Case Study 2 — The Restaurant That Hit Its Labor Target

A constructed composite, built from patterns that recur across independent full-service restaurants. It is not any real business, and every figure in it is illustrative.

Case Study 1 looked at an industry that could not buy the hours it needed. This one looks at the opposite and more common failure: an operator who could buy the hours, decided not to, hit the number he was managing to, and closed anyway. It is the chapter's warning made specific — the "fifth lever" from §19.8, followed all the way to the end.

Background

A 74-seat chef-driven American restaurant in a mid-size market. Opened with a partner-chef and a managing partner. Dinner six nights, no lunch, a small bar. Good press in the opening year, a neighborhood following, a menu built on scratch production — house pasta, whole-animal butchery, stocks made in-house, bread baked daily.

Two years in, the business was working but tight. Here is the second full year, months 13 through 18, as the managing partner saw it:

Read it the way Chapter 1 taught you. Nothing here is catastrophic. Cost of goods sold at 29.6% on a scratch kitchen is genuinely good. Occupancy is fine. But prime cost is 64.2%, which is four points above where a full-service restaurant wants to be, and the whole overage sits in one line.

The managing partner did the obvious arithmetic. Labor at 34.6% against a 30% benchmark is 4.6 points, and 4.6 points of \$1.42 million a year is about \$65,000 — more than half the annual operating profit. He set a target: get labor to 29%.

He was not wrong about the arithmetic. He was wrong about what a labor percentage is.

The operating issue

The reduction took twelve months and it was done exactly the way these things are always done — not by decision, but by accumulation. Nobody ever wrote a memo saying "we are removing eleven positions." Each individual step was small, defensible, and saved real money.

Months 19–21. The two bussers were not replaced when they left. Servers picked up their own tables. The saving was about \$620 a week.

Month 21. Everyone's start time moved back thirty minutes. The pre-service block — the shaded half-hour in Figure 19.6 — came off every shift in the building. Twelve positions, five to six nights: roughly \$430 a week. Nobody complained, because nobody could point to a guest who was affected.

Month 22. The host stand was cut on Tuesday, Wednesday, and Thursday. The managing partner covered it himself when he was there, and a server covered it when he wasn't. About \$260 a week.

Month 24. The pantry cook's position was eliminated and the station folded onto sauté. The four-top line became three. This was the biggest single move, about \$720 a week, and it was the one the chef argued about hardest and lost.

Months 25–27. The second dishwasher came off Friday and Saturday. The prep cook's Saturday shift was cut, and Saturday's prep moved onto the line cooks' pre-service time — which no longer existed, so it moved into service. About \$470 a week.

Month 27. The chef stopped taking two days off and started expediting five nights instead of four, because the sous had left and had not been replaced.

Twelve months. About \$2,500 a week** removed from the schedule, roughly **\$130,000 a year, and every dollar of it was real.

What it shows

Here are the three periods side by side. The managing partner's monthly report showed him the top half of this table. The bottom half he was not tracking.

Period B looked like a triumph, and that is the whole problem

From A to B, labor fell 2.8 points, prime cost fell 2.5 points, and operating profit rose from \$54,824 to \$72,374 — up 32% on essentially flat sales. Every number the managing partner watched moved the right way. Sales were down 1.3%, which he attributed to a soft February.

This is what makes the pattern so dangerous. The first half of a labor reduction genuinely works. There is real slack in most schedules — unnecessary overlap, positions scheduled by habit, hours that produce nothing. Cutting that slack produces exactly the result the arithmetic predicts, and it teaches the operator that the arithmetic is reliable.

It is reliable right up to the point where you stop cutting slack and start cutting the product.

Period C: the number was hit, and the business was gone

Labor reached 29.4%. Target achieved. And:

• Sales fell \$132,000 — 18.8% — against a labor saving of \$55,680. The restaurant traded away roughly two dollars and thirty-seven cents of revenue for every dollar of labor it saved.
• Cost of goods sold rose 2.5 points, from 29.9% to 32.4%. Nobody changed a spec or a price. A kitchen with no pre-service prep time, one dishwasher, and a pantry station folded onto sauté over-portions, over-produces, wastes trim, and 86s items it then has to re-fire. On \$571,000 that is about \$14,275, and it appeared on a line the managing partner was not connecting to the schedule.
• Prime cost did not improve at all between B and C — 61.7% to 61.8%. The labor savings went straight into cost of goods sold. The two halves of prime cost traded, exactly as Chapter 1 warned, and watching only one half made the trade invisible.
• Occupancy went from 8.2% to 10.2% without the rent changing by a dollar. \$58,384 is \$58,384; it only looks worse because the denominator collapsed. This is the same ratio trap as the labor percentage, running in the opposite direction, and it is the tell that the problem is revenue.
• Other operating rose 3.1 points, of which roughly \$18,400 was recruiting, agency placement fees, and training for nine back-of-house departures in six months. Chapter 17's cost-of-turnover arithmetic was being paid in full, on a line labeled something else.
• Operating profit fell to \$26,695, roughly half of the baseline period and barely a third of the period the managing partner considered his best.

The restaurant closed in month 31, in a conversation with a landlord, with a P&L that had shown a profit in every single period.

Outcome

Read the last row of the table again: nineteen of twenty-six weeks with a scheduled position unfilled. By Period C the restaurant was no longer choosing to run short. It could not hire. Word travels in a market this size, and a kitchen known for nineteen-hour prep days and a station folded onto sauté does not attract the cooks who could fix it. The labor reduction became self-sustaining, which is the point at which an operator has lost the ability to reverse the decision.

The chef left in month 29. The menu that had earned the press in year one required a four-station line and a prep cook, and by month 29 the restaurant had neither. What it was serving was a recognizable but degraded version of itself, at the same prices, to a shrinking number of people who had been coming for two years and had begun to notice.

None of that appears on a labor report. All of it appears in the sales line, roughly nine months after the decision that caused it.

The lesson

A labor percentage is an output, and you cannot manage an output directly. When you try, you get exactly what this restaurant got: the number moves, and the thing the number was measuring moves the other way.

Three specific transferable rules.

1. Distinguish slack from product. The first \$1,000 a week you remove from a schedule is usually slack — hours that produce nothing. The next \$1,000 is usually product. There is no line on a schedule that says which is which, so you have to know your staffing guide well enough to say what each position produces. The pre-service half-hour looked like slack. It was the entire quality control system of a scratch kitchen.

2. Watch prime cost, not labor. From B to C, labor fell 2.4 points and cost of goods sold rose 2.5. Prime cost was flat. Every hour of prep and portioning that came off the schedule reappeared as waste, over-portioning, and re-fires. An operator watching only labor saw a triumph. An operator watching prime cost would have seen nothing happening at all, and would have asked why — in month 25, when there was still a business to save.

3. Instrument the things a labor report cannot see, before you start cutting. This restaurant had the data the whole time: ticket times in the point-of-sale, comps in the daily report, review scores in a free dashboard, departures in the payroll file. Nobody put them on the same page as the labor percentage. Had they been on one page, the divergence would have been obvious in month 22 — labor down, ticket times up, comps up, reviews down, all at once, in the same direction, for four months running.

That one page is not hard to build. It is the labor report from Figure 19.9 with four rows added.

Discussion questions

1. Identify the exact point in the twelve-month sequence at which this operator crossed from cutting slack to cutting product. Defend your answer with the data in the table, then say what evidence available at that moment would have revealed it.

2. From B to C, labor fell 2.4 points and cost of goods sold rose 2.5. Explain the physical mechanism — what actually happens in a kitchen with no pre-service prep time and a folded pantry station that converts labor dollars into food cost dollars. Be specific about at least four distinct paths.

3. Occupancy rose from 8.2% to 10.2% without the rent changing. Explain why, and then explain why this particular movement is one of the most reliable early warnings on a restaurant P&L.

4. The managing partner's original arithmetic — labor 4.6 points above benchmark, worth \$65,000 a year — was correct. Where exactly did the reasoning fail? Write the two sentences you would have said to him in month 18.

5. Bellwether's plan requires removing about 110 hourly hours a week in weeks 14–52. That is a larger proportional cut than this restaurant made. Using this case, argue for or against attempting it, and specify what you would measure weekly to know whether you were cutting slack or product.

6. This restaurant closed with a profitable P&L. Chapter 1 argued that a restaurant does not need to be unprofitable to close — it needs to be insufficiently profitable when something ordinary goes wrong. What was the "something ordinary" here, and was it actually ordinary?