How Companies Think About Layoffs

Plus! Moderation; Understanding TikTok; Reputational Bank Runs; Coal; Disney; Diff Jobs

'Tis the season-within-a-season where "Out of stock" or "Arrives after December 25th" starts to show up on more and more online product listings—but not on digital ones! If you've been enjoying The Diff and want to give a gift that will get opened again and again, you're in luck; we're running a 20% off holiday holiday discount for gifted subscriptions.

Programming note: this is 2023's last free edition of The Diff. Paying subscribers get three more posts this week. And we're back to the usual schedule starting Tuesday, January 3rd.

How Companies Think About Layoffs shows the kind of growth trend you never want to see: over the course of 2021, they tracked a total of 15,000 or so tech layoffs. This year, the number is 151,648. To put this in perspective, every quarter starting with Q2 of this year had more tech layoffs than all of 2021—and last quarter's total was higher than the Covid-era peak in Q2 2020.

It's worth understanding the layoff process from start to finish: how do companies decide to do a layoff (as opposed to freezing hiring, cutting other costs, or letting attrition slowly shrink the workforce for them). And that also entails understanding the end goal. The bad news is that layoffs happen, and the structure of layoffs means that plenty of good people will lose good jobs. Sure, it’s frustrating to get laid off and immediately see your former employer hiring for a job suspiciously similar to the one you just got pink-slipped out of (I've been there!)—but companies that do a round of layoffs are generally restructuring for growth—so it’s locally unfair, but generally positive.

The High-Level View

It's helpful to start with the 30,000-foot view before we get to the "this plane is heading straight for a mountain and we need to veer off in a different direction" view.

In a functioning job market, there's inevitable friction that makes it impossible for any company to have exactly the right staffing at any moment in time. Finding the right person can take an unpredictable amount of time, and when the economy is humming along, people have a habit of switching. That's especially true if these employees are underpaid—i.e. the better a deal a company is getting on someone right now, the more likely that person is to leave.[1] So the equilibrium is to slightly overhire. That's an especially easy choice for a growing company, because they can fix short-term mistakes through the bold decision to just slow things down a tad; a company that's 25% overstaffed but growing its revenue 100% annualized will solve the overstaffing issue organically in five months.

That friction also means that a healthy job market for workers is a nerve-wracking one for employers: wages are rising, and companies have to make hiring decisions in light of 1) the probability that many offers will get topped, and 2) the probability that some of their existing team will depart. Fortunately, this can be sustainable for a while when the perceived ceiling on employee productivity keeps rising. That's been increasingly true in software for a while; because when there's a SaaS company and an API for everything, far more scaling problems can be solved with money than ever before.

But this dynamic still relies on assumptions about the real world, and those assumptions won't always be borne out. Companies compete with peers on product quality and pricing, but they compete for talent and capital by being prudently optimistic. Whenever a big company does layoffs, it's common for people to say "Couldn't they have grown a little more slowly and never needed to do that?" And the answer is that there almost certainly was a competitor that grew more slowly and prudently, and that preference for stability over growth is why you haven't heard of them. In a space where every company is prudent and doesn't want to risk a layoff, the first company to defect and go for growth will swallow the entire market.

Outside of the industry cycle, there's a macro cycle, and that one is even harder for companies to avoid. Sometimes, their success is tied to real-world variables that they just don't have much control over. If a business operates in the travel space, there isn't much it can do about a pandemic; if a company sells electronics, it's probably counting directly or indirectly on China's manufacturing. For any B2B business, their revenue is part of somebody else's cost structure, so whether it's tied closely to transactions (e.g. payment processing, marketing) or indirectly to growth (almost everything else), a slowdown in corporate spending will have downstream effects.

When a company's revenue dashboard starts to persistently show that the numbers aren't going according to plan, decisions are required. That often involves diagnosing the problem, and figuring out why products aren't selling—is there new competition? A problem with an existing marketing channel? Or are customers just choosing to spend less? [2]

If there's less demand than expected, the plans the company made around its original demand assumptions are obsolete. And that means that the last capital raise, which they thought was about growth, is retroactively the money they raised to ensure survival.

The Mechanism: Crossing Names Off Lists

The first dreary point to make is that the layoff process doesn't start with layoffs. It starts with more generic belt-tightening. Years ago, I worked at a company that replaced the fancy organic snacks with Lay's, and was confused by the grousing that ensued. You have a great job and you're complaining that the free junk food you're getting isn't as good as the free junk food that you're used to! But it turned out there was a reason behind the grumbling: companies don't like to cut costs, and the last thing they want to cut is lots of headcount all at once, so people who've spent years in big companies become attuned to this kind of thing.

Perks are, as it turns out, partly a financing vehicle. And like many financing methods that early-stage companies use, they can be quite expensive. The basic process goes like this:

  1. A small company with negative cash flow knows that there's an exchange rate between dollars and survival. But not all dollars are created equal! Some kinds of spending can offset more than their cash-comp equivalent—especially if the early team a) has shared interests that make specific team outings or office features especially appealing to them, and b) have high enough morale (or big enough equity compensation) that they're not going to abuse an open-ended benefit like generous expense policies or unlimited time off.
  2. As the company grows, the perceived value of these benefits to each recipient drops. But their cost rises at at least the rate of headcount growth—and probably higher, if anecdotes about employees waiting until catered dinner arrives and then taking home enough food to feed their entire family are to be believed. Meanwhile, the company starts to run into the problem that in-kind gifts tend to be value-destroying relative to cash. (It's seasonally appropriate to cite this paper estimating the deadweight loss of Christmas, here.)
  3. But even when perks are no longer valued, getting rid of them sends a signal that the company either a) is suddenly worried about comparatively small sums of money, or b) is giving up on something that distinguished them from other companies. Both of these are bad for morale.

So generous perks early on are basically a way to borrow high morale from the future. Which can be a good deal! But it's still good to go into it knowing what the costs will eventually become.

When a company is growing and fundraising is easy, it's incredibly hard to resist the temptation to let costs creep up. But they can do so, in many different ways:

All four of those ways are hard to measure, and they’re all focused on the long term. But when capital is less abundant, the long term is, financially, farther away, and risk is more asymmetric to the downside.

So companies start taking a second and third look at discretionary budgets, and start rearranging their product priorities: features that might attract new customers are out; features that retain existing customers, or that reduce internal costs, are in. (This is one reason downturns can be so reflexive: when there are fewer new customers to reach and companies are fighting harder to retain their existing customers, that hurts growth for their competitors, too.)

And, if everyone is flying coach, employees are brown-bagging lunch, the corporate holiday party is more downscale ($, The Information), hiring has been frozen for a while, and margins are still looking bad—well, that’s when the headcount reductions begin.

There are a few different schools of thought for how to choose who stays and who goes. But the general outline is:

  1. Figure out what level of cost cuts would get the company to a more stable position. Stability depends on the business situation; it might mean having a path to cash flow breakeven, or it might mean having high enough margins to make the stock price tick up. (Or, for even more mature companies, it might mean being in a position to buy back lots of suddenly cheap shares.)
  2. Look at specific projects that could be canceled completely. Sometimes this means shutting down something that just got acquired, or canceling an R&D project with returns that aren’t expected for years.
  3. Meanwhile, make a list of unusually hard-to-replace people. This is a case where existing management has a serious advantage compared to, say, a private equity firm that's trying to clean house. Knowing that there's a system that only a handful of people understand, or knowing that there's a big customer who likes dealing with one specific salesperson, is the sort of local knowledge that's essential to reducing the negative impact of cuts.
  4. Making cuts that get refined, from "% of payroll" down to individual names, as they make their way down the org chart. This sometimes happens at the level of managers cutting some share of the payroll expense of their direct reports, but can also happen a level or two up, when it's more driven by performance reviews.

Since firing is uncomfortable and hiring processes are imperfect, there's a set within most layoffs that's mutually beneficial: employees who weren't a fit for whatever reason, and would be more valuable somewhere else, get a severance package that will probably last longer than it will take them to get their next job. In other cases, it's a mutual disaster; a good performer who happened to get a bad performance review, or had bad luck with the team they joined, ends up losing a good job and moving to a less optimal one. The more a company has optimized effectively for fast growth, the more likely it is that laid off employees will fit this category.

Another relevant category is the people who were a great fit for the company when they were hired, but aren't that great of a fit any more. Some people do better in 20-person companies than in 2,000-person companies, and a layoff can be a forcing function for both sides to recognize that.

The Aftermath

It's tautological to say that companies do layoffs in order to cut costs. Of course a business that's spending too much money will need to reset its cost structure. But the important, and more interesting, question to ask is what end state they're aiming for. In some cases, the end state is that the company is going to switch from growth mode to just harvesting cash flows. This can involve pretty brutal headcount reductions of 50% or more. And it's a case where outside managers tend to do a better job, both because they can be unsentimental about people and because they aren't wedded to the idea of running a growth company. (Though there are cases where a successful reset like this comes from within.)

What’s more likely, though, is that the real goal is to reset to a more modest pace of growth, and for an environment in which there isn't a wild bidding war for every employee or sales/marketing channel that could conceivably add to the growth rate.

This is tied to another accurate piece of conventional wisdom: the best time to cut costs is before the situation is desperate, and the optimal number of rounds of layoffs in a given economic cycle is exactly one.

Put those together, and the optimal layoff is a lot more extreme than the change in growth trend that justifies it. A company might see that revenue is coming in 5% below expectations based on their current spending trajectory, might then deduce that the right growth rate is 10% slower than the previous plan, and might further conclude that it's better to aim for 15% lower growth and then accelerate from there than to risk a second or third round of cuts.

A good broad model of layoffs is that companies try to operate with various kinds of slack, and there are tradeoffs between them: a company that hires fast and pays generously can avoid asking anyone to work nights and weekends, but if the cost of this isn't justified by the revenue and funding situation, this increase in employee slack comes at the cost of a radical decrease in financial slack. Another source of slack is customers and suppliers—the same increase in discount rate driven by higher funding costs can be reflected in vendor renegotiations or in tightening up customer refund policies.

When the layoff process is over, if it's done right, the result is a company that's lean. And just like it's easier to maintain healthy habits—physical or intellectual—than to create them in the first place, it's now much easier for such a company to stay on track. It persevered through a difficult experience, and everyone who made it through wants to avoid more of the same, and, importantly, they don't have any of the classic bull market excuses for sloppiness.

Layoffs are not just a way to make the opex numbers trend in the same direction as revenue again. They're also an admission that this relationship got out of whack—that the pace at which a company added costs didn't align with its economic opportunities. That doesn’t mean that the company's broad thesis is no longer intact. Think about it like this: since the core theses of different companies varies a lot, if a company lays off 20% of its workforce, it implies that the hiring decisions it made were only 80% right, i.e. the workforce reduction is just a pace correction, not a pivot to a new destination. And once it's on the right track and has reset costs, the next round of hiring can be done more judiciously, too.

Layoffs are a financially and emotionally expensive form of time travel, a way for a company to go back in time to when it was leaner and more focused, and to rebuild from there. It's an admission of error, but the main point of admitting mistakes is to fix them.

  1. This is one reason the economics of big tech companies are so wonderful: they don't have to have an edge underpaying most of their people, because the cumulative economics of the business are so strong; there's a lot more value to be added at a company with billions of users than millions of users, so there's an economic surplus involved in matching talent with those jobs, and the company can be generous with pay while still keeping most of the upside. On the other hand, these companies can be bad at giving their best employees a share of the variable upside they generate; even when a job is directly connected to revenue, or to the first-order inputs into revenue, compensation isn't a linear function of this. Finance, with its even more measurable outputs, can actually set up compensation structures along the lines of "take home a nominal salary plus 15% of whatever you make for us." Though financial employers generally develop a keen awareness of the dangers of giving anyone, especially a smart and ambitious person, a free call option. ↩︎

  2. Incidentally, this is one reason that good marketing analytics can be incredibly valuable for reasons that go beyond getting a good ROI on ad spend. The first macro indicator a company can get is customer reluctance to spend, so the better a company is at modeling the process by which marketing dollars turn into leads, customers, and repeat purchases, the sooner they'll know to take down job listings. ↩︎

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.



Twitter has banned promotion of, or links to, a roster of competing platforms including "Facebook, Instagram, Mastodon, Truth Social, Tribel, Post and Nostr" and "3rd-party social media link aggregators such as,," threatening to ban users' accounts if they violate the new rule, and then reversed this decision. This seems to have been written very hastily, given that it has an immediate caveat that cross-posting, even from such sites, is okay. (So you can't say "Find me on Mastodon," but you can move 100% of your posting to Mastodon and use a bot that cross-posts your Mastodon posts as tweets.) Meanwhile, Musk posted a poll on whether or not he should step down, promising to abide by the outcome.

Banning Twitter users from linking to competing sites is, of course, a very poor idea, both in the principled sense that limiting users' ability to share information is bad and even in the narrow sense that Twitter doesn't have the communications market share necessary to cripple competitors with this kind of move. It is also, directionally, something other social networks have done before: Instagram limits people's ability to link out to other content, which has created the whole "link-in-bio" industry exemplified by companies like LinkTree. And Facebook has blocked links to competing social networks Minds and Tsu. One issue is that the general pattern you'd expect to see from a social network growing through another social network's graph looks a lot like spam or hacking: when a URL suddenly gets posted more frequently, it might be organic but is probably something artificial, and the faster it grows the more likely it is to trigger automatic filters. So, other companies do this to a degree, but they're less direct about it.

One thing the executives at big platforms have to put up with is that, if they so choose, they could spend all of their waking hours reading criticisms of their own behavior and never run out. This is, among other things, a testament to the power of user-generated content and to the ubiquity of Internet access. It's also a decent tool, since the ebbs and flows of this commentary can tell managers if they've gone too far. For someone who started such a platform, usually controversial product decisions expose them to more public criticism than they've ever gotten in their lives, which tends to make them think twice about drastic changes. But Musk was famous/infamous before, and maybe the vociferous reaction here just feels like background noise.

This rule was quickly revised; part of Twitter's new operating cadence is that it's gotten faster at reversing itself. And in the meantime, it's hard for competing products to be a one-to-one substitute given Twitter's userbase; some users will check Mastodon rather than Twitter first-thing tomorrow, but if Twitter can produce more alerts and updates than Mastodon can, the gravitational pull of a larger userbase will continue to win. The biggest feature launch of Musk's tenure is the populist monarchy approach to governance, which gives everyone a lot to talk about. And that means that making bad decisions and then noisily reversing them can be more DAU- and revenue-accretive than not making a mistake in the first place.

In other Twitter news, the company is apparently seeking funding at $54.20, the same share price as the original deal. I wouldn't be surprised if the goal here were to get a handful of new investors in, maybe those more charmed than alarmed by Musk's post-acquisition behavior, in order to accurately say that the company raised a round at a flat valuation despite comparable businesses losing half or more of their value. (If this were a desperation move, the company would be raising some kind of preferred stock with a nice liquidation preference and other features, rather than common.)

Disclosure: long META.

Understanding TikTok

Does TikTok have an amazing, unique algorithm? Or does it use the same recommendation tools as everyone else, and have a UI that produces more recommendations? An important pair of feature of any recommendation system is:

  1. How long it takes someone to decide whether or not they like a given piece of content, and
  2. How actively they need to seek out the next piece of content.

Reddit's evolution is a touchstone here: the site used to link to lots of conventional long-form content, but since it takes so long to read something and form an opinion, it's hard for that content to reach the front page. (It would be interesting for them to do some kind of recommendation algorithm that estimated reading/processing time and weighted votes accordingly, so an upvote for a two-hour video would matter a lot more than an upvote for an image macro—but at that point, the game is to create long-form content that people will want to upvote as quickly as possible, the way Kindle Unlimited paid based on pages read, so authors used to publish a 3,000-page book where the first page said "check the last page to enter a contest and win a prize!" in order to instantly get credit for 3,000 pages.

Short-form video gets signals rapidly, since people consume so many videos per unit of time. And since it's easy to switch to the next, while continuing to watch is passive, the algorithm can rapidly identify which traits a viewer finds appealing. This is actually an interesting way to think about designing a service: in a case where user-level data makes a big difference, what design would lead users to share as much information as possible?

Reputational Bank Runs

Accounting firm Mazars Group has stopped working with crypto clients, while BDO is also reconsidering its crypto work ($, WSJ). Auditing itself is a kind of extension of credit, where the audit client borrows some trustworthiness from their accounting firm and then transforms this into more traditional borrowing. And, like other kinds of borrowing, it's subject to bank runs. The auditor's calculation might go like this: the client is probably safe, but crypto exchanges with leverage can face a bank run if crypto volatility leads to some customers losing more than 100% of their balance; if that happens while money is flowing out, the exchange can run into liquidity problems, and then the auditors will get blamed. It's safer to exit when this looks like a risk, and potentially return when the situation calms down. And this behavior, which is rational at the firm level, naturally exacerbates both crypto volatility and withdrawals from exchanges.


World coal consumption may set a new record this year ($, FT). Energy is often a story about elasticity. As a general rule, low-emissions energy sources either provide steady access to power but can't change their output much (nuclear, hydroelectric) or provide intermittent access at a low marginal cost (solar, wind). If you want power you can switch on or off on very short notice, you're probably going with gas or spending a lot on batteries. And if you want power you can turn on or off on longer notice—to replace, for example, suddenly much more expensive gas—then the best source is often whatever got shut down recently. And that is disproportionately coal. Moving away from fossil fuels will mean that there are lots of fossil fuel assets that are still in decent working order, and they'll end up being used to address spikes in demand.


The WSJ has a good look at the internal politics that led Bob Iger to become CEO again ($, WSJ). There are lots of takeaways from this: that it's hard to retire from a big institution, and that norms around this are helpful; that managing a company through a crisis means getting blamed for whatever goes wrong (Bob Chapek took over Disney, which makes lots of money from theatrical releases and theme parks, in February 2020); that the media industry has unusually vicious boardroom struggles. But it's also a story of investor relations: one of the criticisms of the Chapek era was that he tried to talk up the qualitative ways Disney was doing better at times when the quantitative metrics were worse. When a CEO has a combination of trust and charisma, this can definitely work, and can help the company pull through during rough patches. But in other cases, investors worry that the distractions imply that the real story is even worse than it looks.

Diff Jobs

Companies in the Diff network are actively seeking talent! If you're interested in exploring growth opportunities at unique companies, please reach out. Some top current roles:

If you’re 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.