Case Study 1: The Labor Market That Did Not Come Back

What happened to restaurant hiring after 2020, and what it taught operators about wages, schedules, and who was actually leaving

Background

In the spring of 2020, restaurants across the United States shed employment on a scale and at a speed without modern precedent. Dining rooms closed by public-health order, and the accommodation and food services sector — which the Bureau of Labor Statistics tracks as its own category — lost employment faster than any other major sector of the economy.

The conventional expectation was that this would reverse. Restaurant labor had long been treated, by operators and economists alike, as an abundant and highly elastic supply: when the rooms reopened, the workers would return, and wages would settle roughly where they had been.

That is not what happened. As dining rooms reopened through 2021, a large number of operators found they could not staff them. Reduced hours, delayed openings, and closed days appeared not because demand was weak but because there were not enough people willing to work the shifts at the wages offered. Reporting from the period is full of restaurants operating four days instead of six, and of dining rooms sitting half-seated while a waitlist formed at the door — the operational signature of a labor constraint rather than a demand constraint.

The operating issue

Three things were happening at once, and separating them is what makes the episode instructive rather than merely memorable.

Some workers had left the industry entirely. An extended shutdown gave a large population of restaurant workers something they had rarely had: a forced pause long enough to look at alternatives. Warehousing, delivery, retail, and healthcare support roles competed for the same labor pool, frequently offering more predictable schedules and, in some cases, comparable or better pay. A worker who spent eight months in a different industry and found the schedule survivable had little reason to come back.

The reservation wage had moved. Even among those who remained available, the wage at which they would accept a restaurant shift had risen. Operators who reposted their pre-2020 wage found the posting produced nothing, which is precisely the failure mode §17.2 describes — a posting that selects only from candidates with no alternatives, in a period when many candidates had acquired alternatives.

The non-wage terms mattered more than operators had assumed. Reporting from the period consistently surfaced schedule volatility, last-minute call-ins, split shifts, and the absence of benefits as reasons workers cited for not returning — often ahead of pay. This was the genuinely surprising finding, because the industry's standing assumption had been that wage was the only lever.

What it shows

First, that restaurant labor was never as elastic as its price implied. For decades, an operator who lost a cook could replace them quickly enough that the replacement cost never became visible — which is exactly the condition under which the \$38,070 in Figure 17.1 stays hidden. When replacement suddenly took eleven weeks instead of three, the cost of turnover stopped being an accounting abstraction and became a closed Tuesday.

Second, that the industry had been running an enormous unpriced liability. A business model that depends on a continuous inflow of replaceable people is stable only while the inflow continues. The moment it stopped, operators discovered that their labor model had a dependency they had never listed as a risk. Chapter 39's risk section exists partly because of this lesson: the assumptions most worth stress-testing are the ones nobody writes down because they have always been true.

Third, that the countermeasures were largely the ones in this chapter — and they were cheap relative to the alternative. Operators who responded successfully generally did some combination of: raising posted starting wages, publishing them in the posting, stabilizing schedules, adding some benefit structure, shortening the hiring process from weeks to days, and formalizing referral programs. None of that is exotic. Most of it is what §17.2 through §17.8 describe as ordinary competence, practiced under pressure.

Fourth — and this is the part operators most often get wrong in hindsight — a real wage reset occurred, and it did not un-occur. Average hourly earnings in the sector rose materially through this period, and while some of that reflected compositional shifts, a substantial part reflected a genuine repricing. An operator building a labor model today on pre-2020 wage assumptions is building on a number the market has moved past.

Outcome

By 2023 and 2024, sector employment had broadly recovered in aggregate terms, but the recovery was uneven and the terms had changed. Wage floors in many markets sat meaningfully above pre-2020 levels. Several states raised statutory minimums, and a number of jurisdictions moved on tipped-wage structures — Chapter 20 covers that framework, which varies enormously and continues to change.

What did not fully return was the assumption underneath the old model. Operators who staffed successfully in this period generally did so by treating hiring as a standing function rather than an emergency response — which is the argument §17.3 makes about pipelines, arrived at the hard way by an entire industry at once.

Lesson

A labor model is a set of assumptions, and the most dangerous ones are invisible because they have never been tested.

Bellwether's plan assumes it can hire eleven positions at market wages, in a specific city, in a specific spring, and that it can replace roughly twenty-seven people over the following year at the costs in Figure 17.1. Every one of those is an assumption. None of them is guaranteed, and the period after 2020 demonstrated at national scale what happens when the supply side of that assumption fails.

The practical response is not pessimism. It is the same discipline this book applies to food cost: name the assumption, price it, and know which way it moves. An operator who has computed their cost of turnover, who posts a wage, who keeps a bench, and who can fill a line-cook vacancy in ten days rather than eleven weeks has converted an existential dependency into a manageable one.

Discussion questions

1. Operators in this period frequently described the situation as a "labor shortage." Workers frequently described it as a "wage and schedule shortage." Both describe the same market. What does each framing cause you to do differently as an operator?

2. Chapter 17 argues that referrals produce the best retention at the lowest cost. Why would a referral pipeline have been more valuable than usual during this period — and why would it also have been harder to build?

3. The chapter's turnover ledger prices a line-cook replacement at \$2,180 assuming a three-week vacancy. Rebuild that figure assuming an eleven-week vacancy. What happens to the annual total in Figure 17.1?

4. Non-wage terms — schedule stability, predictability, benefits — emerged as central. Bellwether runs five dinners and two brunches with 31 people, and Chapter 14 showed a Friday where one absence put the sous on a fifteen-hour day. What schedule commitments could this restaurant credibly make, and what would each cost?

5. An operator says: "Wages went up, so I have to run leaner." Chapter 19 will test whether Bellwether can run at 32.3% labor. Is running leaner a coherent response to a labor shortage, or does it make the underlying problem worse? Argue both sides.

6. This case study relies on sector-level employment data and contemporaneous reporting rather than on a single documented company decision. What are the limits of reasoning about your own restaurant from aggregate data of that kind?