Subscribed: Zero-Marginal Cost Thinking in a High-Marginal Cost World
Subscribed is one of those books that you can dismiss as an extended press release. It’s a book arguing that every business should explore, or ideally switch to, a subscription billing model, and it’s written by the CEO of — a company that sells software to help companies manage subscription pricing!
I have a different attitude. Confluence of interest beats a lack of conflict of interest. Sure, this book is Tien Tzuo talking his book. But if he’s wrong, his entire $200m net worth is at risk. He has some serious skin in the game.
The main thrust of Subscribed is that the economics of subscription pricing beat the economics of selling products for fixed costs, on a couple different axes:
- Subscription products can adapt over time.
- They’re continuous market research: instead of guessing what the market wants, you make a tweak and see how it improves your conversion rates.
- The lifetime value of a subscriber is easier to measure, and you’d generally expect it to be higher than the lifetime value of a repeat purchaser for the same product. It’s easier to sell something to an existing customer, but it’s even easier to sell to someone who will be paying forever unless they opt out.
- As a corollary to this, subscription economics allow a company to massively accelerate its growth once lifetime value exceeds customer acquisition cost by a comfortable margin.
Tzuo gives a couple of examples of big companies that sell hefty industrial products shifting to this model. “The digital company. That’s also an industrial company.” But that’s a tagline from an ad, not from a 10-K. This model has been heavily adopted in just one part of the economy: software. Which happens to be the part of the economy where fixed costs are high but marginal costs are close to zero.
That’s where I get stuck. If the cost of “shipping” a new product is that you switch a new version from your dev server to production, then yes, your product can adapt over time. But if you’re selling turbines, the cost of upgrading to a new turbine is about the same as the cost of building one in the first place. As products digitize, you’ll expect their economics to get more subscription-y, but for many parts of the economy, there’s an irreducible raw materials component and a hard-to-reduce labor component; digitalization doesn’t make machinery use less aluminum and steel.
Furthermore, since the great financial crisis we’ve lived in a low- to no-inflation world in most developed economies. From 2009 to 2016, you could round inflation down to zero and just be thankful it wasn’t negative very often. But that’s changing. The CPI is running at over +2.5%, and whatever you think of the current fiscal/monetary mix’s effect on aggregate growth, its effect on the distribution of that growth leans more towards consumers than capital.
It’s entirely possible that the book came out at exactly the time when we should be questioning zero-marginal cost thinking, rather than embracing it. Even in my own life, I see marginal cost as the main determinant of whether or not I subscribe to something. I pay a subscription for streaming music, public transportation, Internet, and housing — every one of these is a capital-intensive business where the incremental cost of providing the service to me is nearly zero.
Rising input costs and tight labor markets don’t invalidate the subscription thesis; they just turn it from a no-brainer into a tough call. To really explore this more deeply, we have to think about two things: the economic reality of subscription companies, and their real balance sheets.
You want to break every cost down into three buckets:
- Truly fixed costs are the costs that you’d pay even if you sold nothing to nobody. For low- to no-marginal cost businesses, these costs are systematically understated, because they tend to be winner-take-all markets. The true cost of building a successful software company in a given field is the amount of money invested in the winner plus all the competitors who failed. These fixed costs are the costs of exploring the market, determining how big it is, and deciding which company’s vision/framework/team should be the winner. From the winner’s perspective, ROIs are always extraordinarily high; a successful software company doesn’t need a lot of fixed assets, and earns back its early operating losses over time. But to the losers, the costs are high and the rewards are nil. There have been some businesses, like broadband in the 90s where arguably the exploration costs were well in excess of what was economically justified, at least given the information people had at the time. Even if things look good for the ultimate winners, the sector as a whole is only going to get capital allocated to it if the aggregate performance looks okay.
- Incremental customer acquisition costs: At one point, Tzuo gives us an overview of how he thinks about the economics of customer acquisition. Basically he sees renewals as a sort of annuity, whose value is far in excess of the first year’s revenues. This strikes me as exactly the right approach. If it takes $50 to acquire a customer who makes you $50 in gross profits in year one and has a 20% annual churn rate, it’s economically irrational to turn a profit. At a 10% discount rate, that annuity is worth over $160. Staff up your sales team!
- Incremental servicing costs: Here’s where things get tricky. For a software company, these sales costs are the costs of helpdesk support, ongoing sales commissions, payment processing, and maybe a little cloud compute and bandwidth. It’s rare for them to make the difference between success and failure; when a company is small, it might be perfectly rational for the CFO to round them down to zero for planning purposes. However, this only applies to software companies. If you ship your product on a truck rather than over a wifi, those costs add up. If your product takes a lot of raw materials and labor to assemble, they add up faster. And if they’re really heavy-duty and need a specialized manufacturing process, scaling is even harder: when your factory is at 90% of capacity, growing 20% means selling more than you can make. (And if your factory’s economics also depend on scale, then you grow your capacity in Integer-X increments, so 20% growth means you now have two factories — running at 55% of capacity apiece.)
The liquidity of software businesses illuminates a lot about economics, but a lot of the problems software companies can ignore are the problems that make or break companies in other industries.
Now let’s turn to some exciting balance sheet analysis. There’s an accounting term called “goodwill,” which is defined as either a) the assessed value of all of a company’s intangible assets — its customer relationships, its brand name, its esprit de corps, its farsighted management team. Or b) the price the business is sold for, minus the tangible assets. b) is the definition your accountants actually have to compute; a) is an ex-post justification for it.
Goodwill is real, but it’s also really slippery. These assets do have value, and they don’t appear on the balance sheet. At a software company, customer contracts do appear on the balance sheet, but network effects don’t; Slack can book deferred revenue from a customer who paid upfront for a year of service, but what they can’t book on their balance sheet is the assumption that if your company isn’t using Slack, it isn’t serious. Nor do they get to place a value on the Slack widget ecosystem. The fact that people at offices around the world say “Slack me when you’re ready to meet” is worth an enormous amount.
As a shorthand, you can define “execution” as the process of turning those intangible assets to cash in the bank. That means embracing the negative takeaway, too: on a long timeframe, a successful business burns through a decent chunk of its customer loyalty, since ultimately cash, not warm fuzzies, is what pays the bills and rewards shareholders. When you hear about software companies trading at nosebleed valuations compared to their trailing sales or billings, that’s what’s going on: they’ve accumulated a massive Invisible Balance Sheet through brand recognition, vendor lock-in, the sales team’s esprit, the fact that their product managers get a heads-up from other vendors when the API’s about to change, etc.
The high incremental margins of software companies are the main reason they can be valued so much based on their potential compared to their backward-looking metrics. Of course, plenty of other companies have intangible assets that are enormously valuable and don’t show up on any balance sheet. But when your gross margins are 15% rather than 85%, there’s much less room for error — and to realize a billion dollars of incremental profit, you need about $6.6bn in revenue and $5.0bn in operating expenses, while the 85%-margin player needs $1.18bn in revenue and $180m in operating expenses. Estimating either that revenue number or incremental cost number to +/-10% for a software company gets you to where you need to be, but for a lower-margin company, that magnitude of variance is the difference between profit and loss.
This presents a paradox of technology companies: in a sense, their valuations reflect the fact that the business is less risky than a traditional business on a similar trajectory — there’s more underlying uncertainty, but the model is more tolerant of this uncertainty.
Ultimately, I found Subscribed provocative, but it left too many questions unanswered. The big mistake companies made in the 1990s was jamming an offline model onto a website — let’s just scan our catalogue and add some hyperlinks! Genius! I worry that the massive wealth creation in the tech sector since then will lead to the opposite problem: we’ll take models that work great when your marginal cost is bandwidth, apply them in areas where shipping and manufacturing are a big chunk of the cost, and find out they don’t work so well after all. Technology changes what’s possible, but you still have to think it through.
 One thing Tzuo knocks is the “event marketing” approach, where you lauch a really big ad campaign to get everyone talking about your product, instead of moving by increments. Arguably, marketing has returns to scale: a Super Bowl ad exposes your brand to 100 million people, and getting people talking about something requires high market share. If a third of US adults have heard of your company, any two people have ~10% odds of having heard of your product. On the other hand, it’s essential to be suspicious of anything that ad creatives really want to be true. A friend of mine is a big-shot advertising creative, and he posts lots of really neat, creative ads on LinkedIn. I have never in my life seen one of these ads in the wild; the stuff advertising creatives find really appealing is apparently not the stuff media buyers think will perform.
My best guess here is that, if you’re marketing online, you can get Super Bowl-level mindshare for just the subset of people you care about; you can use LinkedIn ads, for example, to target every head of HR at a tech company on the East Coast with ads for your HR product. But if you’re selling consumer goods, your market is measured in the tens of millions of people, not the thousands, and the flip-side of all that great ad targeting is that a general-interest ad is bidding for digital real estate against people with narrower audiences.
 And, in a digitizing economy, “capital” often means bidding up the wages of knowledge workers in geographies that benefit from agglomeration effects. We urbanized knowledge workers do not have an especially high marginal propensity to consume, at least compared to somebody who is praying every day that their car doesn’t finally go kaput. And what we do spend on, on the margin, is stuff like real estate, education, and healthcare; strip those out of the CPI and you get a more deflationary world.
 There’s a possibility that new markets will benefit from investors’ well-established preference for lottery ticket-like assets, but building a business plan on extrapolating irrationality into the indefinite future is both bad business and bad, period.