Case Study 2: Three Ways a Forecast Breaks

A composite study of the recurring failure patterns in restaurant revenue projections — and the decision at the end of it that has no clean answer

This case is a labeled composite. The three restaurants below are constructed teaching examples. They are not real businesses and their numbers are illustrative. The patterns they demonstrate are real, recurring, and described consistently by lenders, restaurant accountants, and operators who have written a plan and then lived inside it. Every figure in this case is internally consistent and computes; none of it is anyone's actual financials.

Background

Case Study 1 looked at the forecast from the reader's side: what a lender demands and why a package fails. This one looks at it from the writer's side, which is where the damage is actually done.

The uncomfortable fact underneath this chapter is that most first restaurant forecasts are wrong in the same direction and by roughly the same amount. They overstate. Not by a little. The three patterns below each produce an overstatement in the twenty-five to thirty percent range, and they produce it through completely different mechanisms — which is why an operator can avoid one and be destroyed by another.

Each pattern is presented with the arithmetic that produced it and the arithmetic that should have.

Pattern one: the market-share forecast

The composite. A 90-seat casual American concept, dinner six nights a week, an intended check average around $34. The plan's revenue section reads, in its entirety:

"Metro-area restaurant spending is approximately $380 million annually. At a conservative 0.45% market share, the restaurant projects year-one revenue of $1,710,000."

The operating issue. That paragraph contains no seats, no turns, no check average, and no calendar. It cannot be checked, and — this is the part that matters — it could not be managed against either. There is no Monday on which anyone can ask whether the restaurant is on plan, because "market share" is not a thing a general manager can observe.

Build the same restaurant from the bottom up and watch what happens.

The plan says $1,710,000. The physical build says $1,241,136. The gap is $468,864 — 27.4% of the headline.

Worse, run the arithmetic backwards and find out what the plan was silently assuming. To produce $1,710,000, weekly revenue must be $32,885, which is $5,481 per service, which at a $34 check is 161 covers, which on 90 seats is 1.79 turns. Six nights a week. Every week of the year.

The plan never wrote "1.79." Nobody in the room ever said it out loud, so nobody ever had the chance to say that is not what this room does. A top-down forecast does not eliminate the turns assumption. It hides it.

Pattern two: the opening spike as the baseline

The composite. An 80-seat restaurant opens in April to genuine local enthusiasm. The first six weeks average $34,000 a week. The owner, reasonably pleased and unreasonably confident, revises the plan: $34,000 × 52 = **$1,768,000**, and staffs the building accordingly.

The operating issue. An opening is not a baseline. It is a curiosity event, and curiosity is finite. Here is what the year actually did.

Check it: 6 + 20 + 14 + 12 = 52 weeks; $204,000 + $530,000 + $322,000 + $252,000 = $1,308,000.

Against the annualized spike of $1,768,000, that is **$460,000 short — 26.0%.**

And here is the part that costs real money, which is not the revenue miss at all. The owner hired to $34,000 a week. The schedule was built for a room doing 34 covers an hour, and the schedule is sticky: those are people with lives, and cutting them is slow, painful, and damaging to a culture three months old.

Watch the arithmetic. A labor budget of 30% on $34,000 is $10,200 a week. Run that same $10,200 against $26,500 of actual revenue and labor is 38.5% — eight and a half points over target. Held across the twenty settling weeks, the overstaffing costs $2,250 a week, or **$45,000**, which is very close to the working-capital reserve the business opened with.

The revenue miss was survivable. The revenue miss plus a cost structure built for the wrong revenue is what closes restaurants, and it is the exact mechanism Chapter 1 described as the fatal pattern.

Pattern three: the year with no calendar in it

The composite. A 64-seat restaurant, dinner six nights, a $38 check. The plan projects 2.0 turns, which the owner describes as "achievable, since we're only open four hours."

$64 \times 2.0 = 128$ covers × $38 = $4,864 per service × 6 = $29,184 a week × 52 = $1,517,568.

The operating issue. Two separate errors, and they compound.

The room cannot do 2.0. With a 105-minute average table time and a four-hour service window, the theoretical ceiling is 240 ÷ 105 = 2.29 turns — but only if the room fills and empties in perfect lockstep, which no room does. Real rooms have a 6:00 that is quiet, a 7:30 that is full, and a 9:00 that is thinning. Chapter 24 works this properly; the practical number for a concept like this is somewhere near 1.45.

The calendar has 52 weeks in it and the restaurant does not. Two closed weeks — a holiday, a hood cleaning, a compressor — is a normal year, not a bad one.

Rebuild with both corrections: $64 \times 1.45 = 92$ covers × $38 = $3,496 per service × 6 = $20,976 a week × **50** weeks = **$1,048,800.**

Against the plan's $1,517,568 that is **$468,768 short — 30.9%.**

Notice that neither correction was a judgment about the concept, the food, the location, or the owner. Both were arithmetic about a physical room and a calendar. The most expensive errors in a restaurant forecast are usually not errors of optimism about the business. They are errors of optimism about time and space.

What it shows

Put the three side by side and a shape appears.

Three different mechanisms, three different operators, three overstatements clustered between twenty-six and thirty-one percent.

That clustering is not a coincidence and it is not a law of nature. It is what happens when a forecast is built from a desired outcome rather than from a physical constraint, because the desired outcome is always roughly "enough to make this work," and "enough to make this work" is reliably about a quarter above what the room will actually produce.

Each pattern also has a specific tell that a reader can spot in under a minute:

• The market-share forecast contains no seats, turns, check, or days. Ask for them and the plan cannot answer.
• The annualized spike has a single flat weekly number and no seasonality. Ask what February does and there is no answer.
• The no-calendar forecast has a turns number nobody has converted into minutes. Ask what the average table time is and how many minutes are in the service, and the arithmetic collapses on the spot.

Each tell is also a question you can ask of your own plan before anyone else does.

The contested decision

Now the part that has no clean answer, because this is where honest operators actually get stuck.

Take pattern three's restaurant. The honest bottom-up build is $1,048,800. Its fixed cost base is $480,000 a year, every marginal sales dollar leaves about 48¢ after variable costs, and its debt service on the loan it is applying for is $74,000 a year.

Run it honestly:

• Contribution: $1,048,800 × 0.48 = **$503,424**
• Less fixed costs: $503,424 − $480,000 = $23,424 of operating profit
• Less debt service: $23,424 − $74,000 = −$50,576

The honest forecast does not service the debt. It is not close.

Now solve for the number that does. Required contribution is $480,000 + $74,000 = $554,000, which at 48¢ on the dollar needs $1,154,167 of sales — about 10% above the honest build, and about 24% below the number already sitting in the plan.

So here is the decision, and every one of these options is taken by real operators every year:

1. Present $1,517,568. It is what the plan says and it will fund. It is also a number the owner now knows the room cannot produce, which makes it a misrepresentation to a lender and a personal guarantee taken against a fiction.
2. Present $1,154,167 — the number that exactly services the debt — and construct a bridge for it. Defensible only if the bridge is real. Note how seductive this option is: it feels honest because it is far below the original number, while being reverse-engineered from a loan payment, which is the exact failure the whole chapter warns about.
3. Present $1,048,800 and change the ask. Reduce the project cost, shrink the build-out, buy used equipment, negotiate more free rent, or raise more equity — reducing the $74,000 of annual debt service until the honest forecast covers it. This is the correct answer and it is the one that requires giving something up.
4. Present $1,048,800 and change the business. Add a lunch daypart, add a brunch, raise the check, or find a room with cheaper rent. Legitimate — but every one of those is a new forecast with new assumptions, and it belongs in the register rather than in the revenue line.
5. Don't do it. The rarest choice and sometimes the right one.

The register is what makes options 3 and 4 possible. Without it, the operator has a single revenue number and a single ask, and the only visible lever is the number. With it, they have sixteen rows, they can see that debt service is what broke the model, and they can go work on the thing that actually broke.

Outcome

Composites do not have outcomes, so here is the honest generalization instead.

Plans built on any of these three patterns do get funded — that is precisely why the patterns persist. What follows is well described by operators and lenders alike: the business opens, revenue comes in a quarter below plan, and the cost structure was built for the plan. Labor runs high because the schedule was written for covers that did not arrive. The working-capital reserve, sized against a forecast that assumed early cash flow, is consumed in the first two quarters. The operator discovers the problem somewhere between month eleven and month eighteen, which is Chapter 1's step three, and by then the options that remain are all expensive.

The business does not fail because the forecast was wrong. It fails because nobody found out the forecast was wrong until the money to respond was gone.

Lesson

Build the number you can defend, then argue about the distance to the number you want — in public, on the page.

Every one of the three patterns is a way of avoiding that argument. The market-share forecast avoids it by never stating the assumption. The annualized spike avoids it by treating six weeks of data as though it were a year. The no-calendar forecast avoids it by never converting a turn into minutes.

Compare Bellwether. Its base case produces $1,410,760 and its plan says $1,550,000, and the plan prints both numbers, itemizes the $139,240 between them into four named claims, discloses $1,240 of rounding, puts the whole thing in a register as the lowest-confidence row in the document, and then shows in the sensitivity analysis that this single assumption carries about 27% of the plan's operating profit.

Bellwether's forecast may still be wrong. Forecasts are. But it is wrong in a way that is visible, measurable, and, most importantly, catchable in week eight rather than month sixteen — because somebody wrote down what they believed, in dollars, before the room ever opened.

Discussion questions

1. All three composites overstate by roughly the same amount through completely different mechanisms. Propose an explanation for the clustering. Is there a reason a plan's overstatement would tend toward a quarter rather than toward five percent or toward double?

2. Pattern two's real damage was the labor schedule, not the revenue miss. Trace exactly how a revenue assumption becomes a cost commitment, and identify the last moment at which the commitment could still have been unwound cheaply.

3. Of the five options in the contested decision, option 2 is described as "seductive because it feels honest." Explain the trap precisely. How would you tell option 2 from a legitimate bridge like Bellwether's?

4. Option 3 — reduce the ask — is called the correct answer. What does an operator actually have to give up to take it, and why do you think it is chosen so rarely?

5. Pattern one's plan never wrote down "1.79 turns." Design a single-page test you could apply to any restaurant forecast, in under ten minutes, that would surface a hidden assumption of that kind.

6. Suppose you are the front-of-house partner and your chef partner has just presented option 1. What do you say? Write the actual sentences. Then consider what it would take to have made that conversation easier — and where in the process it should have happened instead.