The Relationship-Transactional-Relationship Business Cycle
The revenue model for many companies looks a bit like this:
- Initially, the product is new enough and undefined enough that revenue-producing activity is basically a form of consulting: figure out the customer's problems, figure out how the new technology can help solve them, and build a bespoke product for which plausible prices are in an order-of-magnitude range. This model applies equally well to physical industries (the first car companies built and sold one-offs, and had to design many of their components) and to digital ones (Google and Yandex both initially monetized through enterprise search products; one of Facebook's earliest ad deals was giving Apple the then-unique privilege of having a group that spanned multiple campuses).
- As the business scales, the mass-produced or self-serve models become the obvious winner. There's no reason for a business to be held back by hiring and training salespeople when the end customers clearly understand the product just fine.
- But the self-serve model has a paradoxical side effect: less transactional friction means that variance in customers' growth rates is magnified. It also creates demand for intermediaries: any self-serve powerful enough to do everything an expert user wants is probably baffling to the average new user. So a company with monopolistic economics with its self-serve product often ends up dealing with something approaching monopsony from its most effective customers. At that point, the relationship stops being so transactional, even if the typical interaction is still a transaction without a human in the loop. Relationship managers on both sides handle edge cases, and also coordinate the division of labor (i.e. which features are value-added products offered by loyal intermediaries, and which are built-in features) and the rollout of new features.
The transition from the second to the third stage of the cycle is a fraught one, because the second stage makes for an easy pitch to both investors and employees. "We don't have a single salesperson in our entire organization!" means "We have structurally higher incremental margins than the competition" to investors, and "nontechnical people will have zero input into your priorities" to employees. When a company doesn’t want to talk to customers, and wants the product to speak for itself, they’re embracing a higher-variance model. The default outcome is failure; the product speaks for itself, and doesn’t have much to say. But when iit works, it leads to faster growth, because growth isn’t bottlenecked by sales headcount. And the companies that embrace it are also less bottlenecked, since they don’t have a relationship to manage on their side.
But in many cases the move to stage three is a necessary one. The self-serve market is naturally limited since some customers don't want to interact that way. But the more important concern is that a low-friction, low-touch approach also makes it hard for customers to plan ahead, and encourages big customers to have a somewhat adversarial view. In a narrow sense, the existence of online marketing agencies represents inefficiency in the business models of Google and Meta, and cloud computing consultants imply that AWS and Azure are leaving money on the table. But if the companies that offer these services act as if that's true, their biggest customers will also be reluctant to trust them.
Transaction economics include the flow of object-level decisions—do we buy this Google click, spin up that EC2 instance, or accept this Stripe transaction—and a stock of expectations and trust slowly built up on both sides. It's essentially a form of reputational capital, and a company that's betting most of its revenue or operations on a counterparty that they can't have a conversation with is, in some abstract sense, undercapitalized.
The best companies are capital-light almost by definition: they can fund growth through their own profits or through raising outside funds on advantageous terms. But as growth companies mature, they often get more capital-intensive (the topic of an upcoming post for paying Diff subscribers—upgrade to read it this week. Some of that capital is tangible, like datacenters and warehouses. And some of it is intangible capital embodied in relationships and norms; in accounting terms, a fraction of this shows up in operating expenses and cost of goods sold, but it's a material share of every mature company's capital in the sense that an equivalent product without the equivalent relationship would not be competitive at the same price.
This pattern shows up across many different kinds of companies:
- AWS is surprisingly diligent about asking small customers what they need. Steve Yegge talks about this a bit in a recent interview: even when Grab was small, AWS was curious about what they needed. For a transaction business with high margins, upside from a customer relationship is pretty close to equity—get a buyer locked in before they grow 100x, and the revenue associated with them will also grow around 100x.
- In asset management, a prime broker relationship is at one level a purely transactional one: it's a prime broker, not a prime buddy. But even though most interactions with a prime broker are trades, brokers do try to get to know their customers, figure out what kinds of research and corporate access they want, and, at times, decide who will be the one to get hard-to-borrow shares of popular shorts based on something other than the immediate financial payoff.
- API-first companies often reach the point where they need salespeople to get bigger customers comfortable with long-term commitments. And, as it turns out, it can be better to get a guaranteed stream of income over multiple years at a discount than to insist on the default pricing schedule. As a user's needs get more complicated, they evaluate products in more dimensions, and care about things like long-term reliability and the future product roadmap over the immediate question of whether they might save 5% going with someone else.
- And if you treat campaign promises in exchange for votes as the relevant transaction, even political systems do something similar to this. The set of policies appealing to 50.1% of the electorate at any given moment will not necessarily correspond to the proper priorities over a generation—and when the electorate is aware of the most pressing problem, which generally happens during wars and economic crises, they don't necessarily have the best solutions in mind. So a certain amount of government activity moves to institutions that aren't as directly sensitive to immediate changes in popular views, whether in the form of lifetime appointments to the Supreme Court, the tradition of central bank independence, or career civil servants at agencies. However annoying the FDA, IRS, etc. can be, if they had 100% employee turnover and a new rulebook every election cycle, they'd be much more annoying.
One simple way to think about all of this is that once there's product-market fit, transaction costs drop, and the lower some transaction costs are, the more leverage there is in reducing them further. But at some point, the transaction itself matters less than the relationship.
Eventually the roadmap starts to matter for big customers, and retention becomes valuable for both sides: the vendor obviously wants their customer to keep coming back, and the customer doesn’t want to build something on their own that a vendor would have been happy to build for them. This requires some soul-searching for the companies in question: hiring salespeople, building relationships, and cutting long-term deals that aren't available to anyone who signs up for the basic product are all departures from the instinctive behavior of the companies that face this problem. It means conceding that there are parts of the full stack that the company can't control, even if the process they're interacting with wouldn't be possible without them. But, paradoxically, diversifying an approach to sales is actually a way to focus: it's a way to say that high-touch relationships will allow big customers and intermediaries to build the myriad domain-specific wrappers, edge-cases-as-a-service, and rebrandings of generic offerings that are the essential ingredients to ubiquitous distribution. It's a process of figuring out where the most scalable sources of value are, and focusing entirely on them.
Disclosure: Long META, AMZN, MSFT.
Eventually this stops applying, especially in the current environment where companies see more upside from improving margins than improving revenue. But it's directionally true that the right way to think about small, money-losing customers is that their investors think there's a nonzero chance that these companies will someday be big customers who can reliably pay their bills and will be paying much more substantial bills. ↩︎
A Word From Our Sponsors
More and more investors use Daloopa to build and update their financial models. Daloopa captures every reported indicator—from GAAP metrics in SEC filings and press releases down to key performance indicators mentioned in management commentary, or buried in a footnote. Daloopa allows one-click model updates within minutes of new data releases. And with links back to source material, you can always double-check the math and get the numbers in context.
Daloopa is all about increasing idea velocity, and ensuring that investors spend less of their time looking for numbers and manually entering them into spreadsheets, and more time coming up with good trades.
Metrics at Twitter
Marc Andreessen once paraphrased Andy Grove's advice on metrics as "For every metric, there should be another ‘paired’ metric that addresses the adverse consequences of the first metric." Point being that it's a good practice, for any "true north" metric, to brainstorm all of the terrible consequences of prizing that metric above all others. For many ad-driven businesses, the obvious first-order metric is some proxy for usage: more usage means more ad slots, and more data for targeting those ad slots, so it has a nonlinear payoff. But the natural consequence of this is prizing the lowest-effort forms of usage; Facebook had to deliberately tweak its algorithms to downrank political content, since endless partisan flamewars generate lots of pageviews but also make people hate the site.
Elon Musk is going through the same process with Twitter, and has been talking up the metric of "unregretted user-minutes" ($, WSJ) with advertisers. From the standpoint of the public good, unregretted minutes are certainly what you’d aim for—a product that produces compulsive, addictive behavior with no redeeming features is not good for the world. But for an ad-driven product, it's also close to value-maximizing, both because it means that advertisers' content will be paired with less risky content and because it means the app is appealing to users beyond a set of hard-core addicts.
When a payment method has high enough penetration, one thing merchants end up doing is dropping alternatives in order to simplify their own operations. The only thing less convenient than an all-cash business is a some-but-not-very-much-cash business, where cash management is an inconveniently intermittent concern. And since the vast majority of consumers don't rely on cash (and the ones who do rely on cash tend to spend a bit less), it can make economic sense for establishments to switch to cashless. But it makes even more sense to be cash-free by default and to outsource cash management to someone else, so reverse-ATMs, which accept cash and dispense gift cards. It's a way to get the network effects of using every possible payment method, without the operational cost of directly supporting them.
Amazon is working on using generative AI to create ads, including video ads ($, The Information). More purely ad-focused companies are naturally working on the same thing, but one advantage Amazon has is its first-party data. It's hypothetically possible for Amazon to reach the point that every ad is created to target a single user, based on the purchase data that only Amazon has—and with hard sells tempered by Amazon's economic incentive to keep customers around.
Elsewhere in AI
Waymo is doubling its self-driving car service area in Phoenix and expanding elsewhere. Self-driving cars have taken a surprisingly long time, but solving them is both a technical and regulatory problem: it's much easier and safer to offer self-driving cars in an area where pedestrians and non-autonomous vehicles aren't allowed, since so many accidents involve either pedestrians or surprising interactions between autonomous and human-driven cars. So this is a business where economies of scale can be surprisingly strong. But this means it's also a category where ride-hailing apps can justify burning money in order to delay competitors' growth, since the ride-hailing business represents an existential threat to them.
A financial crisis is a multi-week event that turns out to have multi-decade buildup, and the current banking situation is no exception. Patrick McKenzie has a great analysis of how we got to where we are, with an emphasis on how the business of acquiring and paying for deposits has changed. Depositors turn out to be fairly insensitive to interest rates, but very sensitive to the quality of banking apps, which subtly increased banks' net interest margins in a rising-rate environment while increasing their run risk if that rate environment affected their balance sheets. One reading of this is that, despite banks' lackluster performance over the last decade, they were in fact over-earning since they weren't baking in as much rates risk as they were really taking. There are regulatory environments where banks can prosper like this, usually at the cost of capital misallocation, and the tradeoff of a flexible and semi-privatized banking system is that creative, entrepreneurial bankers will invent new ways to suddenly lose all of their money.
Companies in the Diff network are actively looking for talent. A sampling of current open roles:
- A high-growth provider of market data to both retail investors and institutions is looking for an account executive. A background in finance and experience working with RIAs/FAs is ideal. (US, remote)
- A VC backed company reimagining retirement wealth and building a 401k alternative is looking for product/GTM/bizops generalists. (NYC)
- A firm using NLP and other ML tools to give retail and institutional investors access to custom-tailored portfolios is looking for a data engineer. (NYC)
- A profitable AI startup is looking for ML engineers to help build new services to help small companies accelerate their growth. (SF)
Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up.
If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.