Case Study 2 — The Restaurant That Hired Perfectly

A labeled composite: 130% turnover in a building where nothing was wrong with the hiring

This is a constructed composite, built from a pattern many operators will recognize rather than from a single real business. Every figure in it is illustrative and internally consistent, and the per-position separation costs are Chapter 17's, as allocated in §21.7. It is the restaurant Chapter 17 introduced at the end of its own analysis and handed to this chapter, because Chapter 17 could not fix it with anything Chapter 17 owns.

Background: a hiring operation you would copy

Call it a 110-seat full-service American restaurant in a strong urban neighborhood. Eight years old. \$2,100,000** in annual revenue, an average check of **\$40, roughly 52,500 covers a year, 44 employees, labor running 33.5% (\$703,500), food cost a respectable 30.1%. Prime cost in the low sixties. Not a distressed business. A working one.

The general manager was, genuinely, excellent at hiring. Chapter 17's entire toolkit was in place and in use:

• Job descriptions that described the actual work, the actual schedule, and the actual pay range, written from a task inventory rather than copied from a job board.
• Sourcing through three channels — a standing referral bonus, two culinary programs, and a relationship with a workforce nonprofit — rather than a permanent ad and hope.
• Structured interviews with a fixed question set, scored, conducted the same way for every candidate.
• Stages that were paid, time-boxed, and evaluated against a written rubric.
• Onboarding with a first-shift buddy, a training checklist, and a thirty-day check-in that actually happened, on the calendar, every time.

By any measure Chapter 17 could apply, this was a well-run hiring function. Time-to-fill was short. Offer-acceptance was high. New hires arrived competent and stayed through the training period.

And the restaurant turned over 57 people in a year on a 44-person roster — 130%.

The operating issue: what 130% actually cost

Here is the bill, at Chapter 17's per-position figures.

• 57 ÷ 44 = 130% turnover.
• \$78,750 ÷ \$2,100,000 = 3.75% of revenue.
• \$78,750 ÷ \$703,500 = 11.2% of the labor line.
• \$78,750 ÷ 52,500 covers = \$1.50 per cover.

A dollar fifty. Every guest who walked in the door paid a dollar fifty toward replacing the person who used to work there, and not one of them got anything for it.

For scale: this restaurant's operating profit in a good year might be five points — \$105,000. Turnover was consuming three-quarters of it.

The GM's read on the situation was the natural one and it was wrong: we have a hiring problem, and we need to hire better and faster. So the hiring function got sharper — which is why it was so good by year eight — and the turnover did not move, because hiring was never the broken part. Every one of these 57 people was hired well. They were also, every one of them, standing in a building that gave them a reason to leave.

The diagnosis: five findings, all of them the same finding

Chapter 17 named the five failures. Here they are again with the mechanism underneath each one.

1. Schedules posted Thursday for a Monday start

Four days' notice. Read §21.1's callout again for the mechanism, because it is the most important single item on this list.

An employee who cannot know their hours four days out cannot hold a class, arrange childcare, book a doctor, or take a second job with a fixed schedule. They can only take a second job with a flexible schedule — which means they now have two flexible employers and have to choose between them weekly. They will choose the one that treats them better, and if you are the one posting Thursday, you are not it.

The restaurant then experienced the result — callouts, no-shows, people unavailable at short notice — and drew the conclusion the entire industry draws: nobody wants to work anymore. What was actually happening is that the restaurant had structurally designated itself the backup employer and was receiving backup-employer behavior.

This is a self-inflicted wound and it costs nothing to close.

2. Reliability rewarded with more work at the same wage

The hero premium, exactly as §21.4 describes it. Two people on this roster — a line cook and a bartender — took nearly every callout shift between them, because they were the two who said yes. Neither had ever received a dollar, a title, a preferred day off, or the option to decline without awkwardness.

Both left within eleven months of each other. Together they cost \$2,850 + \$2,000 = \$4,850 to replace directly, and the eight weeks after each departure were the eight weeks in which the GM was personally covering shifts instead of doing the GM's job.

Note the perverse allocation this creates. The least reliable employee on a roster receives, in effect, a benefit: they are never called, because you have learned better. The most reliable receives a penalty. Nobody designed this. It is simply what happens when the callout list is "whoever is likeliest to say yes."

3. No promotion path

Eight years old, 44 employees, and there was no document anywhere in the building that told a busser what a server needed to be able to do, or told a line cook what would make them a lead. Promotions happened — the restaurant was not cruel, and good people did move up — but they happened when somebody left and a gap opened, which means they were events, not a path.

The difference matters enormously to a twenty-four-year-old deciding whether to stay another year. An event you cannot predict is not something you can plan around. A path is.

Figure 21.2 explains why small restaurants fall into this: a classical brigade had nine rungs and a 110-seat kitchen has four or five. The rungs were compressed out and nobody deliberately rebuilt them.

4. No real exit inquiry

One exit interview was conducted out of 57 separations, and it was conducted by the departing person's direct supervisor.

Fifty-six times, this restaurant generated its single most expensive piece of management information — a person who knows exactly what is wrong here and no longer has any reason to protect anyone — and threw it in the bin. Over eight years, call it four hundred data points discarded.

The cost of collecting them: thirty minutes each, on the clock, on a fixed form, by somebody other than the supervisor. §21.7 prices this at \$0.

5. A thirty-day check that asked "how's it going?"

The check-in happened. It was on the calendar. It was never skipped. And it produced nothing, because the question was unanswerable.

"How's it going?" costs the employee nothing to answer, reveals nothing, and lets both parties leave the room feeling that a conversation occurred. Thirty days in, a new employee has not yet decided whether it is safe to say a true thing, so they will say the cheap thing, and a question that permits a cheap answer will always receive one.

Compare §21.7's three questions. Every one of them is constructed so that a one-word answer does not work.

All five findings are the same finding. Nothing on this list is about who was hired. Every item is about what happened to a person after they were hired — the schedule, the reward for reliability, the visibility of a future, and whether anyone ever asked them a question that could receive a real answer.

What was tried first, and what it cost

Before the diagnosis, the restaurant tried three things. They are worth pricing, because they are what most operators try.

An across-the-board raise of \$1.00 an hour. Thirty hourly staff averaging 1,400 hours a year:

30 × 1,400 × \$1.00 = **\$42,000, plus a 12% payroll burden = \$47,040 a year.**

It bought approximately one quarter. Separations slowed in the first thirteen weeks and returned to trend by the second. Nothing about the schedule, the callout list, or the path had changed, and a dollar an hour does not compensate for not knowing next week.

A holiday party, roughly \$45 a head for 44 people = **\$1,980.** Fine. Not a lever. It is a thank-you and should be budgeted as one.

An employee-of-the-month program, \$100 a month = **\$1,200 a year.** It distributed recognition to twelve people in a year and implicitly withheld it from thirty-two, and by month five the staff could predict the winner.

Total spent on retention before the diagnosis: \$50,220. Separations that year: 57.

The fix, priced

Here is the same bundle as §21.7, scaled to a \$2,100,000 restaurant with 44 people.

\$30,382 — 1.45% of revenue**, against **\$78,750 of turnover cost.

The target. Separations from 57 to 33 — a 42% reduction, weighted toward the expensive positions, because the schedule and the ladder bite hardest on cooks and bartenders. That takes turnover from 130% to 75% (33 ÷ 44), which is merely industry-typical rather than good.

What you avoid: 2 sous/lead (\$12,000), 3 line cooks (\$8,550), 2 bartenders (\$4,000), 2 prep cooks (\$3,000), 7 servers (\$7,000), 2 hosts (\$1,820), 3 bussers (\$2,400), 3 dishwashers (\$2,100).

Avoided turnover cost: \$40,870. Cost of the bundle: \$30,382. Net: +\$10,488 — a return of 1.35× on the turnover line alone.

What it shows

One. The return on a retention program is a function of how bad you currently are. This is the finding worth carrying out of the case, and it explains an apparent contradiction in the chapter.

Bellwether, at 87% turnover, spends \$19,716 and avoids \$19,400 — it breaks even. This restaurant, at 130% turnover, spends \$30,382 and avoids \$40,870 — it returns 1.35×.

Same levers. Different answer. The levers are worth more where the bleeding is worse, which means the worst-run restaurant in your neighborhood has the highest return available on this spending and is the least likely to make it. If your turnover is already at or below industry norms, buy these levers for the second-order returns — food cost, ticket times, the owner's hours — and be honest that the turnover line alone will not carry them. If your turnover is 130%, the turnover line alone carries them easily and you are leaving ten thousand dollars a year on the table.

Two. Spending on retention is not the same as spending well on retention. This restaurant spent \$50,220 on a raise, a party, and a plaque, and got one quarter. The structured bundle costs \$30,382 — forty percent less — and addresses the five things people actually named on their way out. The difference is not generosity. It is diagnosis.

Three. Turnover cost is a per-cover number, and per-cover numbers are how you should argue this. \$1.50 a cover is a sentence a chef-owner will actually feel. "Our turnover is 130%" is a sentence they will nod at and forget. Chapter 17 built the per-position figures; this chapter's contribution is dividing by covers, because covers are the unit every operator thinks in.

The limits: what this fix cannot do

Be honest about three things, because a case study that ends in triumph is a case study you should distrust.

Turnover does not go to zero, and it should not. Some separations are structural and healthy: students graduate, people move, people leave the industry, and some people should be let go. A realistic floor for a full-service restaurant in most American markets is somewhere in the 40–55% range, not zero, and an operator chasing zero will start retaining people they should have replaced. The 75% target above is deliberately unheroic.

The levers take two to four quarters to show up. Turnover is a lagging measure of a lagging process. The schedule change is visible to staff in fourteen days and visible in the separation count in perhaps six months. An operator who buys the bundle and cancels it in month four because "it didn't work" has bought the cost and none of the return.

Some of these people were going to leave regardless. Of the 57, an honest read is that maybe 15 to 20 were leaving no matter what the schedule said. That is exactly why the target is 33 and not 12. Any lever priced against an unrealistic prevention count will look better on paper and worse in March.

And one thing this fix cannot touch at all: if the restaurant is genuinely understaffed — if the roster is short of the work, the way Chapter 19 found Bellwether's to be by 84 hourly hours a week — then no scheduling discipline produces a sustainable week. It only redistributes an impossible one. You have to fix the roster or you have to fix the work. Culture absorbs the gap until the people absorbing it leave, and then it converts into the table at the top of this page.

Discussion questions

1. This restaurant's hiring function was excellent by every measure Chapter 17 applies, and its turnover was 130%. Write the two metrics you would add to a hiring dashboard so that this failure mode would be visible from the hiring side rather than only from the P&L.

2. The across-the-board dollar cost \$47,040 and bought a quarter. The structured bundle costs \$30,382 and is projected to cut separations 42%. Construct the strongest argument for the raise anyway — there is one — and then say what you would do with it.

3. \$1.50 per cover is the turnover cost here; Bellwether's is \$1.05. Compute what the per-cover figure would be for a quick-service operation doing \$1,400,000 on 95,000 covers with 60 separations, using this chapter's per-position costs and a mix you construct. What does the comparison tell you about where retention spending pays hardest?

4. The case argues that the return on retention spending is a function of how bad you currently are. State the practical implication for an operator whose turnover is already at 55%. Should they buy these levers? On what basis?

5. Fifty-six of fifty-seven exit interviews were never conducted. Assume you inherit this restaurant next Monday and cannot interview anyone who has already left. Design a substitute instrument that uses only people who are still there, and state what it can and cannot tell you.

6. The GM in this case diagnosed a hiring problem and got better at hiring for eight years. Name the specific report or document that would most likely have corrected the diagnosis earlier, and say why nobody was looking at it.