Case Study 2 — The Cost of Starting at Rung Seven

⚠️ This case is a clearly labeled constructed composite.

No restaurant described below exists. The operation, its people, its numbers, and its sequence of events are assembled from patterns this book has already taught — Chapter 11's variance decomposition, Chapter 15's bar leaks, Chapter 21's per-separation costs, and Chapter 34's investigation ladder. Nothing here is attributable to any real business, and no figure in it should be quoted as data. It exists to price a mistake that is difficult to price from real records, because restaurants that make this mistake do not publish what it cost them.

Background

Case Study 1 argued that the audit trail arrived in restaurants framed as surveillance. This case is what that framing costs when an operator acts on it.

The composite operation (constructed): a 96-seat neighborhood restaurant in its fourth year, food and beverage both healthy, with a bar that carries a genuine share of the business. Beverage sales run \$40,000 a period** across thirteen periods — **\$520,000 a year. Spirits and cocktails are about 57% of that. The target pour cost is 22.0%, which puts ideal beverage usage at \$8,800 a period.

The restaurant counts its bar. It counts it to the tenth of a bottle, on the last night of every period, by the tenths method Chapter 15 teaches. That is more discipline than most operations have, and it is why this particular failure is instructive: the counting was not the problem. The reading of the count was.

The operating issue

Here is the sequence, period by period.

FIGURE CS34.2-A — Pour cost, seven periods            [constructed composite]

  PERIOD   POUR COST   VARIANCE      WHAT HAPPENED
                       vs. 22.0%
  ────────────────────────────────────────────────────────────────────────────
   P3        22.7%       $280        Noted on the report. Nobody did anything.
   P4        24.1%       $840        "We have a theft problem." The phrase is
                                     used out loud, in the office, on P4's
                                     report day.
   P5        25.4%     $1,360        Six cameras installed over the bar and the
                                     liquor room. One bartender's closing
                                     shifts reviewed on video.
   P6        26.2%     $1,680        Termination. Counsel consulted afterward.
   P7        26.4%     $1,760        Variance unchanged.
   P8        26.1%     $1,640        Two more bartenders resign within nine
                                     weeks. Waste log goes blank.
   P9        26.2%     $1,680        Someone finally walks the ladder.
  ────────────────────────────────────────────────────────────────────────────
   P4-P9 leak that ran while the investigation looked at people:      $8,960

  Variance = (actual pour cost - 22.0%) x $40,000 of period beverage sales.

Read Period 7 twice. The variance did not move. A person was removed from the building and the number that removal was supposed to fix went up eighty dollars, and then stayed exactly where it had been.

That is the fact this case turns on, and it is the fact an operator who starts at rung seven almost never gets to see — because in most versions of this story the variance falls a little by coincidence, the owner concludes the problem was solved, and the actual cause runs for another two years underneath a resolved narrative.

What the ladder found

In Period 9 the restaurant's bookkeeper — who had never been consulted, and who is the only genuinely independent party in the operation's financial life — asked whether anyone had checked the cocktail cards. Nobody had.

One afternoon. Four documents the restaurant already owned. Here is the decomposition of Period 9's \$1,680.

FIGURE CS34.2-B — Period 9 pour variance, decomposed   [constructed composite]

  RUNG  CAUSE FOUND                                          AMOUNT   DOCUMENT
  ─────────────────────────────────────────────────────────────────────────────
   1    Cost cards stale. The well vodka and the sweet          $238   cocktail
        vermouth both moved on price in the spring; the                cards vs.
        cards still carried the old numbers.                           invoices
   2    Two seasonal cocktails ran eleven weeks and were        $274   specials
        never costed, so their pours have no theoretical               log vs.
        counterpart at all.                                            POS
   3    Mix drift. A $12 house cocktail with a 2.0 oz           $165   POS mix
        spirit build took share from the 1.5 oz well                   report
        highball. Nobody changed anything; guests did.
   4    Purchasing. One case of well gin invoiced above         $152   invoices
        the quote sheet, and a credit promised at the                  vs. quote
        door that never posted.                                        sheet
   5    Free pour. A week of measured pours found the           $636   a jigger
        house averaging 1.85 oz against a 1.5 oz spec.                 and one
        0.35 oz over, at $0.757 an ounce blended, across               week
        roughly 2,400 spirit pours a period.
   6    Unlogged waste. Breakage behind the bar, plus a         $141   the bins,
        CO2 regulator that failed on a Sunday and foamed              a receipt,
        out most of a keg before anyone noticed.                       a memory
  ─────────────────────────────────────────────────────────────────────────────
        EXPLAINED ON RUNGS 1-6                                $1,606
        UNEXPLAINED RESIDUAL                                     $74
        TOTAL PERIOD 9 VARIANCE                               $1,680

  The residual is 0.84% of $8,800 of ideal usage - inside the honest error of
  a tenths-method bar count.

  Every document in the right-hand column was in the building on the day the
  phrase "we have a theft problem" was first said out loud, six periods earlier.

\$1,606 of \$1,680 — 95.6% — was on paper the restaurant already owned. None of it required anyone's dishonesty, and none of it required a camera. The single largest line, \$636 a period, is the free pour Chapter 15 costs at Bellwether: it annualizes to \$8,268**, and the jiggers that close it cost **\$233 a year, a return of about 35 times.

The residual — \$74 a period, **\$962 a year** — is the entire quantity that was ever, at any point, unexplained.

What it cost

Now price the decision to begin at rung seven. Every figure is constructed; the structure is the lesson.

\$20,920 against a residual of \$962 a year. The investigation cost roughly twenty-two times the entire unexplained quantity it was chasing — and it did not resolve the residual either, because at the end of it nobody had learned anything about the residual at all.

Four of those five lines deserve a sentence each.

The leak is the largest cost and the least visible one. Every period an operator spends looking at people is a period the actual cause keeps running. The chapter's fourth argument for the ladder order — evidentiary cost — is usually read as a fairness point. It is also a speed point: rungs one through four are settled in an afternoon from documents you already have, and rung seven takes months. The ladder is faster, and faster is worth \$1,680 a period here.

The cameras were not wasted, exactly — they were bought for the wrong reason. Cameras at a receiving door and over a liquor room have legitimate uses, and this operation now has them. But \$2,760 was spent to answer a question that four pieces of paper would have answered for nothing, and it was spent at the moment of maximum certainty and minimum evidence. That timing is the tell.

The legal sequencing is the cheapest error to avoid and one of the most expensive to make. §34.9 puts it plainly: a conversation with an employment attorney before the first interview is cheap; the one after the termination is not. Employment law varies enormously by state and locality, and the exposures attached to a mishandled investigation — wrongful termination, defamation, wage claims among them — are real. Verify locally, and use an attorney rather than instinct.

The turnover is the part operators do not expect. One person was terminated. Three positions turned over. Chapter 21's argument is that the reliable people leave first, because they are the ones with options — and a bar staff that has just watched a colleague removed on a suspicion, in a room that suddenly has cameras and no explanation, updates its beliefs about the building immediately. The two resignations were not a protest. They were a reasonable inference.

And the unpriceable item is the worst one. After Period 7, the waste log went blank. Not maliciously — nobody makes a decision to stop filling in a form. The staff simply, correctly, inferred that written records were being read against people, and the rational response to that inference is to stop creating written records. Which means Periods 8 and 9 contain an unknown quantity of genuine, ordinary spoilage and breakage that is now numerically indistinguishable from theft, permanently, in the only records that exist.

The chapter's sharpest sentence describes exactly this: if you punish the waste log, you convert waste into theft — not in reality, but in your numbers. This operation did it to itself in nine weeks, and the two affected periods can never be decomposed by anyone, ever.

What this case does and does not establish

Be precise here, because the temptation to round this story off is strong in both directions.

This case does not establish that the terminated bartender did anything. It does not establish that they did not, either. What it establishes is more uncomfortable than either: nobody ever found out, and after the termination there was nothing left to find out with. The video was reviewed by people who had already reached a conclusion, which is the contamination problem §34.6 describes — an investigation that starts with a suspect gathers confirmation, not evidence. No independent explanation was written down and ruled out. Counsel was not involved until afterward. And the one measurement that would have been genuinely informative — does the variance fall when this person is no longer in the building? — came back no, and was not treated as information.

It also does not establish that theft never happens in bars. It does. Chapter 15 names sweethearting as one of five leaks and this chapter puts it in the front-of-house pattern table for a reason. The claim is narrower and stronger: theft is seventh because the base rates put it seventh, and this composite is what the base rates look like when one operator meets them. Ninety-six percent of the number resolved onto rungs one through six. The fact that some restaurant somewhere has a rung-seven problem does not license you to start there in yours.

And it does not establish that this operator was unusually foolish. They counted the bar to the tenth every period, which most operators do not. They noticed a trend at Period 4, which is fast. They acted rather than drifting, which is more than most. Every instinct they had was the industry's instinct, and the industry's instinct came, as Case Study 1 argued, from software marketing and a percentage-of-sales statistic nobody can source. The failure was not character. It was sequence.

The lesson

An investigation that starts at rung seven does not merely risk being wrong about a person. It systematically destroys the information you would need to be right about anything.

Three transferable points:

1. The order is a speed advantage, not only a fairness one. Rungs one through four are paper, settled in an afternoon at zero marginal cost, with no risk of being wrong about a human being. Rung seven is slow, expensive, legally consequential, and frequently ambiguous. Working the ladder in order is how you stop the leak sooner, and the leak is usually the largest number on the page — \$8,960 here against a \$962 residual.

2. Test the conclusion the way you would test any other. The single most useful measurement in this whole story was free: did the variance move? It did not. An operator who commits in advance to that test — "if this is the cause, the number falls next period; if the number does not fall, I was wrong" — cannot fall into the trap this composite fell into, because the trap requires never asking.

3. Protect the information supply above almost everything else. Your variance investigation depends on information only your staff can give you: the walk-in that ran warm, the keg that foamed out, the special that ran three weeks, the portioning drift they assumed you knew about. That supply is destroyed by a single visible episode of records being used against a person, it is destroyed in weeks, and it does not come back on its own. A blank waste log is the most expensive document in the building, and every entry in it is a gift.

Discussion questions

1. Identify the exact moment this went wrong. Name the period, the sentence, and the alternative action — and say how much it would have cost. Then answer the harder version: what would have had to be true about this operation's routines, not its judgment, for that alternative to have happened automatically?

2. The variance did not fall after the termination, and the operation did not treat that as information for two more periods. Design the commitment device you would put in your own operating procedures to force that test. Who writes it down, when, and what happens when the prediction fails?

3. Price the counterfactual. If the ladder had been walked in Period 4, which of the six causes would have been found immediately, which would have taken a week, and which would have needed a period of measurement before it was visible at all? Estimate what share of the \$8,960 was genuinely recoverable, and be honest about the part that was not.

4. The two periods with a blank waste log are marked unpriceable. Argue that this is a cop-out and produce a defensible estimate anyway — state your method and its assumptions. Then argue the opposite case: that putting a number on it would be worse than leaving it blank. Which position do you hold, and does your answer change if you are presenting to a partner rather than to yourself?

5. This operation counted its bar to the tenth of a bottle every period — better discipline than most restaurants have — and still reached the wrong conclusion. What does that tell you about the relationship between measurement and control? Name a control in this chapter that is purely about reading rather than collecting, and say what it costs.

6. Compare this composite to Bellwether's Period 8, where \$997 of a \$1,062 variance was explained on rungs one through six and the residual was \$65. Both restaurants had a real leak; only one of them spent \$20,920 finding it. List every structural difference between the two operations that made the difference — and identify which of those differences an operator can install in a single afternoon.

⚠️ Reminder: this case is a constructed composite.

The restaurant, the sequence, and every dollar figure above are constructed for teaching from the patterns in Chapters 11, 15, 21, and 34. No real business, employee, or event is described, and no figure here should be cited as evidence about the industry. The arithmetic is internally consistent and reproducible — that is its only claim.