Wish: E-Commerce, Fast and Slow

Plus! Online-to-Offline; <$10/month; Movie Lots; Taxes and Incentives; Unfairness

In this issue:

Wish: E-Commerce, Fast and Slow

From time to time, but more rarely in the last year or two, there’s  been a cogent bear case about Facebook: it turns out a huge chunk of  their ad revenue comes from just one company, or just one category. The  exact company varies—early on, it was the daily deal operators (who  relentlessly copied one another’s ad copy) and the gaming companies (Zynga  alone was 12% of Facebook’s revenue the year before the IPO). More  recently, the worries have centered around Wish.

This turns out to be a problem that partly solves itself. Companies  only end up being a large share of Facebook when their unit economics  support lots of expensive direct-response ads. But this dynamic creates a  deep order book for any given Facebook ad. When one advertiser drops  out, there are many more who barely lost the auction for a given user’s  attention.

Wish, whose parent company ContextLogic filed its S-1 last week,  is a great case study in how these economics work out for advertisers,  and is interesting on its own as a case study in revitalizing an old  business model, and undercutting some key e-commerce assumptions.

Wish’s main product is an e-commerce app geared towards consumers who  are sensitive to price and little else. Browsing the app today, I see  wireless headphones for $6, t-shirts under $4, polarized sunglasses for  $2, and a pair of 2TB thumb drives for $6. Wish takes a commission from  merchants, and also gets paid by them for in-app ads and shipping.

The prices are absurdly low, because Wish is making certain  concessions to get them that way. Most of the e-commerce world is  obsessed with faster shipping, partially driven by Amazon. Wish tacitly  concedes that it’s not going to win the battle for same-day shipping,  and goes in the opposite direction: their average time from order to  delivery was 22 days last quarter, a material improvement from 62 days  in pandemic-disrupted Q2. Wish also has a relatively light touch on  quality control issues, intellectual property, and other problems. There  are 4,143 BBB complaints  about their products, some of Wish’s offerings do not try especially  hard to disguise their intended brand-name comp (lots of  “legoe-compatible” toys), and it’s fairly easy to find euphemistic items  like “car fuel filters” (i.e. suppressors for guns).

One viewpoint is that Wish is selling utter garbage. A more realistic  one is that there is a large audience that’s willing to sacrifice  quality, certainty, and delivery time because price is all that matters.  A good ad hoc rule is that consumers satisfice on several criteria, and  then maximize on one; airlines used to complain that they only offered  basic economy seating because people who use price comparison sites  always sort by lowest cost, and while Costco doesn’t give itself much  margin flexibility, it does insist that its Kirkland brand must be 1% better on whatever trait matters most.

Wish’s target demographic is low-income. Their S-1 is full of  references to the global middle class, and to households under under  $18,000 a year. In a Forbes interview, the CEO says that buyers' credit cards are most likely to be declined just before payday.

So Wish is targeting the set of consumers who respond only to price.  As it turns out, this is a large addressable market. It does make the  Wish shopping experience more fraught, since users have to get used to  parsing reviews to figure out which ones are fake, carefully reading  product descriptions to ensure that they’re buying a phone rather than a  case, or a brand-name toy rather than a set of LED display lights for  the toy in question.

But this, too, is part of the model. Wish is rare among US e-commerce  players in that it cites a monthly active user number. Not a count of paying users, but of total users of the app,  regardless of whether or not they pay. As it turns out, Wish doesn’t  just source its products from China; it gets its KPIs there, too:  Alibaba, JD, and Meituan Dianping have all released this number, but in US  e-commerce it’s relatively uncommon. Interestingly, one e-commerce  player that did offer this number was Zulily, which is now privately held. It’s part of Qurate Media, parent company of QVC and HSN.

QVC and HSN pioneered a model of direct response advertising where  the ads were just another form of content. It’s not as entertaining as  other scripted shows, but the revenue per viewer-hour is higher, and if  the ads are entertaining, the advertiser keeps the upside. Wish is  thinking the same way: they claim that users spend an average of nine  minutes a day on the app—low for a consumer entertainment product,  exceptionally high for a product that’s literally one big ad. And within  Wish’s commission-based ads is another ad layer: like Amazon and  others, they’ve found that e-commerce search and product pages represent  excellent inventory for ads, so they let sellers buy ads for their  products on the site. As I’ve argued on Amazon, this is a way to price  discriminate: it’s hard to get sellers to reveal their unit economics  directly, but ad auctions show them automatically, so in-site ads are a  way to charge every seller their expected profit. Wish plans to branch  out from there; the S-1 alludes to the possibility that they’ll sell ads  for external categories like travel and finance.

And where does this audience come from? Wish made $1.30bn in  commission revenue in the first three months of the year. They spent  $1.13bn on marketing in the same period, almost all of which was for ad buys.  Like TikTok, they’ve paid up to acquire an audience that they can use to  train their models so the next few million people they acquire are more  profitable. As is increasingly traditional in S-1s, they offer a cohort  analysis in terms of the gross profit per customer by year:

In one sense, Wish is an audience arbitrage: they find people who are  browsing aimlessly on Facebook, and convince them to browse aimlessly  on Wish instead. They slowly ratchet up how much they monetize those  users, while also convincing merchants to spend more of their own gross  margins running ads on Wish.

Wish’s very marginal customers and low-margin merchants create  another interesting opportunity. Because Wish collects revenue upfront  but remits it to merchants later on, the business actually generates  some float. In a zero interest rate environment, that’s not an automatic  moneymaker, but it does mean that Wish can and does use working capital  lending (in the form of early payment to reliable merchants) and  buy-now-pay-later financing as one more tool to keep merchants and  customers on the platform. The same traits that make Wish’s customers  and merchants less desirable to other e-commerce operators give Wish  more ways to attract them through very low-level financial engineering.

It’s interesting to consider the long-term steady state of the  business, which is currently halfway between TikTok and QVC. For some  users, the Wish product just clicks; it’s entertainment shopping, with a  low enough price point that the hit from gambling on dubious products  isn’t so bad. But for a pickier audience, Wish is a terrible deal:  taking four weeks to send customers a product that breaks right away is  exactly what Amazon is trying to avoid. So Wish may end up churning  through a large audience of not-quite-target customers, and retaining a  smaller audience of diehard loyalists. This is not by any means a  disaster. Qurate, for example, can handle a heavy debt load and return  lots of capital to shareholders. Its audience isn’t huge, but at this  point, they know exactly what they’re getting, and Qurate can reliably  predict what they’ll spend.

Wish is the sort of company that will eventually get a catalyst-free  writeup from a major short seller. The risk factors are basically a  prompt for a diligent researcher. For example, Wish says it has 500,000+  merchants and that it frequently gets IP infringement complaints. All a  short seller has to do is order some suspiciously affordable brand-name  products, see which ones are obvious fakes, and include photos in their  writeup. Or order some especially cheap off-brand products, and  highlight the worst of the lot. Some short-selling due diligence is  expensive, but the very nature of Wish’s product makes it cheap. The Luckin Coffee short report involved immense manual labor—reviewing receipts, lurking in stores, obtaining video footage of the business in action. A Wish short report can budget $100 or so for buying a grab bag of products (anyone planning such a report should factor shipping times into the expiration date of their put options).

The most helpful way to look at Wish is that it’s filling gaps. Slow  shipping is cheaper than fast shipping, because the most lucrative  customers are time-sensitive; low-quality products are damaging to most  e-commerce companies' brand names, because they set high customer  expectations. Even within Wish’s own revenue lines, there are examples  of this: Wish offers one variant of its merchant ad product that  promises rapid-but-expensive traffic instead of more gradual visits;  they’ve replicated AWS’s spot-versus-reserved pricing, with the same  price hit for immediacy. Like anyone else in the space, they’re making a  fundamental bet that when they touch a large amount of economic  activity, they’ll be able to detect where the margin opportunities are,  and exploit them.



The WSJ rolls up some data on a trend I’ve cited off and on this year: big tech companies are buying more commercial real estate ($):

While Facebook, Microsoft and Google have said they would  support working from home beyond the pandemic, that hasn’t appeared to  have dulled their appetite for warehouses, data centers, retail stores  and even more office space. This year alone, the five tech giants have  expanded their real-estate footprint by more than a quarter, their  fastest rate over the past decade.

At many tech companies, “dogfooding” is seen as an important part of  the process. Yahoo couldn’t offer a search engine that Yahoo employees  preferred over Google, while Microsoft employees use plenty of Outlook,  Excel, and Powerpoint. But dogfooding can be done to excess—profiles of  early Facebook point out that employees were generally not big  users of Facebook, because playing with an addictive social product got  in the way of building one. The real estate buildout is a reflection of  the same phenomenon: online interactions are a superior substitute for  many in-person ones, but the more talented employees are, the more  valuable the bandwidth of in-person interaction is.

I wrote a bit more about the long-run steady state of working from home here in May.


In yesterday’s post,  I mentioned the observation that there are very few viable subscription  products that charge under $10/month. This makes every exception worth  looking into: European neobank N26 now has a €4.90/month pricing tier.  In this case, it looks like pandemic-induced price discrimination. This  account tier is similar to the company’s core €9.90/month product, but  without travel insurance, so presumably the plan is to get users at the  current pricing and upsell them as soon as travel fully recovers. A bank  is particularly well-positioned for this kind of upsell, since N26 can  suggest an upgrade as soon as they start seeing transactions for hotels  and plane tickets.

Movie Lots

Netflix is investing $1bn into a studio in New Mexico,  helped along with some significant incentives from the state  government. This looks like a story about California’s network effects  unwinding, and that’s probably part of it—it’s easier to convince film  workers to leave LA when they’ve already left—but part of it is due to  cost pressure. Private equity firms have been buying up film lots ($, WSJ), and Netflix has always been very careful about controlling sources of cost inflation.

Taxes and Incentives

New York is considering raising parking fines and offering a cut to people who report illegally parked cars.  Governments have been relatively slow to realize what consumer Internet  companies figured out early: any labor-intensive surveillance activity  can be carried out by, or at least significantly assisted by, a  sufficiently large userbase. In most tech contexts, this means using a  flagging/reporting function to a) triage moderation decisions, and b)  train automatic moderation and spam-filtering features. But even absent  training, it’s a useful technique. Since cities, especially large and  expensive ones, will be in fiscal trouble for a long time post-pandemic,  they’ll be likely to experiment with new revenue sources.


Tyler Cowen has a bullet-biting piece on why we should prefer unfair vaccine distribution  until there are enough doses for everyone. The benefits of vaccination  for a given  population are nonlinear, in a Dickensian sense: 65%  vaccination rate, R(t) of 0.99, result happiness. 64% vaccination rate,  R(t) of 1.01, result misery.