Case Study 15.1: The Bar That Ran Out at 9:15

📜 From the Field

A Tier 3 illustrative composite. The couple, the venue, and the figures are constructed; the arithmetic reproduces from §15.2 and §15.3. The pattern — a bar sized by a rule of thumb that was right for a different wedding — is among the most common failures in the trade and it is entirely preventable.

The situation

Charlotte Nkemdirim and Adam Sowerby. 165 guests. A converted mill with an events licence, its own bar licence, and a policy of permitting client-supplied alcohol at $0 corkage.

That last fact is unusual and it is why they booked it. Their budget is $61,000 and the bar allocation is **$4,800.**

The planner, Devon, has done eleven weddings. For the bar he uses the figure he has always used, which came from his first employer and which he has never questioned:

"One bottle of wine per two guests, one case of beer per twenty, and a bottle of spirits per twenty-five. It's never let me down."

At 165 guests that produces:
Wine 83 bottles
Beer 8 cases (192)
Spirits 7 bottles
Estimated servings 415 wine + 192 beer + 112 spirits = 719

Devon buys it, adds $220 of mixers and garnish and $90 of ice, and comes in at $1,840** — well under allocation. The venue's own bartenders are $58/hour each; he books three for six hours: $1,044.** Total **$2,884** against $4,800.

He tells the couple they are $1,900 under and they put it toward the band.

The day

The ceremony is at 3:30 and ends at 4:05. The bar opens at 4:05. Dinner is at 6:30. The band starts at 8:45. The bar closes at midnight.

That is a seven-hour and fifty-five-minute bar.

Time What happened
4:05–5:30 Cocktail hour. Long queue but it moves. Roughly 300 drinks go out
5:30–6:30 Extended drinks reception because the photographer runs over. Another ~140
6:30–8:15 Dinner. No wine on the tables — everything from the bar. ~150
8:15–8:45 Speeches, then a gap. ~60
8:45–9:15 Band starts. The floor fills. ~90
9:15 The white wine runs out. Then the beer, at 9:40
9:40 Devon sends the best man's brother to a supermarket 20 minutes away
10:35 He returns with what a supermarket has at 10 p.m. on a Saturday
12:00 Bar closes

Total consumption, as counted afterward: 1,040 drinks.

Devon bought 719.

What the forecast would have said

💰 Run the Numbers: the forecast Devon did not do

Drinkers: 165 guests, 9 children, 11 non-drinking = 145.

Duration: 4:05 p.m. to midnight = 7 hours 55 minutes. Call it 7.9 hours.
First hour: 145 × 2 290
Hours 2–7.9: 145 × 6.9 1,001
Base 1,291

Adjustments: mixed ages (0), no shuttle but the venue has rooms (+10%), September indoors (0), long extended cocktail period +0.5/guest (+73), dancing-heavy (+10%).
Base 1,291
Rooms on site, +10% +129
Extended reception +73
Dancing, +10% +129
Adjusted 1,622
Shrinkage, +10% +162
Forecast 1,784

The forecast is 1,784. Actual consumption was 1,040. Devon bought 719.

Three numbers, and the gaps between them are the whole case.

🔍 Why Does This Work? The forecast over-shot actual consumption by 71%, and it still would have prevented the failure.

This is the point most planners miss about forecasting. The purpose of the model is not to predict consumption precisely. It is to establish an order of magnitude that a rule of thumb cannot reach.

Devon's rule produced 719. The model produces 1,784. Actual was 1,040. The model is wrong and the rule is catastrophically wrong, and the difference is that the model's error is in the safe direction and is recoverable at the retailer — 83% of a wine over-buy comes back as a refund.

And the model's over-shoot is diagnosable. It came from applying the full 1.0/hour rate across a 7.9-hour bar, when the two hours of dinner run at roughly half that. Correcting for dinner:

1,784 − (145 × 1.75 hrs × 0.5) = 1,784 − 127 = 1,657 — still high, and now the residual gap is the model's conservatism plus a guest list that drank less than its demographics suggested.

A planner who forecast 1,784, bought 1,300 on a returns agreement, and understood where the 484 of slack was, would have had a fine evening and a $600 refund.

The four errors

1. The rule of thumb was duration-blind.

"One bottle per two guests" contains no reference to how long the bar is open. It was learned at a company that did four-hour hotel receptions. Devon applied it to a bar that was open for nearly eight hours and it was wrong by exactly the ratio you would expect.

2. Nobody counted the hours.

The bar opened when the ceremony ended, not when the reception started — and nobody wrote down that this was 4:05 p.m. The bar's operating window was never a number in any document. It emerged on the day.

3. Wine was not on the tables.

165 guests, no wine on the table, plated dinner. Every glass across a 105-minute dinner was a bar transaction — 150 drinks and roughly 90 separate trips, at exactly the moment the room should have been seated. (§15.3a.)

4. There was no re-order plan.

The most fixable one. A retailer who agrees to a returns policy will usually also agree to hold a reserve — a further ten cases, uncollected and unpaid, available until close of business Saturday. It costs nothing. It converts a catastrophe into a phone call at 6 p.m.

Instead, a guest drove to a supermarket at 9:40 p.m. and came back with what a supermarket has at ten on a Saturday night, which is the wrong wine at the wrong price, and — worth stating plainly — a guest at a wedding drove a car at 9:40 p.m. on an errand the planner set them. Chapter 9's liability chain runs directly through that decision.

Discussion Questions

1. Devon's rule "had never let him down" across eleven weddings. Explain how a genuinely bad rule survives eleven trials.

Discussion **Because it is not a rule about consumption. It is a rule that encodes a specific event shape**, and eleven of Devon's twelve weddings had that shape. His first employer did hotel receptions: a four-hour bar, a package, wine on the tables, a hard 11 p.m. close. At four hours with 130 guests, "one bottle per two" produces 325 servings against a forecast of roughly 400 — **short, but within the venue's own inventory**, which absorbed the gap invisibly. Devon never saw a failure because a hotel bar cannot run out. **The mill is the first venue where the inventory was his.** The rule had been failing for eleven weddings and had been silently corrected by someone else eleven times. **The general form, and it is worth carrying:** a heuristic that works is not evidence that the heuristic is right. It may be evidence that something else in the system is absorbing its error. **Ask what would have to be true for the rule to fail, and then ask whether you would have noticed.**

2. The model over-forecast by 71% and the case argues it should have been used anyway. Is that defensible, or is it advocating a bad tool because it fails safely?

Discussion **Both, and the honest answer has to hold them together.** The model *is* imprecise. Its error here was large enough that a planner who trusted it literally would have bought 60% more alcohol than the wedding drank, which at a venue without a returns policy would be $700 wasted. **What makes it defensible is not the number. It is that the number is decomposable.** Devon's rule produces one figure and no explanation. The model produces a figure and seven lines, and each line can be argued with: - *Is 7.9 hours right?* — It is, and writing it down is most of the value. - *Is the dinner hour really 1.0?* — No, it is about 0.5. **Correct that line and the forecast falls to 1,657.** - *Is a +10% dancing adjustment right for this crowd?* — Arguable, and now somebody is arguing about it. **A wrong number you can interrogate beats a wrong number you cannot.** That is the actual claim, and it is narrower than "the model works." **The correct practice**, which §15.3 states and which this case demonstrates: **forecast high, buy at 70–80% of forecast on a returns agreement, and hold a reserve.** The model sets the ceiling; the purchase decision is separate; the reserve covers the gap between them. Devon's failure was not that he lacked a model. **It was that he had no ceiling, no returns agreement, and no reserve** — three independent safeguards, none of which existed.

3. The bar's operating window was never written down. Where should it live, and who is responsible for the fact that it was 4:05 rather than 5:00?

Discussion **It lives in three places and it has to agree in all of them:** the bar service's contract (as billable hours), the run sheet (as a line with an owner), and the forecast (as the duration input). The reason it was 4:05 is that **nobody decided the bar would open when the ceremony ended.** It opened because there was a bar, guests walked past it, and a bartender was standing there. **The default state of an unmanaged bar is open.** **Whose responsibility:** the planner's, unambiguously, and it is a five-word line on a run sheet — *"Bar opens 5:00, not before."* **And there is a real argument for opening it late.** Fifty-five extra minutes at 145 drinkers is roughly 145 drinks — **$400 of alcohol and, more to the point, 145 drinks consumed by people who then have to get through another seven hours.** §15.6's rate argument applies at the front of the evening as much as at the end. **The counter-argument, which is also real:** guests arriving from a ceremony with nothing in their hands and nothing to do is a bad twenty minutes, and this is exactly what passed drinks are for. **The answer is not a closed bar; it is a controlled opening** — trays circulating from 4:05, bar opens at 4:45.

4. Devon sent a guest to a supermarket at 9:40 p.m. Analyse this decision.

Discussion **It is the worst decision in the case, and it is the one most planners would make.** **What is wrong with it:** 1. **A guest drove.** At 9:40 p.m., from a wedding, on an errand the planner asked for. If that car had been in a collision, the chain of exposure runs through the person who asked — and Chapter 9 §9.1's rule is *coordinate, do not perform*, which extends to *do not deputise a guest to perform.* 2. **The best man's brother was drinking**, almost certainly. Devon did not ask. 3. **It took 55 minutes** and returned inadequate product. The problem was not solved; it was moved. 4. **It made the failure public.** A guest sent on an emergency alcohol run tells nine people why. **What should have happened instead**, in order: - **The reserve** (see error 4) — a phone call at 6 p.m. when the trend was visible, not at 9:40 when it was a fact. - **Failing that: the venue.** The mill has its own bar licence and therefore its own stock. **Buying 40 bottles at the venue's retail price at 9:15 p.m. is expensive and it is the correct answer** — perhaps $500 against a $1,900 underspend that had already been given to the band. - **Failing that: a taxi**, paid for, to an off-licence. Not a guest, not a car, not free. - **Failing all of that: tell the couple, close the wine, and run beer and spirits.** A bar that has run out of one thing at 9:15 and says so is a smaller failure than a bar that produces a crisis. **The general rule: never solve an event problem by putting a guest in a car.** There is always another answer and it is always cheaper than the one where you are wrong.

5. The couple gave the $1,900 underspend to the band. Whose decision was that, and what should Devon have said?

Discussion **The reallocation was the couple's decision to make and Devon's information to supply — and the information was wrong.** He told them they were $1,900 under. **They were not under; he had under-bought by roughly $900 of alcohol.** The surplus was an artifact of a bad forecast, and a reallocation made on a bad forecast is a decision made on false information, which is Chapter 3's territory rather than Chapter 6's. **What he should have said:** > "We're showing about nineteen hundred under on the bar. Before you spend it, give me two days — **I want to re-forecast against the actual hours**, because the bar's open from the end of the ceremony and that's nearly eight hours, which is longer than I first costed. **I'd rather tell you the real number than a good one.**" **And then the harder version, once he had re-forecast:** > "I got this wrong and I want to correct it before it matters. I costed the bar on a rule of thumb rather than on the hours, and the hours here are unusual. **The real number is about thirty-two hundred, not eighteen-forty.** That still leaves you about sixteen hundred under, and I'd want to hold three hundred of it as a reserve we don't spend — I can arrange stock we don't pay for unless we open it." **Chapter 3's bad-news promise is what makes the second version possible**, and this is exactly the kind of error it was designed for: a mistake found by the planner, before the client feels it, corrected with a number attached.

Connections

  • Chapter 3 — the bad-news promise, and correcting your own error before it lands
  • Chapter 5 — where the count of non-drinkers comes from
  • Chapter 9 §9.1 — coordinate, do not perform, and its extension to guests
  • Chapter 11 — the queue that moved but should not have had to
  • Chapter 14 — the dinner block, and why 105 minutes of dinner produced 90 trips to a bar
  • Chapter 26 — the run sheet line that says "bar opens 5:00, not before"
  • Chapter 28 — the reserve, as an instance of the general failure playbook