Chapter 34: Financial Controls: Theft Prevention, Cash Handling, Inventory Variance, and Closing the Leaks

"I didn't have a theft problem. I had a nobody-was-looking problem. Those look identical from the outside for about eleven months, and then one of them turns into the other." — constructed; an operator, on the year he stopped counting

Overview

A controls chapter in a restaurant book almost always opens the same way, and the way it opens is wrong. It opens with a number — the industry loses three to four percent of sales to employee theft — and a photograph of a bartender's hand on a bottle, and from that point on the reader is being trained to look at the people who work for them as a risk to be managed. It is bad ethics and, more usefully for our purposes, it is bad accounting. That number cannot be sourced. And the money it points at is mostly somewhere else.

Here is what I actually believe after a lot of years of counting walk-ins at eleven at night. Controls are not an accusation. They are the systems that let you know what happened. That is the whole definition. A control does not catch a thief; a control produces a record, and a record is what lets you tell the difference between a compressor that failed on Sunday, a cook who started plating seven ounces in March, a special that was never costed, an invoice that crept eleven cents a pound, and — occasionally, genuinely — a person taking something. Without records, all five of those look the same on your bank statement: less money than there should be.

And here is the part that matters most, which Chapter 11 already proved with arithmetic and which this chapter will not contradict. When Bellwether's ideal-versus-actual food cost came apart, the variance decomposed into six named causes before theft was even on the list. Prices that no longer matched current costs. Specials nobody costed. Menu-mix drift. Purchasing. Over-portioning. Waste. Theft was seventh. Chapter 13 ranked the same causes a second way — by how fast you can check them — and landed in the same place. That ordering is the spine of this chapter, and it is not a courtesy. It is what the numbers keep saying.

The absence of controls is the actual danger, and the danger is not that it lets a thief operate. It is subtler than that. The absence of a record is what turns a small dishonesty into a habit — a free round on Thursday that nobody ever mentioned becomes a free round every Thursday, and eighteen months later a decent person is doing something they would have been horrified by at the start, because nothing ever pushed back. That is on the operator. A restaurant where nothing is counted is not a restaurant that trusts its people. It is a restaurant that has left its people alone with a temptation and no structure, which is a different and less admirable thing.

So this chapter builds the structure. Cash first, because it is the oldest and, honestly, no longer the biggest. Then the point-of-sale audit trail, which is where nearly all of the money now lives. Then the daily sales report, the one page that makes a night provable. Then inventory variance — thresholds, investigation, and what a pattern actually looks like as opposed to a bad week. Then the theft patterns themselves, each one described by the report it shows up on rather than the trick that produces it. Then receiving. Then the surprise count and the culture question it raises. And finally, soberly, how to respond when you think something is genuinely wrong — because a wrongly accused employee is a serious harm, and an investigation run badly creates real legal exposure and destroys the room.

In this chapter, you will learn to:

• Define internal control and separation of duties, and build a defensible control environment in a building with 31 people, four salaried managers, and no possibility of textbook segregation.
• Write cash-handling procedures — bank, drop, count, tolerance, deposit — and explain why cash is a shrinking but not vanishing share of the risk.
• Read every exception report a modern POS produces, and state precisely what each one proves.
• Build a daily sales report that ties in two directions and reconciles to a bank account.
• Set variance thresholds that catch drift without generating noise, and tell a pattern from a period.
• Apply the seven-rung investigation order, with theft last, and defend that order on grounds of both fairness and accuracy.
• Respond to a suspected loss in a way that protects the business, the record, and the person.

Learning Paths

🏗️ Opening — write §34.1 through §34.5 into your operating procedures before you open, not after. The habits set in the first ninety days (Chapter 9) are the habits you keep; a restaurant that opens without a daily sales report will still not have one in year three. 📋 Managing — this is your chapter. §34.3 and §34.4 are Monday-morning work, and §34.9 is the one you hope never to need and must read anyway. If you manage someone else's building, §34.8's culture argument is how you keep controls from costing you your best people. 🍸 Beverage — §34.5's thresholds are tighter for you, because liquor counts to the tenth and Chapter 15 already showed the leaks are larger per dollar. §34.6's sweethearting entry is yours; Chapter 15 named the leak, and this chapter names the behavior and the control that closes it. 🚚 Small Format — a truck or a pop-up has worse separation of duties than Bellwether, not better: often one person authorizes, records, holds, and reconciles. §34.1's practical substitutes and §34.8's blind count are the only tools you have. Use them harder.

34.1 Internal control in a small business: separation of duties when you have four managers

An internal control is any procedure that makes a business's own records reliable enough to act on. That is the definition, and notice what it does not say. It does not say "prevents theft." A control's product is trustworthy information. Theft prevention is a side effect — a real and valuable one, but downstream of the actual job.

Every control in this chapter does one of four things. It authorizes (somebody with standing approves the transaction). It records (the transaction lands in a system). It takes custody (somebody physically holds the money or the product). Or it reconciles (somebody compares two independent records and confirms they agree).

Separation of duties is the principle that no single person should hold two of those four functions for the same asset. The reason is not suspicion; it is that a person who both records a transaction and holds the asset has no external check on their own error, and error is far more common than dishonesty. If the same person counts the walk-in and orders the food, an ordering mistake and a counting mistake cancel each other out on paper and nobody ever learns anything.

In a corporation with a controller, a purchasing department, and an internal audit function, this is straightforward. In a 68-seat restaurant with four salaried people, one of whom owns the place, it is impossible, and I want to say that plainly rather than pretend otherwise. Most restaurant management books present the segregation-of-duties matrix as though it were achievable and then move on. It is not achievable at this scale. What is achievable is a set of practical substitutes, and knowing the difference is the actual skill.

FIGURE 34.1 — The four control functions, and why four managers cannot hold them
                                                          [the Bellwether plan]

  AUTHORIZE    who may approve it?         a comp, a purchase, a pay rate, a void
  RECORD       who enters it?              the POS, the invoice, the timeclock
  CUSTODY      who physically holds it?    the cash, the walk-in, the liquor room
  RECONCILE    who checks the two match?   the DSR, the bank, the count sheet

  THE TEXTBOOK RULE:  no one person holds two of these for the same asset.

  BELLWETHER, HONESTLY — 31 people, 4 salaried:

                            AUTHORIZE  RECORD   CUSTODY  RECONCILE
    chef-owner                 ●●●       ○        ●●●        ●
    front-of-house partner     ●●●       ●        ●●         ●●
    sous chef                  ●         ●        ●●●        ○
    service manager            ●●        ●●       ●●         ○
    outside bookkeeper         ○         ●●       ○          ●●●

    ●●● primary   ●● substantial   ● some   ○ none

  Read the rows, not the columns. The chef-owner authorizes nearly everything AND
  has primary custody of the most valuable inventory in the building. That is two
  functions on one person, and there is no fifth manager to hand one to. The sous
  chef receives the delivery and counts the walk-in — also two.

  The textbook answer is unavailable. The practical answer is the bookkeeper's
  RECONCILE column and the blind surprise count: the two places where a genuinely
  independent second look costs almost nothing.

The seven practical substitutes

When you cannot separate the functions, you compensate. These are the seven that actually work in a small restaurant, roughly in order of value per dollar.

1. An outside reconciler. The single most valuable control available to an independent operator, and most of them already pay for it and don't use it as a control. Your bookkeeper never touches cash, never signs an invoice, never approves a comp, and never writes a schedule. That makes them the only genuinely independent party in your financial life. Give them the daily sales reports, the deposit slips, the bank feed, and the invoices, and ask them to tell you what does not tie. Not to produce a statement — to produce a list of exceptions. It costs you a conversation.

2. Cross-review between the two people at the top. The chef-owner and the front-of-house partner each read the other's exception reports. Neither one can fix their own numbers unobserved. This is not distrust between partners; it is the reason partnerships in this business survive. The partnership that never looks at each other's numbers is the partnership that ends badly, and almost never because anyone stole anything — usually because one of them was quietly wrong for a year and nobody noticed in time to fix it.

3. Rotation. Whoever ordered a category this period does not count it next period. Whoever counted the bar last period counts the walk-in this one. Rotation breaks the accumulation of undetected error, and it has the side benefit of teaching four people the whole building.

4. Blind records. A count sheet with the expected quantity pre-printed on it is not a count sheet; it is a suggestion. Give the counter the item list and nothing else. §34.8 develops this.

5. Owner review of exception reports. The owner cannot be inside every process. The owner can read every exception summary, every week, forever. And here is the thing about that: most of the control value is not in the reading. It is in the fact that everyone knows the reading happens. A report that is produced and never read is worse than no report, because it manufactures a false sense of coverage.

6. Mandatory absence. Some controls only work when the usual person is not there. If one manager has closed every Friday for two years, nobody knows what a Friday close looks like without them — including whether it looks the same. Rotate the close. Make people take their days off. This one also happens to be a retention control, which we will come back to.

7. Systems that log. A POS audit trail, a timeclock with individual credentials, a key log on the liquor room, a camera at the receiving door. These substitute a machine for a second human. They are the cheapest separation you can buy, and they are the reason a modern small restaurant can run a credible control environment at all.

⚖️ Code and Compliance

The schedule and the payroll approval must not sit on the same desk — and this is a wage-and-hour problem before it is a theft problem.

Chapter 20 established that restaurants generate more wage-and-hour liability per dollar of revenue than almost any other industry, and that nearly all of it is unintentional. Here is where that collides with internal control.

If the manager who supervises a server can also edit that server's punches, you have created two exposures at once with a single permission setting:

• The control exposure. Time records are the input to your second-largest cost line. A person who can both create and edit that input has no check on their own error.
• The compliance exposure, which is larger. A supervisor with punch-edit authority is exactly the mechanism by which off-the-clock work happens — a shift trimmed to avoid an overtime hour, a pre-shift meeting edited out, a closing crew clocked out and still cleaning. Under the Fair Labor Standards Act, unpaid work time is generally recoverable as back wages, and the remedy scheme commonly permits liquidated damages in an amount equal to the back wages, with a lookback of two years that extends to three for willful violations — plus attorney's fees. State law is frequently more generous to the employee than federal law.

The control that addresses both: every punch edit requires a reason code, is visible to the employee, and is reviewed by someone who does not supervise that employee. At Bellwether that reviewer is the outside bookkeeper, who flags any edit that reduces paid time. It costs nothing and it is the single highest-value permission decision you will make in your POS.

One more, and be honest about it: buddy punching by employees is real and it is small. Fifteen minutes a week for six people at a \$16 blended wage is 78 hours a year — about **\$1,248. A supervisor shaving ten minutes a shift off eight people across five shifts a week is roughly 347 unpaid hours — about \$5,547** in wages, which can double under liquidated damages, before fees. Point your controls at both. Notice which one your instinct went to first.

Employment law varies enormously by state, county, and city, and it changes. Verify locally, and use an employment attorney for anything consequential.

What a control environment is for, restated

Prime cost is the number that keeps you open, and Bellwether's plan says 60.0% — blended cost of goods sold at 27.8% and labor at 32.3% on \$1,550,000 of sales, producing \$261,020 of operating profit before \$69,500 of debt service. Chapters 11 through 33 built that number. This chapter is what keeps it there. A plan can produce a 60% prime cost; only a control environment can produce it in month twenty-nine, when the person who wrote the cost cards is busy, the sous who counted the walk-in has left, and three specials have run uncosted since April.

Controls are the maintenance system for every number in the rest of the book. That is the least glamorous sentence in this chapter and the truest one.

34.2 Cash handling: banks, drops, counts, and the two-person rule

Cash is where every restaurant's control conversation starts, and it should be where it ends.

Let me define the vocabulary first, because it is genuinely specific to this business and everyone uses it as though you already know it.

A bank is the fixed amount of currency and coin a drawer starts a shift with, so that change can be made — at Bellwether, \$250 per drawer, two drawers, one at the bar and one at the server station. The bank is counted in and counted out and never varies; a bank that drifts upward is somebody's convenience becoming somebody's opportunity.

A cash drop is the mid-shift removal of currency from the drawer into a safe, leaving the bank and enough working change. Bellwether's rule is a drop whenever the drawer holds \$500 over its bank, and a mandatory drop at the end of brunch on Saturday. The point of a drop is not really theft; it is that a drawer holding \$900 on a Friday night is an armed-robbery target and an uncounted liability at the same time.

The count is the end-of-shift reconciliation of what is physically in the drawer against what the POS says should be there. Over/short is the difference. And the two-person rule is exactly what it sounds like: the count happens with two people present, off the floor, in one place, and both of them initial the sheet.

Cash handling procedures, taken together, are the written sequence covering all of it: who issues the bank, when drops occur, who counts, what tolerance is acceptable, who prepares the deposit, who transports it, and who reconciles it against the bank statement.

FIGURE 34.2 — The path of a dollar, and the control at each handoff
                                                        [the Bellwether plan]

   GUEST
     │
     ├── orders ──────────────► SERVER / BARTENDER
     │                            CONTROL: ring it before you pour it or plate it
     │                                     │
     │                                     ▼
     │                              POS TICKET on a numbered check
     │                            CONTROL: no product leaves without a ticket
     │                                     │
     ├── pays ────────────────►  TENDER
     │                            CONTROL: itemized receipt, every time, no exceptions
     │                                     │
     │                    ┌────────────────┴────────────────┐
     │                    ▼                                 ▼
     │              CARDS  ~91%                         CASH  ~9%
     │                    │                                 │
     │                    ▼                                 ▼
     │            PROCESSOR BATCH                       DRAWER
     │            CONTROL: batch closed                 CONTROL: drop at $500
     │            by the POS, not by hand                        over bank
     │                    │                                 │
     │                    ▼                                 ▼
     │            SETTLEMENT, 1–2 days                  TWO-PERSON COUNT
     │            CONTROL: matched to the               CONTROL: over/short logged
     │            batch report, not to sales                    and initialed x2
     │                    │                                 │
     │                    │                                 ▼
     │                    │                             SEALED DEPOSIT
     │                    │                             CONTROL: logged, bag number
     │                    │                             recorded, rotating pair
     │                    │                                 │
     │                    └──────────────┬──────────────────┘
     │                                   ▼
     │                          THE BANK ACCOUNT
     │                                   │
     │                                   ▼
     │                   DAILY SALES REPORT — ties POS to bank, nightly
     │                                   │
     │                                   ▼
     │                   OUTSIDE BOOKKEEPER — reconciles weekly;
     └──────────────────  never touched the money at any point above

   The control is never the safe. The control is the HANDOFF. At every point where
   money or product changes hands, one document must be created that a DIFFERENT
   person will read.

The sequence, written out

A cash-handling procedure that lives in a binder nobody opens is not a control. Here is the version that fits on one laminated card by the office door.

Two of those lines are the ones operators skip, and they are the two that matter most.

The first is recording overs. An over feels like good news, so people pocket the difference into the next drawer or leave it in the tip jar or simply do not write it down. But an over is a control failure with exactly the same information content as a short — it means the drawer and the POS disagree, which means somebody mis-tendered, mis-keyed, or gave the wrong change to a guest. If you only record shorts, you have taught your staff that the count is a punishment rather than a measurement, and you have destroyed half your data. A drawer that is \$8 over on Tuesday and \$8 short on Wednesday is a training problem you can fix. A drawer that is only ever recorded when it is short is a mystery you will never solve.

The second is the outside reconciliation. The person who prepared the deposit must not be the person who confirms it landed. That is the one true separation of duties a small restaurant can actually achieve, and it costs a weekly email.

👨‍🍳 On the Line

The closing manager's last twenty minutes, honestly.

It is 11:35 on a Friday. The last two-top is finishing coffee. The dish machine is running its final rack, the sous is breaking down the line, and the closing manager has to do three things that all want to be done first.

Here is the order that works, and the reason for each:

1. Close the POS batch (2 minutes). Do this before you count anything. The batch closes card transactions for the day and freezes the numbers you are about to reconcile against. If you count the drawer first and a server closes one more check while you are counting, you have created a discrepancy out of nothing and you will spend fifteen minutes chasing it.

2. Count with a second person, off the floor, door closed (8 minutes). Not at the host stand. Not with servers walking through. Two people, one sheet, both initials. The second person is usually the closing server or the bartender, and here is the thing nobody tells you: they want this. A server who carried a drawer and left without a witnessed count has no protection if the number is wrong tomorrow. The two-person rule protects the staff at least as much as the owner. Say that out loud when you introduce it and it stops feeling like suspicion within a week.

3. Fill in the daily sales report while the building is still open (8 minutes). This is the one people defer to the morning, and deferring it destroys most of its value. The DSR's job is to make the night provable while the people who handled the money are still there. If the drawer is \$40 short and you find out at 11:45, you can ask three people a question. If you find out at 10 the next morning, you have a mystery and a memory, and the honest answer is you will write it off.

What goes wrong. The manager who is also expediting, the manager who has to drive a cook home, the Friday when the POS batch fails and the report has to be pulled by hand. On those nights the count gets done and the DSR gets deferred, and that is a survivable compromise as long as it is the exception and somebody notices. Which is why the DSR has a second signature on it the next morning. The second signature is not about the numbers. It is about whether the first signature happened.

Tolerance, and what to do about a short

Set a tolerance and publish it. Bellwether's is ±\$5.00 or 0.5% of cash sales, whichever is greater. On a Friday with \$612 of cash tenders, 0.5% is \$3.06, so the \$5 floor governs; on a Saturday with \$1,100 of cash, the tolerance is \$5.50.

A single night outside tolerance gets noted and nothing else. Three nights outside tolerance in a four-week period, or any single night over \$25, triggers a review — and the review looks at the process first: Was the bank too small, forcing servers to make change out of tips? Was the drawer shared between two people with no attribution? Did the POS have a modifier that mis-prices a common item? Was there a new hire on the drawer? In my experience four out of five recurring shorts resolve into one of those four answers, and the fifth resolves into a training gap. That is not naivety; it is base rates. Investigate the system, then the person, in that order — the same ordering principle that runs the whole chapter.

⚠️ Where the Money Leaks

Cash is now the smaller risk and the bigger ritual.

This is the most expensive misallocation of attention in the industry, and almost every operator makes it.

At Bellwether, cash runs roughly 9% of sales — about \$139,500 a year through the drawer. Assume you lose 1% of it to shorts, mis-tenders, and every other cash cause combined. That is \$1,395 a year.

Now put that next to the other numbers this book has already computed:

Where the money actually goes	Per year
Measured food variance, 2.0 points of food sales (§34.5)	\$22,282
The five bar leaks from Chapter 15	\$16,169
Comps over the 1.0% target (§34.3)	\$8,680
Receiving error at 1% of the order guide (§34.7)	\$3,348
Cash, at a 1% loss rate	\$1,395

Cash is 2.6% of that list. And it routinely gets more than half the control attention in the building, because it is physical, because it is the part managers were trained on, and because arguing about \$1.75 in a locked office at midnight feels like control in a way that reading a void report on Monday does not.

Count the drawer. Do it every night, two people, no exceptions — the ritual has real value for staff protection and for personal safety, and a business that cannot count \$612 cannot be trusted to count \$25,488 of walk-in. But understand its scale. Then go read the reports where the money is.

There is one thing cash controls do that nothing else does, and it has nothing to do with money: they keep people safe. A nightly deposit taken by the same person at the same time out the same door is a pattern. Vary the pair, vary the timing, use the bank's night drop, and never send someone alone at 1 a.m. with a bag. I have watched an operator lose a manager over this — not to theft, to a mugging in a parking lot — and no amount of over/short discipline compensates for it.

34.3 The POS as an audit trail: voids, comps, discounts, and reopened checks

Chapter 26 established the point-of-sale system as the system of record — the place where every transaction in the building is captured, and the source of every number you manage by. This section is about its second job, which most operators never use: the POS is also the audit trail.

A POS audit trail is the system's own log of every action that changed a transaction, and who authorized it. Not just what was sold — what was altered. Every void, comp, discount, price override, item deletion, reopened check, tip adjustment, and drawer opening, with a user ID and a timestamp attached. It exists in every modern cloud POS. It is on by default. And in most independent restaurants nobody has ever opened it.

An exception report is any report drawn from that trail that shows you the abnormal rather than the normal. Sales reports show you the business. Exception reports show you the edges, and the edges are where both your errors and your losses live.

The five exceptions that matter, and precisely what each one proves

Define these carefully, because operators mix them up constantly and the distinctions have real consequences on the P&L.

A void removes an item from a check before the check is closed. It is a correction of a mis-ring: the server hit "hearth chicken" and meant "hearth trout," or the kitchen never made it. A void does not reduce sales, because the sale never happened. Chapter 31 made this point in the flash report and it is worth repeating: voids are invisible on your P&L. They are purely an audit item — a signal about training, menu design, and printer routing, and, at the margins, about product that left the kitchen and then left the record.

A comp removes an item from a check after it was made, and the restaurant absorbs the cost. The guest received the food; nobody paid for it. A comp reduces sales. It appears on your P&L as revenue you did not earn, and the product cost stays in your COGS — which is why an uncontrolled comp line simultaneously depresses sales and inflates food cost percentage. It moves both halves of the ratio in the wrong direction at once.

A promotional discount reduces the price of an item the guest still pays for — a neighborhood card, a Restaurant Week price, an industry-night twenty percent. A discount reduces sales, like a comp, but it is a marketing expense with an expected return, where a comp is usually a service recovery expense or a relationship expense.

A reopened check is a check that was closed and paid and then reopened. Legitimate reasons are common: a tip written illegibly, a card run for the wrong amount, a split done wrong, a guest who came back for one more drink. It is also the only place in the POS where a completed, tendered transaction can be changed after the fact, which is why it is simultaneously the highest-value and least-read report on this list.

A no-sale is a drawer opening with no transaction attached. Legitimate: making change for a guest or for another drawer, correcting a mis-tender. What matters is never the individual event but the rate, the user, and the hour.

Comp / void / discount authorization is the policy layer over all of it: who may do each of these, up to what dollar limit, with what reason code, and who reviews it. That policy is the actual control. The reports are only the evidence that the policy is being followed.

What Bellwether's week looked like

Chapter 31's weekly flash report carries the figures. Comps of \$486, which is 1.56% of gross sales against a target of ≤1.0%. Promotional discounts of \$310, or 0.99% — on target. And voids of \$128 across 41 items, which do not reduce sales and are an audit item.

Start with the comps, because there is real money in that gap.

🧮 Run the Numbers

What 0.56 points of comps costs.

Bellwether's comps ran \$486** on gross sales of **\$31,196 in the sample week — 1.56% against a 1.0% target. The gap is 0.56 points.

In the week: 1.0% of \$31,196 is \$312. Actual was \$486. **The week's excess is \$174.**

At plan volume of \$1,550,000, 0.56 points of sales is:

\$1,550,000 × 0.0056 = **\$8,680 a year.**

Put that against the plan's operating profit of \$261,020: the comp overage alone is 3.3% of the entire year's profit, given away sixteen dollars at a time by people who were, in almost every instance, trying to do the right thing for a guest.

Now the part that changes what you do about it. Comps do not distribute evenly across the week:

Service	Gross sales	Comps	Comp % of gross
Sunday brunch	\$2,590 | \$24	0.93%
Tuesday dinner	\$3,282 | \$18	0.55%
Wednesday dinner	\$3,744 | \$46	1.23%
Thursday dinner	\$4,393 | \$61	1.39%
Friday dinner	\$6,807** | **\$187	2.75%
Saturday brunch	\$2,940 | \$22	0.75%
Saturday dinner	\$7,440 | \$128	1.72%
Week	\$31,196** | **\$486	1.56%

Five of seven services are at or near target. The entire overage is Friday and Saturday dinner — \$315 of the \$486, on the two nights when the kitchen is at capacity and the managers are least able to stop and think.

The weekly number told you that you have a comp problem. The daily breakdown told you that you have a Friday-and-Saturday kitchen-capacity problem being paid for out of the comp line. Those require completely different responses, and only one of them is a control response.

That last observation deserves its own paragraph, because it is the single most common misdiagnosis in this part of the business. Pull Friday's comps apart by reason code:

\$76 of Friday's \$187 — 41% — bought back tickets that ran long. That is not a controls failure. That is a kitchen throughput problem, and the fix is in Chapter 14's ticket-time work and Chapter 19's staffing guide, not in this chapter. A comp policy tightened in response to that number would make the guest experience worse without recovering a dollar, because the underlying tickets would still run long — you would simply have stopped paying to fix them.

⚠️ Where the Money Leaks

The \$16 line is the only one that is actually a control problem.

One comp on Friday closed with no reason code entered. Sixteen dollars. Nobody should lose sleep over sixteen dollars.

Lose sleep over the percentage. \$16 of \$187 is 8.6% of Friday's comp dollars that are permanently un-reviewable. Not disputed — un-reviewable. There is no version of this you can go back and analyze, because the field is blank and the memory is gone.

Here is what I have watched happen, more than once. Un-coded comps start at 5%. Nobody says anything, because it is five percent and it is sixteen dollars. Six months later they are 30%, because the staff correctly inferred that the field is optional. At 30% of a \$25,000 annual comp line, \$7,500 a year of giveaway has no explanation attached to it — and the operator, staring at a rising comp number with no reasons behind it, will eventually start suspecting people. Which is precisely the outcome this chapter exists to prevent.

The rate at which reason codes go un-entered is the single best measure of whether your comp policy is real. Measure it weekly. Report it as a percentage, not as dollars. Configure the POS to refuse a comp without a reason code if it can, and if it cannot, review the blanks by name every Monday and ask the question kindly the first three times.

Voids: what a distribution looks like, and what it does not mean

Forty-one voids for \$128 across a 767-cover week is a low, healthy number — an average void of \$3.12, which tells you these are mostly modifiers and mis-rings, not entrées. The total is unremarkable. The distribution is the report.

FIGURE 34.3 — Void distribution, one week: 41 voids / $128
                                                     [the Bellwether plan]

    Server 1     ███ 3
    Server 2     ██ 2
    Server 3     ████ 4
    Server 4     ██████████████████ 18     <-- the only line worth a conversation
    Server 5     ███ 3
    Server 6     ██ 2
    Bartender 1  █████ 5
    Bartender 2  ████ 4
    ──────────────────────────────────────
    TOTAL        41

  Eighteen voids against a house average of about five is a real outlier and it
  should absolutely be looked at. Here is the discipline: it is a QUESTION, and
  it has four ordinary answers before it has an extraordinary one.

    1. New hire.       Two weeks in, still learning the modifier tree.
    2. Menu design.    One dish with six required modifiers that everybody
                       mis-rings — and this person sells the most of it.
    3. Routing.        Their station's printer sends to the wrong expo, so they
                       void and re-ring to correct it.
    4. Instruction.    A manager told them to void and re-ring for something,
                       and never told anyone else.

  In my experience one of those four is the answer better than four times in
  five. You find out by ASKING, in a normal voice, on the floor, in a way that
  assumes the ordinary answer — because it usually is the ordinary answer, and
  because the cost of implying otherwise to a good employee is enormous.

The reason voids matter at all, past training, is narrow and worth stating precisely: a void removes an item from the record but does not un-cook the food. If product physically left the kitchen or the bar and the item representing it was voided, the sale is gone from your records and the product is gone from your walk-in. That mismatch does not appear on the void report. It appears in inventory variance — which is §34.5, and which is why this book keeps insisting that the count is the control and the POS is only the evidence.

Reopened checks and no-sales: the two reports nobody reads

Bellwether's week: six reopened checks, all by the two salaried FOH managers, all with reason codes, four of them tip corrections. And 22 no-sale drawer opens, fourteen of them during the two brunch services — exactly where you would expect them, since brunch is the cash-heavy daypart and change gets made.

Both of those are clean. But here is what you are actually looking for, and it is not the total:

Notice that in every row of that table, the signal is a shape, not an event. One reopened check proves nothing. A reopened check that changed a tender type from card to cash on a Tuesday at 1 a.m. is not proof of anything either — but it is a question that deserves an answer, and if the answer is good you have lost four minutes.

That is the correct posture toward every exception report in this section. The report generates questions. Only people answer them, and most of the answers are boring.

🤝 Hospitality

Do not let a control policy kill the save.

Chapter 23 worked the arithmetic of a comp and reached a conclusion this chapter has to respect: a guest whose problem is fixed generously and immediately, by the person standing at the table, is more likely to return than a guest who never had a problem at all. The comp is the cheapest retention instrument in the building.

Now imagine a control policy that requires a manager's code for every comp. It is Friday at 8:40. A table waited nineteen minutes for a dessert. The server knows exactly what to do — take the dessert off — and cannot, because the manager is in the walk-in. The table waits eleven more minutes for a manager to appear and apologize for something the server already apologized for. The nine-dollar dessert now costs you the visit, and the server has learned that the restaurant's systems prevent them from doing their job.

Tiered authorization solves it, and it is the right design in every restaurant I have seen it in:

Who	Limit per check	Requirement
Server / bartender	up to \$15	mandatory reason code; no manager needed
Shift manager	up to \$75	reason code plus their own manager ID
Chef-owner / FOH partner	above \$75	reason code plus a note on the DSR

Bellwether's week had roughly 34 comped items averaging \$14.29 — most comps fall under the server limit, which is exactly right. The person at the table fixes the table. The control is not the permission gate; the control is that every one of those 34 items carries a reason code and gets read on Monday, and that a server whose comp rate is triple the house average gets a conversation that starts with "walk me through your Fridays."

A control that makes hospitality slower has to justify itself against the revenue it costs. Most of them cannot.

34.4 The daily sales report: reconciling sales, cash, cards, and deposits

The daily sales report (DSR) is one page, produced every night the restaurant is open, that accounts for every dollar the building took in — in two independent directions. Down one side, what was sold. Down the other, how it was paid. The two must agree. When they agree, the night is provable. When they do not, you stop and find out why while the people who handled the money are still there.

If I could install exactly one control in a restaurant that had none, it would be this. Not the cameras, not the inventory software, not the safe. This.

Chapter 31's flash report carries the week's version as a single line: expected deposits plus card settlement of \$32,528. The DSR is where that line is built, one night at a time.

🧾 Read the Numbers

```text FIGURE 34.4 — "The daily sales report" [the Bellwether plan] THE ARTIFACT One page, closed by the closing manager at 11:40 p.m. and reviewed and initialed by a second manager before 10 a.m. the next morning. Delivered to the outside bookkeeper in a weekly packet of seven. THE CONTEXT A Friday in June. 143 covers, patio open (84 seats, 1.70 turns), average check $46.01 — dead on the plan's $46 dinner assumption. Two drawers at a $250 bank each. Three servers, one bartender, one host, four on the line at peak.

               SALES                                     CONTROL NOTE
                 Gross sales                  6,807.00
                 less comps                    (187.00)   12 items, reason codes attached
                 less promotional discounts     (40.00)   neighborhood card, 2 checks
                 ────────────────────────────────────
                 NET SALES                    6,580.00    the only revenue line here
                 Sales tax @ 7%                 460.60    LIABILITY — not your money
                 Gift certificates sold          75.00    LIABILITY — not revenue
                 Charged tips                 1,163.40    OWED TO STAFF — not revenue
                 ────────────────────────────────────
                 TOTAL TO ACCOUNT FOR         8,279.00

               TENDERS
                 Cash                           612.00
                 Cards (118 transactions)     7,442.00    includes the charged tips
                 Gift certificates redeemed     150.00
                 House account, billed monthly    75.00
                 ────────────────────────────────────
                 TOTAL TENDERED               8,279.00    TIES. If it doesn't, stop here.

               CASH RECONCILIATION
                 Opening banks (2 x $250)       500.00
                 + cash tenders per POS         612.00
                 - cash tips to the pool         88.00
                 - petty-cash paid-outs          52.00    produce run, receipt attached
                 ────────────────────────────────────
                 EXPECTED IN DRAWER             972.00
                 COUNTED, two people            970.25
                 OVER / (SHORT)                  (1.75)   0.29% of cash — in tolerance
                 DEPOSIT SEALED, bag #4471      470.25    (drawer less the $500 re-bank)

               CARD SETTLEMENT EXPECTED
                 Gross card tenders           7,442.00
                 less processing, 2.6% + $0.10 x 118
                                               (205.29)   2.76% effective
                 ────────────────────────────────────
                 EXPECTED IN BANK, 1-2 DAYS   7,236.71

               EXCEPTIONS TONIGHT
                 Voids                        7 items / $22.00
                 Comps                       12 items / $187.00   (2.75% of gross)
                 Comps with NO reason code    1 item  / $16.00     <-- the line that matters
                 Reopened checks              1 / tip correction / mgr ID / initialed
                 No-sale drawer opens         3 / all before 6 p.m.

WHAT IT SHOWS The night ties. Every dollar that crossed the counter is accounted for in two independent directions, and both columns land on $8,279.00. It also shows something an operator has to internalize: of that $8,279, only $6,580 is Bellwether's revenue. $460.60 belongs to the state, $75.00 belongs to whoever holds that certificate, and $1,163.40 belongs to the staff. Cash was 8.7% of sales-plus-tax. The over/short is $1.75 on $612 — noise, correctly logged anyway. WHAT IT DOESN'T It does not prove nothing was stolen, and it cannot. A ticket that was never rung does not appear on a report assembled from tickets; that leak surfaces in inventory, not here. It does not show whether the 12 comps were justified — only that 11 of them were coded. It does not show ticket times, which are the actual cause of $76 of tonight's comps. And the settlement line is a PREDICTION: until it is matched against the bank in two days, $7,236.71 is a promise, not a deposit. THE DECISION Sign it. Initial the exception block. Then two things before service tomorrow: (1) ask the closing manager — not accusingly — why one comp closed without a reason code, and re-train if the answer is "I forgot," because 8.6% of tonight's comp dollars are now permanently un-reviewable and that share only grows; (2) put Friday's 2.75% comp rate on Monday's flash report NEXT TO the ticket-time report, because those two numbers are the same number. THE LESSON The DSR's job is not to catch a thief. Its job is to make one night's money provable while the people who handled it are still in the building. One number you can prove tonight is worth ten you can argue about in April. ```

The three reconciliations hiding inside one page

Read Figure 34.4 again and notice it is really three separate ties stacked on one sheet. Operators who see this stop making the classic errors.

Tie one: sold against tendered. \$8,279.00 = \$8,279.00. This is the identity that makes the page a control. It catches mis-tenders, mis-keys, a card run twice, a check closed to the wrong tender type, and a drawer that was short-changed. It cannot catch a sale that was never entered — nothing built from the POS can.

Tie two: cash expected against cash counted. \$972.00 expected, \$970.25 counted. This is the only place in the restaurant where a physical asset is compared to a record on the same night.

Tie three: card tenders against what the bank receives. This one is not finished tonight. It completes in one to two days, and it must be matched against the processor's batch report, not against sales. Which brings us to the error I see most often.

⚠️ Where the Money Leaks

Comparing your bank account to your sales, and concluding something is wrong.

An operator looks at the bank feed for the week. Money in: something over \$38,000. Then they look at the flash report: net sales of \$30,400. And they think, good — more came in than I sold.

Or the other version: they look at one day's expected settlement of \$7,236.71 against sales of \$6,580, cannot make the numbers behave, and stop trying.

Neither number is wrong. They are different quantities, and the DSR is the bridge:

Layer	The week
Gross sales	\$31,196
less comps and promotional discounts	(\$796)
Net sales — the only revenue	\$30,400
plus sales tax collected at 7%	\$2,128
= the flash report's expected deposits + settlement	\$32,528
plus charged tips (owed to staff)	rides on top
plus gift certificates sold (a liability)	rides on top
less card processing at roughly 2.76% of card volume	comes off
= what actually appears in the bank	none of the above

The money crossing your bank account is never equal to your sales, and it never will be. It is larger by the tips and the gift-certificate proceeds — neither of which is yours — and smaller by the processing fees. Chapter 31's \$32,528 is the sales-and-tax figure your deposits and settlements have to account for, not the balance you will see.

Chapter 33 made this point about cash and profit diverging. Here is the control version of the same theme: an operator who cannot separate revenue from receipts will eventually make a disastrous cash decision — most commonly spending the sales tax, which was collected daily, looked like income for four weeks, and was never theirs. \$2,128 a week is on the order of **\$110,000 a year** sitting in your account belonging to the state.

Seven DSRs make a flash report

The last thing the DSR does is roll up. Bellwether's operating week is six days — dark Monday, dinner Tuesday through Saturday, brunch Saturday and Sunday — which is seven services and seven DSR sheets.

Every figure on the bottom row is a number Chapter 31 already published. That is the point of the exercise: the flash report is not a separate document. It is seven DSRs added up. If your flash report is assembled from the POS's weekly summary rather than from the nightly sheets, you have a report and not a control — because nothing in it was ever compared to a physical count of anything.

\$30,400 of net sales is a normal-to-good week; the plan's weekly average is about \$29,800. Nothing here is a crisis. The whole apparatus exists so that the week you do have a crisis, you find out on Tuesday.

🔍 Check Your Understanding

1. A server closes a check, reopens it and lowers the tip by \$40, then re-closes it. Which of the three ties on the DSR catches this, and which does not?
2. Your DSR shows \$612.00 of cash tenders and \$972.00 expected in drawer, but the drawer counted \$1,022.00. Name three ordinary explanations before you name an extraordinary one.
3. Why is a void invisible on the P&L, and a comp not?

(1: Tie one — sold against tendered — will not catch it, because lowering the tip lowers both the charged-tip line and the card tender by the same \$40 and the page still balances. Tie three catches it only if you match the processor batch or the signed slips to the adjusted amounts, which is exactly why tip adjustments belong on the reopened-check exception report. 2: A bank issued at \$300 instead of \$250 on one drawer; a \$50 drop left in the drawer rather than dropped; a paid-out slip written but the money never removed; a guest who overpaid a cash check and did not wait for change. Any of those beats a theory about a person. 3: A void removes an item that was never sold — there was no revenue to reduce. A comp removes revenue from a sale that did occur while the product cost stays in COGS, so it depresses sales and inflates food cost percentage simultaneously.)

34.5 Inventory variance: setting thresholds, investigating, and what a pattern looks like

Chapter 11 owns the arithmetic here and I am not going to re-derive it. The usage formula — beginning inventory plus purchases minus ending inventory — gives you what you actually consumed. The cost cards and the POS item mix give you what you should have consumed. The gap between them is the whole subject.

Inventory variance is that gap, expressed at the category level, in dollars and as a percentage of theoretical usage, over a defined period. Chapter 11 called the underlying comparison ideal-versus-actual food cost and decomposed one instance of it into named causes. This section is about the layer on top: how large a gap has to be before you act, how to tell a bad period from a trend, and what the report looks like on the desk.

That layer is where most operators fail, and they fail in one of two symmetrical ways. Either they set no threshold at all, in which case every variance is either ignored or panicked about depending on mood. Or they set one absurdly tight — "anything over a hundred dollars gets investigated" — and within two periods they have investigated eleven things, found nothing eight times, exhausted themselves, and quit. A threshold that generates more work than you will actually do is worse than no threshold, because it teaches you to ignore your own reports.

The report

🧾 Read the Numbers

```text FIGURE 34.5 — "Period 8 food variance report" [the Bellwether plan] THE ARTIFACT Four-week ideal-versus-actual food usage report by category, period 8 of 13. Built from the cost cards, the POS item mix, the purchase journal, and a BLIND physical count taken the last night of the period. THE CONTEXT Bellwether, period 8. Food sales $86,400. No menu change and no price change in the period. No new hires in the kitchen. One Wednesday pork special that ran three weeks and was never costed.

               CATEGORY                IDEAL      ACTUAL   VARIANCE    % OF   FLAG
                                       USAGE      USAGE                IDEAL
               Meat & poultry         $9,850     $10,912    +$1,062   +10.8%   ●●
               Seafood                 3,420       3,708       +288    +8.4%   ●
               Produce                 4,180       4,466       +286    +6.8%   ●
               Dairy & eggs            2,640       2,701        +61    +2.3%
               Dry goods & pantry      3,910       3,864        -46    -1.2%   ?
               Bakery & desserts       1,488       1,551        +63    +4.2%
               ──────────────────────────────────────────────────────────────
               TOTAL                 $25,488     $27,202    +$1,714    +6.7%
               As % of food sales      29.5%       31.5%    +2.0 pts

               ●● over threshold and rising    ● over threshold
                ? negative variance — recount before you do anything else

WHAT IT SHOWS Two full points of food cost, worth $22,282 a year at this volume, and it is not spread evenly: meat and poultry is 62% of the total variance on 39% of the ideal usage. Three categories are over threshold and three are not, which is the most useful thing on the page — it turns "my food cost is high" into a specific place to stand in the walk-in. The negative variance in dry goods is NOT a gift; a category that comes in under theoretical is almost always a count error, a mis-extended invoice, or product counted into the wrong category. WHAT IT DOESN'T It does not name a cause, and it cannot. It cannot distinguish a bad period from a trend — one period is a data point, not a pattern (Figure 34.6). It cannot see anything that happened to product that was never counted or never rung. And it is only as good as the count: a hand count of a 40-SKU walk-in carries perhaps half a point of honest error, which on $25,488 of ideal usage is about $127. Anything inside $127 is arithmetic, not information. THE DECISION Work the ladder (Figure 34.7) on MEAT AND POULTRY ONLY, starting with the cost cards and the uncosted special — both of which are paper, both of which can be settled before service tomorrow. Put produce on the waste log for two periods and do nothing else with it. Recount dry goods. Do not convene a meeting about seafood. THE LESSON A variance report tells you WHERE to look and never WHAT you will find. It is a flashlight, not a verdict. ```

Two points of food cost is \$22,282 a year — 1.4% of sales, and **8.5% of the plan's \$261,020 of operating profit.** It is exactly the kind of number Chapter 1 described as cost drift: individually trivial, collectively equal to a good chunk of the year.

The thresholds I would actually recommend

Here is the structure I use, and the reasoning behind each line matters more than the numbers.

The greater-of rule is the part worth understanding, because it is what keeps the system usable. Run Period 8 through it:

• Meat & poultry, ideal \$9,850: 5% is \$493, which beats the \$250 floor, so the test is \$493. Variance \$1,062. Over — by more than double.
• Seafood, ideal \$3,420: 5% is \$171, so the \$250 floor governs. Variance \$288. Over, narrowly.
• Produce, ideal \$4,180: 5% is \$209, so the floor governs. Variance \$286. Over, narrowly.
• Dairy & eggs: variance \$61 against a \$250 floor. Clear.
• Bakery & desserts, ideal \$1,488: 5% is only \$74 — and this is exactly why you need the dollar floor. Without it, a \$75 swing in a small category would trigger an investigation every single period, forever. Variance \$63. Clear.
• Dry goods: −1.2%, which trips the negative rule. Recount, do not investigate.
• Whole book: 2.0 points against a 1.0-point trigger. The book gets reviewed.

The percentage catches the big categories where 5% is real money. The dollar floor protects you from the small categories where a percentage is noise. Neither works alone.

What a pattern looks like

This is the distinction that separates operators who use variance reports from operators who own variance reports.

FIGURE 34.6 — What a pattern looks like, and what it doesn't
     Category variance as a percentage of that category's ideal usage, three periods
                                                          [the Bellwether plan]

  MEAT & POULTRY
    P6   ████ 4.3%          ($418)     under threshold
    P7   ███████ 7.1%       ($690)     over
    P8   ██████████ 10.8%   ($1,062)   over, and rising
    → One category. One direction. Monotone across three periods.
      THIS IS A PATTERN. Investigate it, and expect to find a cause on
      rungs 1 through 6.

  SEAFOOD
    P6   (███) -2.8%        (-$96)     under theoretical — count? credit?
    P7   ████████████ 11.7% ($402)     over
    P8   ████████ 8.4%      ($288)     over
    → Large. Noisy. Crosses zero. Seafood does this everywhere, because
      yield varies by the fish and spoilage is unforgiving.
      THIS IS VOLATILITY, NOT A PATTERN. Fix the spec and the count first,
      then look again in two periods.

  PRODUCE
    P6   ███████ 7.4%       ($310)     over
    P7   ████ 4.0%          ($168)     under
    P8   ███████ 6.8%       ($286)     over
    → Hovering right at the line, period after period, never dramatic.
      THIS IS A CHRONIC WASTE PROBLEM, not an investigation. Start the
      waste log. Nothing here needs a meeting or a conversation about a
      person, and holding one would be a serious mistake.

  THE RULE:  one period is a DATA POINT.
             two of three in the same direction is a PATTERN.
             three periods bouncing across zero is your COUNTING ERROR
             plus the natural volatility of the product.

Say the rule out loud, because it is the whole section: a single period over threshold gets a look. A pattern gets a project. Operators who treat every period as a pattern burn out and stop counting. Operators who never look for the pattern discover it in month twenty-nine, which is Chapter 1's entire survival curve in one sentence.

Working the meat-and-poultry pattern

Bellwether's chef-owner works Figure 34.6's pattern the way Chapter 11 taught, in order, and the result is worth reproducing because of how ordinary it is.

The bird line is the interesting one, because it is a cost that is nobody's fault and belongs to no category cleanly. The frozen cost card prices the Hearth Chicken's protein at \$5.60 — half of a 3.5 lb air-chilled bird at \$3.20/lb, or 1.75 lb. When the birds arrive at 3.8 lb average, half a bird is 1.9 lb, and at the same \$3.20/lb that is **\$6.08. A \$0.48 per plate overage, on about 108 plates a week and 432 in the period: 432 × \$0.48 = \$207.36.**

Note what did not happen. The supplier did not cheat anyone — you paid by the pound and you received the pounds. The guest was not shortchanged; the guest got eight percent more chicken. What happened is that the plate is now bigger than the card and the price did not move, which is a portioning problem with a purchasing cause. Figure 11.3's own "WHAT IT DOESN'T" line warned about exactly this: the birds that come in over-weight and blow the portion.

The fix is a Chapter 13 fix. A spec is a weight range, not a nominal weight — write it as "3.25–3.60 lb, 3.5 lb average" — and receiving weighs a sample from every case rather than signing for a count. Then you have a documented basis for the credit conversation with the supplier, which you did not have before.

And the residual: \$65 unexplained on \$9,850 of ideal usage is 0.66% — well inside the honest error of a hand count. There is nothing to investigate. There never was. If the chef-owner had opened this investigation at rung seven, they would have spent a week damaging a relationship over \$65 of arithmetic noise and never found the \$997 sitting in plain view on four pieces of paper.

🧮 Run the Numbers

The scale on the pass costs nothing and returns \$3,874 a year.

The ribeye is running 12.9 oz against a 12 oz spec — 0.9 oz over, plated by eye at the height of service by a cook who is doing four things at once and is not being careless.

Beef at \$13.80/lb. 0.9 oz is 0.05625 lb. **0.05625 × \$13.80 = \$0.7763 per plate.**

At 96 ribeyes a week — 384 in the period — that is \$298** a period, and **\$3,874 a year (\$298 × 13).

What it costs to fix: a digital portion scale Bellwether already owns, moved from the prep table to the end of the line; a portion chart taped inside the reach-in door; and ten portions weighed once a week by whoever is on the pass. Incremental cost: zero.

Compare that to Chapter 15's bar arithmetic, which is the same lesson in a different room. The half-ounce over-pour costs \$7,631 a year — 1.76 points of pour cost. The jigger that closes it costs \$233 a year. That is a 32.7-times return, and it is available in every bar in America for the price of a conversation about how drinks get made.

This is what "closing the leaks" actually means. It is almost never surveillance. It is a scale, a jigger, a spec written as a range, and a cost card someone updated in April.

Chapter 15's full accounting is worth restating here because it is the most concentrated leak in the building: the five bar leaks together total \$16,169, which pushes pour cost from 22% to 25.7% and moves the whole restaurant's prime cost from 60.0% to 61.1%. One room, one point of prime cost. Chapter 15 owns those leaks; this chapter owns the controls around them — the jigger, the tenths-method count on the same night every period, the item-level reconciliation of pours against sales, and a bar that gets counted by someone who does not order for it.

34.6 Common theft patterns, front and back, and how each one shows up in the numbers

Before you read a word of this section, read Figure 34.7. Everything in §34.6 is rung seven, and the six rungs above it are not a formality.

FIGURE 34.7 — The investigation ladder
                                         [after the Chapter 11 decomposition]

  #   CAUSE                       WHAT YOU EXAMINE            EVIDENCE    HARM IF
                                                              REQUIRED    YOU'RE WRONG
  1   Prices vs. current costs    cost cards vs. invoices      paper       none
  2   Uncosted specials           the specials log vs. POS      paper       none
  3   Menu-mix drift              the mix report                paper       none
  4   Purchasing                  invoices vs. the order guide  paper       none
  5   Over-portioning             a scale and one service       observed    small
  6   Waste and spoilage          the waste log and the bins    observed    small
  ═══════════════════════════════════════════════════════════════════════════════
  7   THEFT                       a person                      testimony   SEVERE
                                                               + records

  Chapter 11 decomposed a $2,728 variance into six named causes and put theft
  SEVENTH. Chapter 13 ranked the same six by how fast you can check them and
  arrived at the same order. This is not politeness. It is method.

  Rungs 1-6 are settled from documents you ALREADY OWN, in an afternoon, without
  a single conversation about a single human being. Rung 7 costs you a
  relationship even when you are right, and costs far more than the variance
  when you are wrong.

  YOU DO NOT SPEND YOUR MOST EXPENSIVE INVESTIGATIVE TOOL FIRST.

Why the order is both fairer and more accurate

The fairness argument is obvious enough that it does not need me: you should not suspect a person of a crime before you have checked your own paperwork. Let me spend the space on the accuracy argument instead, because it is the one that persuades operators who think fairness is a luxury.

First, base rates. Every variance I have ever personally decomposed resolved primarily onto rungs one through six. Period 8 above is typical: \$997 of \$1,062 explained by a stale cost card, an uncosted special, an unchased credit, a scale nobody used, and birds that came in heavy. Not one dollar of it was anybody's dishonesty. If you begin at rung seven you are beginning at the least likely explanation, which in any other domain we would recognize immediately as bad reasoning.

Second, contamination. An investigation that starts with a suspect does not gather evidence; it gathers confirmation. Once the chef-owner has decided the problem is the new prep cook, every ordinary thing that cook does becomes data. Uncomfortably often, the operator then finds "evidence," acts on it, and never learns that the ribeye was running 12.9 ounces the whole time. The stale cost card is still there after the firing, and so is the variance — and now you have no prep cook and no credibility.

Third, the information supply. This is the one that costs the most money and it is almost never discussed. Your variance investigation depends on information that only your staff can give you: the walk-in that ran warm on Sunday, the case of thighs that went off, the special the sous ran for three weeks, the portioning drift they have noticed and assumed you knew about. The moment your staff learn that reporting a problem produces an investigation of them, they stop reporting problems. Your waste log goes blank, your temperature log gets pre-signed, and your variance becomes permanently unexplainable — which will make you more suspicious, which will make them less forthcoming. That spiral is entirely self-inflicted and I have watched it destroy two otherwise healthy restaurants.

Which produces the single most important sentence in this chapter:

If you punish the waste log, you convert waste into theft — not in reality, but in your numbers.

Fourth, evidentiary cost and quality. Look at the last two columns of Figure 34.7. Rungs one through four are settled with paper you already have, at zero marginal cost, with no risk of being wrong about a person. Rung seven requires evidence about a human being, which is slow, expensive, legally consequential, and frequently ambiguous. A disciplined operator spends cheap, reversible, high-quality evidence before expensive, irreversible, low-quality evidence. That is not softness; it is how any competent investigation of anything is sequenced.

Front of house: the patterns and where each surfaces first

With that established — the patterns are real, and an operator who cannot recognize them is not being kind, they are being negligent. What follows describes each one at the level of what a report shows and what control closes it. That is deliberately where the description stops.

Sweethearting deserves a paragraph of its own, because Chapter 15 described the behavior and deliberately did not claim the term. Here it is: sweethearting is the deliberate giving away or under-ringing of product for someone the employee wants to favor — a friend at the bar, a regular who tips well, another employee off shift. It is the most common front-of-house loss in a bar-driven room and it is the hardest to see, for one structural reason: it leaves no trace on the sales side. No void, no comp, no discount, no reopened check. The check average looks fine. The void report looks fine. Every exception report in §34.3 is clean, because nothing was excepted — the transaction simply never existed.

Which means sweethearting can only be seen from the inventory side, by comparing what was poured against what was sold, item by item, on the ten or fifteen items that carry the most cost. That is Chapter 15's tenths-method count meeting the POS mix report, and it is the single reason a bar gets counted every period rather than every quarter.

And the control that actually prevents it is neither a camera nor a report. It is a rule that everyone knows and everyone follows: every drink is rung before it is poured. No exceptions, not for the owner's friends, not for the health inspector, not for the neighboring chef who came in at midnight. A rule with one exception is not a rule, and the staff will calibrate to the exception rather than the rule within about three weeks.

Back of house

The last two rows are the ones that matter most, and they are the two that get misdiagnosed as theft more than anything else in this business.

Staff meals are a policy vacuum, not a crime. Thirty-one people eat in your building. If nobody has written down what they may eat, they will eat what seems reasonable, and what seems reasonable drifts upward, and none of it appears anywhere except as unexplained usage. I have seen an operator convinced they had a theft problem in proteins when what they actually had was a family meal that had quietly become "whatever is on the line." The fix costs one page: what is available, what is not, when, and how it gets recorded. Ring it at zero with a staff-meal code and it becomes a number you can see and manage instead of a variance you cannot explain.

Unlogged waste is the same failure with a sharper edge, because a blank waste log does not merely hide waste — it converts waste into apparent theft. Every dollar of product that spoiled and was never recorded shows up in exactly the same place, on exactly the same line, as a dollar of product that walked out the door. They are numerically indistinguishable. The only thing that separates them is a piece of paper somebody was willing to fill out, and whether they are willing depends entirely on what happened the last time somebody did.

⚠️ Where the Money Leaks

Where the dollars are, versus where the attention goes.

Put every quantified risk in this chapter on one list, using Bellwether's own numbers.

```text FIGURE 34.8 — Dollars at risk vs. minutes of attention [the Bellwether plan; annual figures]

DOLLARS AT RISK PER YEAR food variance, measured, 2.0 pts of food sales ████████████ $22,282 the five bar leaks from Chapter 15 █████████ $16,169 comps over the 1.0% target █████ $8,680 receiving error at 1% of the order guide ██ $3,348 cash loss at 1% of currency through the drawer ▌ $1,395 time-clock abuse, employee side, estimated ▌ $1,248 ─────── TOTAL $53,122

TYPICAL ATTENTION, IN MINUTES PER WEEK counting drawers and arguing about $1.75 ████████████ ~90 min counting the walk-in ████ ~40 min reconciling invoices to the order guide ██ ~15 min reading the void and comp reports ▌ ~5 min reading the exception summary ▌ ~0 min

The two lists are upside down relative to each other. That inversion is the most common control failure in this industry, and it costs far more than theft does. ```

\$53,122** is **3.4 points of sales**, **20% of the plan's \$261,020 operating profit, and — after \$69,500 of debt service — 28% of what is actually left. Not one dollar of that list requires a dishonest employee to exist.

Now price the controls that address it:

Control	Annual cost	Addresses
Jiggers and a pour-cost review (Ch. 15)	\$233 | \$7,631 of over-pour
Portion scale on the pass (owned already)	\$0 | \$3,874 of over-portioning
Weekly count, 2 people × 1.5 hr, \$22 blended | \$3,432	the whole \$22,282 variance
Second pair of eyes on the nightly count (the closing manager is salaried; only the second person is incremental)	\$1,144	over/short and the deposit
Outside bookkeeper's reconciliation (already in G&A)	\$0 incremental	the only independent reconciler you have
Reason codes and the Monday exception review, 45 min of salaried time	\$0 incremental	comps, voids, reopens
Blind count sheets and a clipboard	\$40	a count that is actually a count
Total incremental	\$4,849** | **\$53,122 at risk

You will not recover all of it — nobody does. Recover half and you have turned \$4,849 into \$26,561, a 5.5-times return. There is nothing else on your capital list that returns that, and it requires no loan, no construction, and no permission from anyone.

🔍 Check Your Understanding

1. A bar's liquor variance runs 6% of ideal usage for three consecutive periods while the void report, the comp report, and the per-bartender check averages all look completely normal. What does that combination suggest, and what is the first thing you check?
2. Why is it more accurate — not just kinder — to check the cost cards before you consider theft?
3. Your waste log has been blank for six weeks and your variance is up 1.4 points. What is the most likely explanation, and what is the wrong conclusion?

(1: A loss on the inventory side with nothing on the sales side means product moved without a transaction — the signature of sweethearting or unrecorded sales. But the first thing you check is still rung one: the pour spec, the cost of the pours on the cocktail cards, and whether a recipe changed. Chapter 15's over-pour arithmetic produces the identical report shape and is far more common. 2: Because base rates favor it — nearly all variance decomposes onto rungs 1–6 — and because an investigation that starts at rung 7 gathers confirmation rather than evidence, and typically leaves the real cause running. 3: The most likely explanation is that waste is occurring and is not being recorded, which makes it numerically indistinguishable from theft. The wrong conclusion is theft. The right move is to find out why the log is blank, and the answer is very often that somebody got in trouble the last time they filled it in.)

34.7 Receiving and vendor fraud: the twenty minutes that decides your food cost

Chapter 13 called receiving the highest-leverage twenty minutes of the day, and the leak ladder in §34.6 puts a number on why: **\$3,348 a year at a one percent error rate** on Bellwether's \$334,776 of annual purchasing. One percent. That is not a scandal; it is a back door held open by somebody who was busy.

The uncomfortable part is that receiving errors are the only item on the ladder that a third party profits from — which makes them the one category where somebody outside your building has an incentive for the failure to continue.

The four things that actually go wrong

Short deliveries. The invoice says forty pounds; thirty-seven arrive. Nobody weighed it. This is by far the most common, and it is usually not deliberate — a case gets left on a truck, a count is estimated at four in the morning. But an operation that never checks will be short more often than one that does, and neither party has to intend anything for that to be true.

Substitutions. You specified an air-chilled bird at \$3.20 a pound and a water-chilled one arrives at the same price. You specified 14-ounce chops and 16-ounce ones arrive, with the invoice reflecting the weight. Chapter 13's specs exist precisely so this is checkable, and a spec nobody checks against is a document, not a control.

Invoice games. The price on the invoice is not the price on the quote sheet. A credit issued last week never appears. A delivery is billed twice — usually an error, occasionally not, and indistinguishable without reconciliation.

Collusion. Rare, real, and the reason receiving should not be permanently assigned to one person who also places the orders. It is also the pattern most likely to be misdiagnosed, because it looks exactly like carelessness until you compare invoices to the quote sheet across several weeks.

⚠️ Where the Money Leaks

"We'll check it later" is the whole failure — and it happens for a structural reason.

The delivery arrives at 9:40 on a Tuesday, mid-prep, in the middle of the one uninterrupted stretch the kitchen gets. Signing the sheet takes four seconds; checking it takes eighteen minutes. The driver has eleven more stops and is standing there.

Every incentive in that moment points at signing. Which is why receiving discipline is not a matter of telling people to be careful — it is a matter of scheduling. If deliveries land during prep and nobody is assigned to receive them, you have designed the failure.

The fix is unglamorous and nearly free: a standing delivery window, a named receiver on the schedule for it, a scale by the door, and the printed order guide on a clipboard. At Bellwether's volume that is about twenty minutes a day of an existing hourly person — roughly \$1,530 a year against \$3,348 of exposure, and it catches the temperature failures Chapter 25 cares about at the same time.

The rule that makes it real: you cannot sign for what you did not check. A signature is an assertion. Teach it that way.

What receiving control actually is

Four practices, in order of value:

1. Weigh and count against the spec, not against the invoice. The invoice tells you what they intend to bill. The spec tells you what you ordered. Checking the delivery against the invoice only confirms the invoice is internally consistent.
2. Never sign a blind delivery. If paperwork is missing, record your own count, have the driver initial it, and reconcile when the invoice arrives.
3. Reconcile invoices to the quote sheet weekly, not monthly. Prices move; that is normal. Prices moving without notice is what you are looking for — and at monthly cadence you find it three weeks after you could have called about it.
4. Track credits to closure. A credit memo issued is not a credit received. An unreconciled credit is money you have already decided you are owed and never collected.

⚖️ Code and Compliance

Two things worth knowing before you accuse a vendor of anything.

Most supply agreements carry a dispute window — often short, sometimes seven days — after which a billing error is deemed accepted. That is a commercial term rather than a legal trap, and it is the practical reason weekly reconciliation matters more than the principle does.

And a pattern is not proof. If invoice prices are drifting above the quote sheet, the overwhelmingly likely explanation is a price change communicated to somebody who no longer works here, or a rep quoting from a stale sheet. Raise it as a reconciliation question, in writing, with the documents attached. If it does not resolve, escalate commercially — change vendors — long before anyone reaches for the word fraud. Accusations of fraud against a business partner carry their own legal consequences, and this is a place for counsel rather than instinct.

34.8 Surprise counts, audits, and the question controls actually raise

A control that runs on a published schedule measures compliance with the schedule. A control that runs unannounced measures the thing itself.

That is the whole argument for surprise counts — an unscheduled inventory of one high-value category, conducted by somebody who does not normally count it. At Bellwether the obvious candidates are the spirits well, the by-the-glass wine core, and the proteins in the walk-in: three categories carrying most of the value and nearly all of the variance.

That last line is the one most often skipped and the one that matters. A surprise count recorded only when it finds something is not a control, it is a search. The value of the boring results is that they establish what normal looks like — which is the only reason the interesting one is legible when it arrives.

👨‍🍳 On the Line

What a surprise count feels like to the person being counted around.

Here is the thing nobody writes down. The first time an owner walks into a bar at 4:15 with a clipboard and starts weighing bottles, the bartender's stomach drops — not because they have done anything, but because being measured unexpectedly feels like being suspected.

That reaction is not irrational and it does not fade on its own. It fades when the count is routine, universal, and explained: when it happens to every category on a rotation, when the result is shared with the person whose section it is, and when a clean count is acknowledged out loud.

An owner who counts only the bar, only after a bad week, and never mentions the outcome has not built a control. They have built a signal that they suspect the bartender — and Chapter 21 already priced what that does. The reliable people leave first, because they are the ones with options.

Say what you are doing and why, before the first time. "We count one category a month, everybody's section, so that when a number moves we know whether it's real." Nine seconds, and it converts the same physical act from an accusation into a system.

The question controls raise

There is a real tension in this chapter and it deserves stating plainly rather than resolving cheaply.

Controls impose a cost on trust. Every one of them — the two-person drop, the void authorization, the surprise count, the second signature at receiving — is a small statement that the restaurant does not rely on any individual's word alone. In a business Chapter 21 argued runs on relationships, that is not free.

The resolution is not to skip them. It is to be honest about what they are for, and the honest framing is this: controls protect the innocent at least as much as they catch the guilty.

• A manager who counts the drawer with a second person cannot be accused of a shortage.
• A bartender whose section is counted on the same rotation as everyone else's has evidence that their variance is normal.
• A cook whose portioning is measured can demonstrate that the food-cost problem is a purchasing problem.

In a restaurant without controls, an unexplained loss lands on whoever the owner happens to suspect. That is a far worse place to work than one with a clipboard. Say so, and mean it, and most of the tension dissolves.

34.9 Responding well: investigation, evidence, and the decisions you make once

Sooner or later a number will not explain itself, the first six rungs will come back clean, and you will be looking at something that appears deliberate. What you do in the following week decides whether the restaurant ends up with a resolved problem — or a lawsuit, a departed innocent employee, and a leak that is still running.

Read this section twice, because nearly every instinct in the moment is wrong.

What to do

Stop and document before you act. Write down what you observed, when, and what the numbers show. Assemble the POS audit trail, the variance history, the DSR reconciliations, and the schedule. Do this before speaking to anyone — memory reorganizes itself around a suspicion with remarkable speed.

Establish the pattern, not the incident. A single anomalous night is noise. What you need is the same signature across multiple periods, correlated with something: a shift, a terminal, a category, a vendor.

Consider every innocent explanation in writing, and rule each one out on evidence. This is what separates an investigation from a search for confirmation, and it protects you legally as well as morally.

Involve counsel before, not after. Employment law varies enormously by state, and the consequences of mishandling an investigation, an interview, or a termination are substantial — wrongful termination, defamation, and wage-claim exposure among them. A conversation with an employment attorney before the first interview is cheap. The one after the termination is not.

Keep the circle small. A suspicion that circulates through a staff of thirty-one is unrecoverable even if it turns out to be wrong — particularly if it turns out to be wrong.

What not to do

• Do not confront on the floor. Ever, for any reason.
• Do not accuse before the boring explanations are excluded. Chapter 11 put theft seventh for a reason. An accusation you have to withdraw costs you the employee, the team's trust, and possibly a claim.
• Do not interview alone, and do not let it become an interrogation. You are gathering information, not obtaining a confession — and a confession obtained under pressure is worth very little while creating exposure of its own.
• Do not "handle it quietly" by cutting someone's hours until they leave. It is a constructive dismissal problem, it teaches the staff precisely the wrong lesson, and it leaves the control gap open for the next person.
• Do not decide about law enforcement in the first hour. That decision has a legal dimension, an insurance dimension (your policy may require notification), and a human one. Make it with counsel, deliberately, once the facts are established.

🤝 Hospitality

The people hurt most by a leak are usually the ones who did not cause it.

Worth holding on to, because it is what keeps this chapter's framing honest.

When \$53,122 walks out of a restaurant it does not come out of an abstraction. It comes out of the raise that does not happen, the second dishwasher who is not hired so the closing shift runs an hour long, the wage ladder in Chapter 21 that stays theoretical — and eventually, in Chapter 39, the jobs themselves. Chapter 17 established that your people are the product. A business leaking a fifth of its operating profit cannot invest in them.

So controls are not, finally, an expression of distrust toward staff. They are how a restaurant keeps the money that pays them. An operator who can say that honestly — and who has built controls that are universal, routine, and explained rather than targeted and secretive — will find that good employees are not merely tolerant of them. They are the ones who ask why the count did not happen.

🍽️ The Business Plan

Checkpoint 34 of 40 — the Controls section.

The plan can now measure itself. This checkpoint makes the measurements trustworthy.

The exposure, from §34.6:

The control program:

Recover half and \$4,849 becomes \$26,561 — a 5.5× return. Nothing else on the plan's capital list returns that.

The policies adopted:

1. Authorization. Comps require a manager; discounts require a manager and a reason code; voids are reviewed weekly regardless of value, because they touch no financial statement.
2. Cash. Two-person count at close, sealed drop, no banking by the person who rang the sales, and a daily sales report reconciling sales → cash → cards → deposits.
3. Variance thresholds. Investigate at ±1.0 point of food cost or ±\$400 in a period, whichever is smaller; escalate at ±2.0 points sustained across two periods.
4. The investigation order is written down — stale cost cards, uncosted specials, mix drift, purchasing, over-portioning, unrecorded waste, and only then people.
5. Surprise counts. One category a month, irregular intervals, counted by someone other than the usual counter, result recorded whether or not it is interesting.
6. Receiving. Standing window, named receiver, scale by the door, order guide on the clipboard, weekly invoice-to-quote reconciliation, credits tracked to closure.

What this checkpoint does not settle. Whether a 31-person restaurant with four managers — two of them the owners — can sustain genuine separation of duties. It cannot, fully; §34.1 said so. The plan substitutes surprise counts, exception review, and owner attention for structural segregation, and that substitution depends entirely on the owners actually doing it, in a week Chapter 19 already showed is fully committed.

Open questions carried forward:

1. If controls recover half the \$53,122, where does \$26,561 go — margin, wages, or the reserve? (Chapters 21 and 33)
2. What happens to this program in a bad quarter, when counting hours are the first thing cut? (Chapter 39)
3. Who audits the owners? (Chapters 35 and 37 — it becomes a real question at two units)

Conclusion

Controls are not an accusation, and a chapter that reads like one has taught the wrong thing.

They are the systems that let a restaurant know what happened — and their absence is what allows a small dishonesty to become a habit, an honest error to be mistaken for a dishonest one, and \$53,122 to leave a business that clears \$261,020 before debt service.

The arithmetic is the argument: three and a half points of sales, a fifth of the operating profit, and twenty-eight percent of what is actually left — against a control program costing \$4,849. You will not recover all of it; nobody does. Recover half and the return is five and a half times.

But the sequence matters more than the total. Chapter 11 put theft seventh on a list of seven, and this chapter has kept it there for two reasons worth restating together. The first is accuracy: base rates favor the boring explanations, and nearly all variance decomposes onto the first six rungs. The second is that an investigation beginning at rung seven gathers confirmation rather than evidence — and typically leaves the real cause running while a good employee is quietly pushed out.

A restaurant with universal, routine, explained controls is a better place to work than one without, because in a building with no controls an unexplained loss lands on whoever the owner suspects. The clipboard protects the innocent. That is not a consolation prize for the discipline. It is the main reason for it.

Chapter 35 opens Part VIII by asking what to do when all of this works — when the numbers are real, the leaks are closed, and somebody suggests opening a second one.

Key Terms

Internal control — the procedures that make a business's own numbers trustworthy: authorization, recording, custody, and reconciliation, arranged so no single person completes a transaction end to end unobserved. (Ch. 34)

Separation of duties — the principle that the person who authorizes a transaction, the person who handles the asset, and the person who records it should be different people. Impossible to achieve fully in a 31-person restaurant with four managers, which is why the practical substitutes matter. (Ch. 34)

Cash handling procedures — the rules governing currency from drawer to bank: assigned banks, two-person counts, sealed drops, and a prohibition on the person who rang the sales reconciling them. (Ch. 34)

Comp / void / discount authorization — the requirement that each carries a named authorizer and a reason code. A comp reduces revenue after production; a discount is a pricing decision; a void touches no financial statement at all, which is why it is reviewed by count rather than by value. (Ch. 34)

Inventory variance — the difference between theoretical usage (what the cost cards say should have been used) and actual usage (what the count says was used). The most informative single control number in a restaurant. (Ch. 34)

Sweethearting — giving product away to friends, family, or regulars without ringing it. Appears as inventory variance with no corresponding sale, and is distinguishable from unrecorded waste only by whether the waste log is being kept. (Ch. 34)

Over-ringing — charging a guest more than the item's price and removing the difference, or ringing an item and voiding it after payment. Shows up as an unusual void or post-payment adjustment rate on one operator's audit trail. (Ch. 34)

Shorting — delivering less product than was paid for, whether at the bar (a short pour) or at the loading dock (a short delivery). (Ch. 34)

Daily sales report (DSR) — the one-page daily reconciliation of sales to cash to cards to deposits, including comps, voids, discounts, and the over/short. The document that makes every other control checkable. (Ch. 34)

POS audit trail — the point-of-sale system's record of who did what: voids, comps, discounts, reopened checks, deleted items, no-sale drawer opens, and the operator ID attached to each. (Ch. 34)

Vendor fraud — billing that does not match what was delivered or quoted: short deliveries, unauthorized substitutions, price drift from the quote sheet, duplicate invoices, and unissued credits. The one leak category a third party profits from. (Ch. 34)

Surprise count — an unscheduled inventory of a single high-value category, conducted by someone other than the usual counter, with the result recorded whether or not it is interesting. (Ch. 34)

Spaced Review

1. Without looking back: name the seven rungs of the investigation ladder in order, and give the two reasons theft is seventh rather than first.
2. From Chapter 15: a half-ounce over-pour costs \$7,631 a year and the jigger that closes it costs \$233. Explain why this is a control problem rather than a training problem, and state what the daily sales report would have to show for you to notice it at all.
3. From Chapter 31: the flash report showed comps at 1.56% of gross against a 1.0% target, and 41 voids worth \$128. Which figure is more concerning, and why is the answer counterintuitive?
4. From Chapter 11: a period shows \$1,062 of unexplained food variance. Walk the first three rungs of the ladder and state, for each, which document you would pull and roughly how long it would take.
5. The recurring question: the control program costs \$4,849 and puts \$53,122 at stake. Does spending it move prime cost, and in which direction? Then name which of the six leaks would appear in prime cost at all, and where the others hide.