The Counterintuitive Economics of Hiring
Companies are made out of people, and who they hire ultimately determines what kind of business they'll be. This can happen in a very path-dependent way: Sidney Weinberg, who ran Goldman in the mid-20th century and was perhaps the most famous investment banker in the US at the time, got his job in 1907 by taking an elevator to the top floor of an office building, and visiting each office to see if anyone was hiring. He started as a janitor's assistant, but worked his way up.
Janitor's assistant is no longer a partner-track role at Goldman or most other places; as companies have matured and as the labor market has gotten more efficient, the role of chance has shifted a bit. Now, companies invest huge amounts of money and effort into spotting talent, hiring, retaining employees, and replacing them. The incentives here are opaque because most of us will do a lot more getting-hired than hiring. It's easy to see the asymmetry between employees and employers, because losing out on a job is generally a lot worse than losing out on an employee. But in a dynamic market with multiple competing bidders, a company that systematically misses out on the most desirable employees faces constant adverse selection—it's never getting the best possible team but is always competing against it. And companies that get good at talent acquisition but can't do it in a cost-effective way suffer another kind of bleed, where the overhead required to get a good team swamps the benefit of having one.1
So recruiting is one of the fundamental things companies do. This holds true at all levels of the job market. Unskilled workers tend to have more jobs they can switch to, so the market for their time is broader and more liquid; when Amazon raises hourly rates, it's a problem for Uber and McDonald's even though they're not direct competitors. So a company that has this kind of workforce is fundamentally solving the problem of how to handle structurally high turnover, and to balance between overhiring and running short of workers. It's a nontrivial problem: raise wages 5% to account for a temporary labor shortage and you've committed to a higher cost structure for everyone, not just the people you just hired. Some companies accommodate this by building a business around higher wages: Costco and Trader Joe's pay high enough wages that they don't have to worry much about attrition every time a competitor bumps the starting wage by $0.50/hr, but that model works best when companies maximize inventory turnover, which pushes them into other decisions (like strictly limiting the number of discrete products they stock) that have other costs.
The central importance of hiring doesn’t mean companies do every step of it in-house. The usual wisdom is to outsource everything that isn’t a core competency, but the dividing line gets fuzzy. Making great phones is part of Apple’s core competency, but the people who actually assemble those devices work for companies like Foxconn. For recruiting, the nuanced question is not whether or not the entire process can be outsourced, but which parts represent a comparative advantage for the company doing the hiring, and which don't.
Hiring is an optimization problem, and one whose contours start to sound like a good idea for an elaborate German boardgame. The hiring organization’s goal is to find a good person who can do a particular job, and not pay them too much. This happens in a highly adversarial environment, in which those employees are also trying to maximize what they get out of their employer (not strictly pay, but some combination of pay, interesting work, good colleagues, the right location—which might mean an office in a particular city or might mean the freedom to work from home, social status, and good exit opportunities). And that environment is even more adversarial because there are other companies gunning for the same talent, often targeting the same winner-take-all markets. When Apple misses out on a good hardware engineer, they have two problems: the person they didn't get, and the company that got them instead.
A hiring process will have multiple layers of filters, and the optimal fineness of those filters is a function of both the expected value of new information and the opportunity cost of obtaining it. A non-technical person can ask a developer fairly basic questions and get a sense of whether or not they probably know what a given technology is, but would have a hard time distinguishing between “I read about it on Hacker News”-level knowledge and “I use it every day” knowledge. (There are proxies for this, but they’re fakeable.) A more technical interviewer can dig in a bit more, but their time is also more expensive.
I’ve been involved in hiring processes that were, in opportunity cost terms, easily a five-figure investment per interviewee considering the value of the time of everyone who participated. A company that has several high-cost employees take time to prep for interviews and then interview someone, and allows any of them to veto a hire, is spending a huge amount of money on recruiting whether it shows up as a line in the P&L or not. You could try to optimize such a process by doing it serially, having someone interviewed one week by a fairly junior person doing a tech screen, then by others in order of increasing seniority. But subjecting someone to a months-long hiring process that’s also a hazing process is a good way to lose them to a more agile competitor. (And one advantage of the “superday”-style interview process is that some days are like that, just one intellectual challenge after another back to back until you’re exhausted.)
Outsourcing the initial stages of recruiting is less about ceding a core task to an external vendor and more about finding a cost-effective way to do it: the more effective the hiring process has been in the past, the more valuable the time of people who do the filtering in the present. So a recruiter's fee is partly a way to save internal opportunity cost from interviewers.
It's also a way to save on another kind of matching cost: there's substantial variance in company culture, and the more extreme the company culture the more likely it is to be instantly rejected. A company with a friendly, collaborative environment can feel deeply passive-aggressive and lazy to someone who has a more direct approach, whereas a company with a more pointed mode of internal communication can feel needlessly mean. (One of Bill Gates' catchphrases at Microsoft was "that's the stupidest thing I've ever heard," and one of his explicit goals was to hire the smartest possible people. These are a bit contradictory! So maybe some of it was gratuitous, after all. But if you wanted to build software products that would be used by tens of millions of people, and get paid well to do it, Microsoft was once the best option.)
The general rule with any kind of matching business, whether it's recruiting, a dating site, an online travel agency, or a lead-gen service for mortgages, is that there are two forces pushing in opposite directions:
- Leads from middlemen have the worst unit economics, so anything that can reduce reliance on them tends to improve outcomes, but
- Getting unfiltered leads means paying for lots of leads that you'll instantly reject as a poor fit even though they're valuable to someone else. If a single hotel bids whatever it takes to be the top ad result on a Google search for "Hotel," basically none of its traffic will convert; it's much cheaper to pay 20x the cost per click to get leads from Booking.com and Expedia instead, which have been pre-filtered for location, trip date, and price-sensitivity.
This leads to a frequently repeated lifecycle for companies' hiring: very early-stage companies usually do their recruiting entirely in-house, because their founders generally have some kind of network. From an investor’s perspective, this is a good way to get an early read on the quality of a business: if the founder graduated from a decent college and spent a few years working at a bigger company, but they can’t tempt a single ex-coworker or ex-classmate to join them, it may be a problem.2 As the company grows, that network gets tapped out. Even if every new hire introduces a few leads—some companies explicitly ask them to do this!—eventually everyone the company will want to hire in a given extended social circle will get hired. (A company can grow its headcount very cost-effectively if every new hire's referrals produce at least one new hire, but this is incredibly hard to maintain. It seems to have been roughly the case at pre-IPO PayPal, but that just illustrates how hard it is: a strategy that worked for a company that raised its biggest round a week or two before the market crashed is not a strategy that can be reliably replicated; that hit rate is possible when everyone else in the industry is shedding employees.)
As companies scale, they reach a point where there are some roles they can't easily fill, and their choice is either to make recruiting an in-house function or start with external recruiters. External recruiters can be especially useful if the job function is unrelated to what the founders have been doing: if someone starts a company right out of school, or instead of going to school, or after a few years at a FANMG business, they have minimal exposure to the business process side of things—accounting, HR, regulatory compliance, etc.3 And even finding more employees for core functions can be a challenge, because it requires tapping into new networks.
Outsourcing recruiting can mean working with a recruiting firm, but it can also mean choosing investors partly based on their ability to provide referrals. There's a two-way convergence between venture capital and recruiting: When capital is more abundant than opportunities, an edge in investing either means being the first to find unique opportunities or being the qualitatively most compelling option when the quantity of money available at a given valuation is already basically set. For employers, the ones who are making good long-term choices are making a venture bet, and trying to create a talent portfolio that can scale linearly over long periods. The convergence is happening in another way, with full-service VC firms increasingly offering recruiting assistance, whether through an informal process or through a distinct business function. And in the other direction, some recruiters have structured their business to have more of a venture-like payoff from making good placements at good companies. (If that sounds like something you, personally, would be interested in doing, we should chat.) A capital-abundant world is one where finding scarce talent can be the most cost-effective way to get scarce dealflow, so these distinctions will continue to blur.
As a company gets even bigger, the equation shifts again, and they reach the point where they can do more filtering and pay more for leads, knowing that their hit rate doesn't have to be perfect. There are job categories where big companies know that they'll need people soon, even if they don't right away, so it can make sense to hire a bit ahead of demand when the market is tight. At that point, moving more recruiting in-house becomes a better bet. Though there are still cases where it doesn't work, either because there's a temporary need to scale up that doesn't justify adding long-term recruiters to the full-time payroll, or when there's a specialized role that's hard to find organically.
This is not just the time when companies start signing meaningful contracts with LinkedIn and Indeed. It's also when they start to look at hiring as more of a systematic marketing problem: spinning up an engineering blog to talk about the fun and challenging problems they solve is a good way to attract people who delight in solving such problems—and to find the ones who under-index to LinkedIn usage.4
Recruiting is a cyclical business, because it's partly a defensive decision: when big companies are worried that they won't be able to find anyone for a key role, or they fear that a major launch will be delayed at a seven-figure cost because of one missing six-figure hire, they quickly ratchet up compensation accordingly. This also feeds into startup incentives, since acqui-hires are a material source of talent: it's a lot safer to quit your job and start something new when your downside scenario is quitting something new to go back to your old job with a nice bump in compensation. A good read on the 2021 labor market is that it was a short squeeze in talent, and like a lot of short squeezes, that implied objectively overpaying for a while because of the risk of overpaying even more later on. And, also like normal short squeezes, it didn't last forever.
A company's narrow goal in hiring is to get a good risk-adjusted return on talent. Its very broad goal is to define its own future: nobody has gotten rich entirely by underpaying people (though some rich people are a lot richer than they otherwise would be because they kept a lid on compensation). Of all the things you can economize on, compensation is one where it is at least possible to get your money's worth at order-of-magnitude-different compensation levels. Managing a "talent portfolio" is a matter of getting some good work out of everyone and getting extraordinary results from the right people at the right times, with the mix between the two determined by both the circumstances and the particular nature of the business.
If you run a company that will outlive you, or at least that will keep existing after you leave, there's a good chance that one of the hires you make will run the company, and will run it at a time that it's bigger than it ever was when you were around. Even if the founder doesn't directly hire a future CEO—not everyone wants to copy Elon Musk's policy of personally interviewing the first few thousand hires at SpaceX—the hiring process still affects what the company's internal talent pool will look like, and thus who is likely to end up in charge. On a very long timescale, the entire future of every business is determined one warm intro, cold LinkedIn-outreach, or engineering blog puzzle solution at a time.
Disclosure: I own shares of Amazon and Microsoft. I also do recruiting through The Diff, and this piece arose from thinking about why that economic niche exists in the first place.
A Word From Our Sponsors
Tegus is the first port of call for M&A professionals and institutional investors ramping up on an industry or company.
Get access to a database of 35,000+ expert call transcripts, spanning 5+ years, or schedule expert calls through the platform for a fraction of the usual cost.
When thousands of research analysts are pooling their expert calls into an on-demand database, using Tegus is table stakes. It's the leading platform for due diligence and primary research.
See the power of a Tegus subscription, and get up to data parity with your competitors, with a two week free trial through the Diff.
Smartphones and Distributed Law Enforcement
XKCD once pointed out that smartphones tell us useful things about the world by what phenomena they don't capture—every year since the ubiquity of the smartphone is further evidence against the existence of Bigfoot and the Loch Ness Monster. The broader implication is that there are many phenomena that have always existed but that are now much easier to observe and track. And New York City is considering using this to its advantage by giving people bounties for spotting parking violations.
The rise of cheap cameras in general should make it easier to detect crime, and it makes sense to give people an incentive to take advantage of this. (If there isn't a monetary incentive, the effect still exists, but it's concentrated among people who are very prone to noticing, or believing they've noticed, illegal activity, which is not ideal.) But the situation isn't static. The optimal punishment for some crime is a function of both the social cost and the likelihood of getting caught: the more likely someone is to get away with it, the higher the penalty should be, so the risk-adjusted incentive is optimal. And a natural consequence of this is that if it gets much easier to catch someone, the optimal punishment is much lower. The main critique of this kind of thinking is that people who break the law don't rationally weigh costs and benefits, but a steady stream of fines has a way of focusing the mind. And even if they don't, it means that there's essentially a steep tax on being indifferent to the rules, which can perhaps fund things like better driver education or more bike lanes.
Neal Stephenson's global neighborhood watch wasn't wrong, just early.
Actors as IP
A report claims that Bruce Willis has sold the right to his deepfake likeness (though maybe not). This comes just a week after news of a similar deal for James Earl Jones. When there's a new way to monetize existing IP, there's sometimes a land rush as people try to stake cheap claims to something they expect to be in use later. This tends to have mixed results; a few people made money buying domain names early in the dot-com era, but a) many of the big domain name sales were actually sales of a business that used the domain as a way to get traffic and search rankings, and b) anyone who realized this was a potential opportunity in, say, 1995 had much better options for capitalizing on the general trend. Of course, one thing that makes this kind of IP valuable is the very fact that people can own it: once you don't need Bruce Willis to make a Bruce Willis movie, you can rethink other things about the moviemaking process.
Twitch has launched a feature allowing users to pay to highlight their messages in chats. There are some sites that grow because users create, or feel that they're part of, a community. But community is intrinsically hard to scale, and is qualitatively different as it grows. In a small Twitch stream, the streamer can see and respond to everything viewers are saying, but as they get more popular, that gets difficult. If viewers value getting heard rather than merely saying things, Twitch is giving them a financial way to express this. (Perhaps this community-scalability problem is another use case for deepfakes! Some users would prefer a fake relationship with a real streamer, but others might want a more real-feeling relationship with a fake streamer.)
Google announced last week that it was shutting down its game streaming service Stadia. This raises some good questions about how big companies allocate capital to big bets: Stadia was an ambitious project, but it was also a network effects business where game publishers would need to make an investment to see results from the platform. Those kinds of businesses can be scaled quickly, but only at a high cost (the "Fresno Drop" launching credit cards is a classic example). For Google, one of the limits to building these two-sided platforms is the risk that services will get shut down. On the one hand, it would be hard to run a profitable company while maintaining every single service that had ever launched, just to signal to customers that they wouldn't pull the plug. On the other hand, every failed launch makes the next one even harder.
Since tech companies evolve so much, they're partly defined by their attitude towards these launches. Amazon seems to have structured itself around keeping new launches cheap and scaling the ones that work, and Stripe has written at length about the importance of maintaining backwards compatibility in payments APIs. There's a cost to doing this. In a way, Google's launches represent the same instinct applied differently: their services often get killed because they're designed to scale, which means they lose lots of money when they don't scale. So Stadia's shutdown represents the visible cost of a different kind of long-term thinking than the kind that led to its failure in the first place.
Streaming apps tend to focus most of their marketing on big, expensive shows, but some participants are aiming for lots of cheap filler, including Judge Judy spinoffs ($, WSJ). When streaming was getting started, there were two big components to the content decisions: first, streaming rights were underpriced, so it was cheaper than it should have been to assemble a good library. And second, streaming was a better vehicle for testing out riskier shows, both because the risks were lower with more data and because there was less friction for new users to sign up. As streaming has gotten bigger, it's unsurprising that its original content converges more with what's on large, popular channels, because it's grown big enough that the audience for those channels is now streaming, too.
Companies in the Diff network are looking for talent! Some current opportunities include:
- Frontend engineer interested in infosec and crypto, ideally with NextJS experience and good UX skills. (US, remote)
- Recruiter who can help enterprise software companies land key hires. (US, multiple locations)
- Multiple senior engineering roles (ML, infrastructure, systems programming) for a company building ambitious tools to increase developer productivity. (US, DC area)
- A quantitative developer with Python or C++ experience and some exposure to systematic investing. (Multiple roles, some remote)
- Someone with deep knowledge of life insurance regulation and an interest in crypto. (US, remote)
If you're interested in hiring through the Diff network, please reach out.
Since team performance follows a power law distribution, this can still be optimal overall, even if it's bad for individual firms. What it means, as a shorthand, is that picky employers who invest a lot in trying to hire the absolute best possible team will mostly fail, but the ones that succeed will make enough money to pay for elaborate recruiting processes. This is a good description of Jane Street: spending 2x the money to go from recruiting 99.8th percentile to 99.9th percentile people is worth it in a winner-take-all market. ↩
This can be a fuzzy signal, especially as FANG salaries have gone up. From a risk-adjusted compensation perspective, keeping a job at, say, Google, can be a great deal. In part because Google and other big tech companies got sick of losing good people to startups. It’s nominally possible to work at a high-paying job for a few years, then go to a startup, then go back, but it’s hard to outrun the hedonic treadmill. One source of potential energy is that people’s priorities change, and while big companies can afford to pay better than small ones, part of the reason they can pay better is that they get more value out of specialized and well-organized workers, so these jobs are partly a way to convert freedom into money. ↩
This ends up being a retention tool for those big companies: the more people can valuably specialize earlier in the career, the less room they have to be generalists. If you join a company as one of the first few employees, you get some exposure to sales/marketing, business operations, finance, etc. by osmosis. And you'll tend to observe that these things are not magic. But at a big company, they are compartmentalized enough to be basically magical to someone who doesn't work on them directly. ↩
If there are standard ways to recruit people, and they involve posting on high-traffic job sites and starting the filtering process by having candidates talk to non-technical extroverts, there will be a set of compelling potential employees who don't get reached by that process. And these employees can be very valuable, even ignoring the possibility that they're undervalued by the hiring process! In 2004, Google was doing some traditional recruiting, but was also looking to interview people who could find the first ten-digit prime number in the consecutive digits of e.
As the job market has gotten more efficient for the neuroatypical, the negative correlation between interesting work and high pay persists at the low end, but it flips to a positive correlation at the high end. The people making high-six or low-seven figures from working on distributed systems or modeling interest rate volatility tend to be the ones who find those problems genuinely interesting.
An underrated aspect of this rise of neuroatypical employment is that it leads to more robust abstractions throughout the economy—scalable storage and compute, digital payments, and hedging of obscure financial risks are better-defined and more tractable than they were a decades ago, and it’s easier to put a reasonable confidence interval around what they’ll cost if they’re the inputs into something else. And well-defined abstractions are composable, and can lead to new abstractions. So this kind of systematizing, intrinsically globalized, winner-take-most work creates more opportunities for more such work. ↩