but was also empirically wrong. The first thing I read to challenge this well was Interfluidity's essay on financial complexity as a distributed conspiracy to increase investment in the future. The infohazards piece above is a powerful generalization of the concept.

How The Grinchbots Steal Inflation

As the world economy starts to permanently reopen in 2021, it makes sense that pent-up demand and higher savings will lead to a surge in spending, and an increase in inflation. Tyler Cowen doesn't think so, on the grounds that companies don't want to look like profiteers, and would rather constrain supply. For some products, that will probably be the case; I've written before about how cheaper manufactured goods, consumer expectations, and "shrinkflation" make inflation numbers noisier. (In categories with frequent discounting and variable pricing, it will be easier to hike prices; low-end clothing retail, grocery stores, theme parks, and airlines will be able to inflate by omission, by not offering their best deals.)

Simple economics says that artificially low prices lead to shortages, hoarding, and black markets. More complicated economic models say that companies can choose where to set prices, and that they have concerns that extend beyond supply and demand curves. And, as in so many domains, software imposes a simpler model on reality. In this case, "Grinch Bots" enable automated bidding on scarce products like new Playstations, which then get sold at their market value on eBay. This is not a significant share of the economy just yet, but software scales fast. There has already been a multi-billion dollar acquisition of a consumer-facing tool that automates something similar. If people will download an extension that lets them passively comparison-shop for better prices, they're likely to use one that lets them actively comparison-shop for available inventory.

The Hack Expands

SolarWinds has said that fewer than 18,000 customers downloaded the hacked update to their software, but only a few dozen victims have been identified so far—recent additions include Microsoft, which says the attacks are ongoing, and the National Nuclear Security Administration. Before marketers were talking about "the long tail" to mean the back catalogue of obscure products, web pages, and search queries, insurance companies were using "long tail" to refer to risks that kept generating claims for years after the fact. As in insurance, there are open questions about both the number and magnitude of claims. There have been relatively few companies that were utterly destroyed by software breaches (Nortel is, allegedly, the biggest example by far), but software vulnerabilities can cause liabilities far in excess of the value of the software itself. Right now, companies tend to disclose late and settle fast, which works when the data they lost belongs to consumers, and doesn't work nearly as well when the victims include Microsoft and various arms of the Federal government.

(Disclosure: I'm short a little SolarWinds.)

AR and the Physical/Digital Switch

In the year after the Facebook IPO, a popular assumption about mobile advertising was that it was a net negative. Mobile ads didn't monetize well, because entering a credit card number on a mobile phone was a pain. But users were switching. So, if you were building a financial model of an ad-dependent company, you'd bake in a gradual drop in revenue per user due to the mobile shift.

That, obviously, is not what happened. As it turns out, mobile advertising works just fine, and for many purposes it's far superior to desktop.

That flip is useful to keep in mind in other domains. New technologies don't just catch up; they can surpass. A recent example of this is Google's AR-enhanced makeup shopping. While augmented reality's assumptions won't perfectly reflect reality, Google makes it trivial to sample many times as many products in one shopping session; for a subset of customers, the digital shopping experience is better than the analog one could ever hope to be.

Antitrust and Real Interest Rates

The EU has cleared Google's acquisition of Fitbit, with conditions. One of those conditions is a ten-year wait before Google can integrate Fitbit data into ads. Timelines like this are sticky: ten years is a nice, round number. But ten years during which 10-year treasurys yield under 1% is, financially speaking, not the same length of time that it was when rates were higher. As tech companies mature, they get more defensive, and they pile up cash on their balance sheets. Both of these factors encourage them to think much further into the future. If Google can subsidize Fitbit now and use its data later, the discount rate applied to that "later" matters in measuring how strictly their behavior is curtailed.