Private Equity and Software are Competing to Eat the World
Plus! Gnomic VCs, AI companies as software-adjacent, Solomonic financial engineering, the subtleties of fees, and more...
Why I’m Going Paid
As I mentioned last issue—as I’ll be mentioning ad nauseum—I’m putting most of The Diff behind a paywall. This week will be the last fully free week. $15/month or $150/year will get you at least three issues per week, only one of which will be available to subscribers in the free tier.
There’s a very obvious reason to do this: I like getting paid. But I also like the incentives. A paid publication is a double opt-in incentive plan: it gives me a good reason to write more consistently, and it gives subscribers a good reason to read what I write. We’re both committing to a better product.
Subscriptions offer another layer of incentives, too: they change the payoff function of controversy. In ad-supported media, there’s a tendency to say the most outrageous stuff possible for clicks, usually by omitting details that explain the bad guys’ behavior. Subscription-based media can use the outrage-cultivation model, at least in general news, but the drift is slower and thus easier to stop. But the other side of online outrage is innocuous stuff that gets taken out of context, and paywalls significantly reduce the viral coefficient of anger.
Private Equity and Software are Competing to Eat the World
Some friends of mine run a B2B software company focused on a very specific sector of the economy—one of those “Oh, I suppose that exists, and sure, it makes sense that it’s a multi-billion dollar industry, and it doesn’t surprise me at all to find out that everyone in the industry thinks that fax machines are essential and that Excel is high-tech” sorts of situations.
They sell software to independent companies. Meanwhile, private equity firms are buying up and slimming down companies in the same industry.
In a way, these are both the same business—every fragmented market eventually gets gobbled up by Harvard MBAs armed with spreadsheets or Stanford CS majors wielding text editors. In both cases, the goal is to standardize, and implement best practices, to profit from paying a fixed cost that can be amortized over as many end users as possible.
It makes sense that finance and software would compete, because they’re both highly abstract businesses. In software, you make a model of the world and encode it; in finance, you build an operating model and then optimize it. In fact, they often solve exactly the same problems: better management of inventory and receivables, more efficient advertising, optimizing workloads, getting rid of employees who don’t add much value.
So which model wins? The answer comes down to transaction costs: is it cheaper to convince a small business owner to sign up for a subscription product, or to convince them to sell? And that depends on a whole slew of other factors:
How cheap is debt financing compared to equity financing?
How sophisticated are these business owners?
How likely is it that anyone would want to inherit this business from their parents?
Don’t underestimate the last one. One of the greatest beneficiaries of the baby boomer retirement wave is a small company called RCI Hospitality. “RCI” is an abbreviation for “Rick’s Cabaret,” and “Cabaret” is a euphemism for strip club. Strip clubs are a profitable business, but most people don’t want to own one, which means the odds of someone a) owning one, and b) having a kid who wants to inherit it, are quite low. As a consequence, RCI can pay bottom-dollar to buy strip clubs, cut some costs, and turn a healthy profit.
In other cases, the buy-out-vs-sell-to divide is harder to parse. WeWork was, in part, a bet that if you built software for managing and marketing short-term office leases, it was cheaper to just lease the space and resell it than to convince landlords that WeWork’s product was worth using.
Some industries get totally transformed by a software product—recruiting pre- and post-LinkedIn are totally different. Others get transformed when they get bought up by PE operators; Invitation Homes permanently changed the way residential real estate functions, mostly by providing a well-funded marginal buyer.
And even the definition of “Software” is blurry. Hamburger University is basically a piece of technology that very slowly and painstakingly writes how-to-run-a-burger-joint source code into store managers’ brains. Marriott and Hilton are so abstract that they barely even exist; they’re an interface between guests and hotel operators, but they themselves don’t own the assets.
This thesis has two corollaries: for every enterprise software company with a surprisingly slow and expensive sales cycle, there ought to be a private equity-backed rollup. And, conversely, for every rollup that finds it hard to squeeze out incremental cost benefits, there’s a software company waiting to happen.
A New York-based VC is incubating a new business in the fintech space, and looking for a new hire on the business development side. If you or someone you know is interested in early-stage tech, has sales experience, and wants to join something exciting in the very early days, please reply to this email and say so. Fluency in Mandarin a big plus.
I have a new piece in CoinDesk today on why China’s leadership is excited about blockchain. And no, it’s probably not because they love the idea of a permissionless, anonymizable currency system
I asked a famously gnomic and successful VC about this a while ago, and he had another take: VC tweets are vague because the insights are born out of incidents they can’t or won’t talk about. You can occasionally find good takes on the dramatic backstories of successful startups (Hatching Twitter is pretty dishy, for example), and compare investors’ public pronouncements to their boardroom behavior. Ever investor who says “Always invest in smart people who work hard” is saying that because they’ve been part of a board that deposed someone for being stupid or lazy. At least by CEO standards.
In a way, it implies that venture capitalists are nicer than they’d otherwise be. Given the failure rate of startups, and the sample size of investors compared to founders, they came by all that gnomic wisdom the expensive way. It’s nice to know that venture-backed startups are one domain where you can cost someone millions of dollars and when they complain about you it won’t be by name.
Martin Casado and Matt Bornstein of a16z have a great piece on why “AI” companies have worse margins than other software companies. The root of the comparison is that traditional software companies find something with a high fixed cost and low marginal cost, then scale it and get high and rising margins. In AI, the high fixed cost is born by academic researchers, and they publish their results. The business tends to involve a lot more data-labeling and model-training, all of which is a) expensive, and b) scales right along with the top line. It always feels like an AI-based company is a software company, like Google or Microsoft, but it’s really one layer down; an AdWords agency, or a consulting shop where everyone gets really efficient once they learn Excel.
Scott Locklin has more, from a practitioner’s standpoint. Key quote: “I’ll go out on a limb and assert that most of the up front data pipelining and organizational changes which allow for it are probably more valuable than the actual machine learning piece.” The right business plan for an AI company might be to raise money as a software company and then use that capital to transition into a consulting model.
And the first half of this Ben Horowitz interview has even more, including a discussion of Databricks. Databricks is a great product, but it’s also the single most efficient way to convert interesting hypotheses into colossal AWS bills.
On this topic, The Economist pointed out a few weeks ago that China’s AI companies don’t just benefit from lots of local engineering talent—they also benefit from cheap data-labelers in poorer provinces.
And, two more points on China and the limits of scale. First, a detailed writeup of a Chinese academic paper mill. As Google has discovered in its battle with black-hat SEO, any sufficiently scalable way to automatically produce text lends itself to detection. China is a special case in academic publishing, because many non-academic jobs require publications for career advancement; there’s demand for something that looks like an academic paper, but that nobody will read. And, in an example of the demerits of an entirely different sort of scaling: Hainan province is planning to take over HNA. HNA was originally a small airline servicing a tropical island, but spent 2015-2017 frantically levering up to buy a crazy grab-bag of assets—Ingram Micro ($6bn), a quarter of Hilton Worldwide ($6.5bn), 10% of Deutsche Bank, real estate, more hotel brands, 20% of Dufry, the third-largest aircraft leasing company ($10.4bn), and various airlines. They’ve been in trouble for a while, hitting employees up for reverse-payday loans, and now they’re unwinding. HNA seemed to exist to transform non-RMB cash in the Chinese banking system into assets anywhere else in the world. But any company that grows thanks to readily available financing will eventually find that it’s too dependent on financing, and vulnerable to random external events.
The NYT has a great piece on what was either a high-margin retailer or a fraud, involving rare books in France. The basic model was to buy rare books and letters, and sell shares in them to retail investors. The sale contract may or may not have promised a minimum return of 8%; the Times is frustratingly vague on this. Notable because a few US-based companies like Rally Rd. and Otis have started offering shares in old cars, rare books, expensive sneakers, and modern art. The entire business is itself a form of modern art, forcing us to confront the definition of “ownership” in an abstracted and financialized world. If you owned a pair of Air Jordan 1s, you probably wouldn’t wear them, so why not “own” them in a more approximate sense? If owning a 1% share in them gives you more than 1% of the joy of owning the entire thing, slicing collectible objects into shares is a net win for everyone. On the other hand, “collectible” is a misnomer, since the other reason to own it is that if you have it, nobody else does. So these businesses might reduce the aggregate “collectibility” of collectibles, while keeping buyers happy. It’s Solomonic justice meets financial engineering!
I wrote a few months ago about how “deepfakes” are just an easier way to do what the media already do—reprocess video to fit a predefined narrative. We’re seeing the first test of this theory, with a new Bloomberg campaign video that technically violates Twitter’s strict rules against manipulative edits, but doesn’t violate Facebook’s. The Twitter rule is broad enough that it can’t be fully enforced as written—it would ban quoting “We are all Keynesians now” and “Information wants to be free,” both of which are truncated versions of full quotes with very different meanings. Perhaps social media moderation will be a good live case study for the a16z argument that any software company that needs to continuously learn from real-world inputs will have structurally low margins.
Finally, Kris Abdelmessih has some useful thoughts on investment fees. The fundamental point: for a given investment strategy and fee structure, lower volatility generally constitutes a fee increase. It’s very convenient for hedge funds that so many large investors are volatility-sensitive, since that constitutes a wealth transfer to asset managers. As a corollary, if capital allocators start to figure this out, we’ll finally see a swing away from large multistrategy shops and back to more focused operators.