Chapter 37: Multi-Unit Management: Systems, Standards, and Running What You Can't Watch

"The second restaurant is not a restaurant. It is a copy of a document you have not written yet." — constructed; what nobody tells you before you sign the second lease

Overview

You are standing in the second dining room on a Thursday, and the chicken is wrong.

Not bad. The bird is cooked correctly, the skin is right, the plate is hot. But there is too much salsa verde — maybe an ounce too much — the roots are cut on a bias instead of halved, and the ticket went out at eleven minutes instead of fourteen because the cook on that station has decided, sensibly, that faster is better. Every one of those choices was made by a competent person trying to do a good job. Not one of them was wrong on its face. All of them together produce a plate that is not the plate.

Here is what you discover when you go looking for who to blame: there is no one. You never wrote it down. For four years you set the standard by standing eighteen inches away from the pass and saying "less" or "more" or "again," and it worked perfectly, and it produced exactly zero transferable information. Your quality system was your body in a room. You have just opened a second room and your body is not in it.

That is the whole chapter. The multi-unit skill is not cooking and it is not hosting. It is writing down what "good" looks like precisely enough that it happens in a building you are not standing in.

Chapter 35 ran the second-location test on the Bellwether plan and returned a not yet — the business is not profitable once both owner jobs are paid at market, and there is no management bench. Chapter 36 ran the franchise test and returned the same verdict for a different reason: you cannot sell a system you have not written. This chapter is the constructive half of both answers. It describes what the systems answer actually looks like — the manual, the audit, the reporting package, the supervisory layer, the dashboard — and then it does something that will surprise you. It goes through the business plan you have been building for thirty-six chapters and finds that most of an operations manual is already in there. Not because anyone set out to write one, but because you cannot cost a recipe without standardizing it, cannot build a staffing guide without defining the positions, and cannot write a food safety plan without writing procedures. Documentation turns out to be a by-product of measurement.

Chapter 34 ended on a question it did not answer: who audits the owners? In one unit, the owners are the control — they sign the checks, count the walk-in, and read the void report. At two units they are a bottleneck. At three they are a rumour. Answering that question honestly is the fastest route into everything else in this chapter, so we take it first.

In this chapter, you will learn to:

• Separate the four things a single-unit owner does at once — set the standard, teach it, detect deviation, correct it — and explain why growth forces them apart.
• Decide what belongs in an operations manual using a single test, and write a specification, a procedure, and a standard without confusing the three.
• Build a weighted standards audit with critical items, and explain why the score is the least important thing on it.
• Compute span of control as a time budget, and describe what a district manager's product actually is.
• Compute the scale at which central production pays, and name the four conditions that have to hold.
• Compute comparable-store sales and decompose them into traffic, price, and mix — the single most important arithmetic in multi-unit reporting.
• Set exception thresholds with owners and consequences, and name the way each metric will be gamed.

Learning Paths

🏗️ Opening — you are not opening a second unit. Read §37.2 anyway, and read it before you open the first one: writing the standard down while you are building it costs a tenth of what it costs to reconstruct it four years later from three cooks' memories. 📋 Managing — this is your chapter end to end. §37.3, §37.4, and §37.7 are the actual daily work of a general manager who reports to somebody and of every district manager who ever visits you. 🍸 Beverage — §37.3 and §37.5 are yours. Beverage is the easiest program in the building to standardize (a cocktail is a recipe executed with a jigger) and among the hardest to audit, because pour cost travels across units and so does the reason it moved. 🚚 Small Format — §37.5 is the center of your world. For a truck or a stall, the commissary is not an economic choice at all; in most jurisdictions it is a licensing requirement, and Chapter 30 priced it. Read §37.6 too: two trucks are a multi-unit business with the same reporting problem.

37.1 The transition: from operator to operator-of-operators

Multi-unit leadership is the practice of producing restaurant performance through other people in buildings you are not in. That definition sounds bland until you notice what it removes. Everything you are good at — reading a room, tasting a sauce, cutting the floor at nine, catching the server who is in the weeds before they know it — requires you to be present. Presence is the input to every skill that got you here, and multi-unit management is the discipline of operating without it.

The transition is usually described as a promotion. It is better understood as a change of product. A single-unit operator's product is a good service. A multi-unit operator's product is other people's good services, which means their actual daily output is instructions, standards, feedback, and hiring decisions. If you spend your Tuesday expediting at Unit Two, you have produced one good service and zero instructions, and you have made your real job harder, because the unit's manager has just learned that you will do it if they don't.

The four jobs nobody named

Watch a competent chef-owner for a week and you will see four distinct activities that they perform continuously, in the same body, with no boundary between them and no name for any of them:

1. Setting the standard. Deciding what good is. Two ounces of salsa verde, not three. A greeting inside thirty seconds. Fourteen minutes on the hearth, not eleven.
2. Teaching the standard. Making other people able to hit it. Usually by demonstration, occasionally by shouting, almost never in writing.
3. Detecting deviation. Noticing that it slipped. This is nearly always visual and immediate — the plate goes past and something is off.
4. Correcting deviation. Fixing it now, on this plate, before it leaves the pass.

In one restaurant these four collapse into a single continuous act that feels like "running the place." Growth pulls them apart, and it pulls them apart in a specific order that is worth knowing in advance.

FIGURE 37.1 — THE FOUR JOBS AND WHO ACTUALLY DOES THEM        [constructed teaching example]

  THE JOB                    1 UNIT             2 UNITS            4+ UNITS
  ──────────────────────────────────────────────────────────────────────────────
  SET the standard           owner              owner              owner
  (decide what good is)      by taste           by taste           in writing

  TEACH the standard         owner              owner              a training
  (make people able)         by showing         by driving         system + GMs

  DETECT deviation           owner              owner              reports +
  (notice it slipped)        by looking         by looking,        audits +
                                                half as often      a district manager

  CORRECT deviation          owner              unit manager       unit manager
  (fix it today)             on the spot        sometimes          always
  ──────────────────────────────────────────────────────────────────────────────
  WHAT BREAKS                nothing — the      the owner is       nothing, IF the
                             owner IS the       the BOTTLENECK     system exists.
                             system             for all four       Everything, if not.

Read the middle column, because that is where most independent groups die. At two units the owner is still doing all four jobs, but each one at roughly half strength, in two buildings, with twice the administrative load and no help. Nothing has been delegated. Nothing has been written. The only thing that changed is that the owner is now absent half the time from both rooms instead of present in one. Two units is the worst number in this business — too many for one person's presence, too few to pay for a layer of management that would replace it. We will price that precisely in §37.4.

The right column is what a functioning multi-unit business looks like, and notice what happened to the owner's job. They still set the standard. They no longer teach it, detect deviation in it, or correct it. Those three moved into systems: a training program, a reporting package, an audit, and a person whose full-time job is the units they cannot visit.

👨‍🍳 On the Line

The first shift you don't work.

You will take a Friday off eventually, and it is worth knowing what you will find on Saturday, because operators consistently misread it.

Some things will be worse. The walk-in will be stacked differently and something will be buried. The mise on garde manger will be cut by someone who has watched you do it forty times and never been told why the cut is what it is. Two tickets will have gone out of sequence. A comp will have been given that you would not have given.

And some things will be better, which is the part nobody warns you about. Someone will have found a faster way to run the fish station. A server will have handled a complaint in a way you would not have thought of and the table left happy. The prep list will have been reorganized by a cook who is better at organizing than you are.

Here is the skill, and it is the actual skill of multi-unit management: you have to be able to tell the difference between deviation and improvement, and you have to have a mechanism for absorbing the second one. An operator who treats every difference as a failure gets compliance and nothing else, and the good people leave — Chapter 21 explained exactly why they leave first. An operator who treats every difference as innovation has no standard at all and four units that share a logo.

The mechanism is not complicated. It is a written standard, plus a route by which anyone can propose a change to it, plus a named person who decides, plus a date when the document changes. Without the last two it is a suggestion box, and everyone in a kitchen knows what a suggestion box is for.

The presence you are about to give away

Operators plan the second location as a real-estate and construction problem. It is a time-allocation problem, and the arithmetic is unforgiving enough that it is worth doing on paper before you do it with money.

🧮 Run the Numbers

What happens to supervision when you open the second building.

Bellwether runs seven services a week: five dinners, Tuesday through Saturday, plus two weekend brunches. Two owner-partners, both salaried, are in the building for essentially all of them.

Today, one unit. Call it 58 hours a week for the chef-owner and 55 for the front-of-house partner: 113 hours of owner time. Of that, roughly 35 hours is desk work that has to happen whether or not anyone is eating:

Desk task	Hours/week
Ordering, receiving follow-up, vendor calls	6
Inventory and the weekly flash report (Ch. 31)	5
Scheduling (Ch. 19)	4
Payroll and administration	4
Marketing, reservations, private-event inquiries	6
Hiring, reviews, coaching, documentation (Ch. 17, 21)	5
Maintenance, compliance, licensing (Ch. 25)	5
Total desk	35

That leaves 78 hours of floor presence across 7 services — 11.1 hours per service. Someone who owns the place is standing in the room for essentially the entire life of every shift.

Now open unit two. The desk work does not double, but it does not stay flat either: two payrolls, two schedules, two inventories, two flash reports, plus a consolidated report that did not exist before, plus a second landlord, a second health department file, and a second set of vendor accounts. Call it 1.7×: 35 hours becomes roughly 60.

Floor presence: 113 − 60 = 53 hours, now spread across 14 services in two buildings.

$$\frac{53 \text{ hours}}{14 \text{ services}} = 3.8 \text{ hours per service}$$

Supervision per service falls from 11.1 hours to 3.8 — a 66% cut — in the same year you open a restaurant staffed entirely by people who have never worked for you. A brand-new unit with a brand-new team needs more supervision than a seasoned one, not less. You have delivered a third.

To restore 11.1 hours per service across fourteen services you would need 155 floor hours plus 60 desk hours: 215 hours from two people who currently supply 113. The 102-hour gap is roughly two and a half full-time management positions.

There are exactly two ways to fill that gap: a person, or a document. The person costs a salary every year forever. The document costs the owners' time once and then costs almost nothing. Nearly every group that survives its second unit buys some of both, in that order — and nearly every group that fails buys neither and hopes that driving faster will work.

That last sentence is not a joke either. The characteristic behavior of an under-systematized two-unit operator is driving. They are in the car four times a day, they eat standing up, they answer the phone during service, and their answer to any problem in either building is to physically go there. It works for about eleven months, at which point both restaurants are slightly worse than the original was, the owners are exhausted, and the numbers show a group that is bigger and less profitable.

37.2 The operations manual: what has to be written down, and how detailed is too detailed

The operations manual is the written record of how this restaurant does what it does — its standards, specifications, procedures, and the decisions that have already been made — organized so that a competent stranger can find the answer without asking you.

Read that definition again and notice the last clause, because it is the whole functional test. A manual is not a rulebook and it is not a training document, though it feeds one. A manual is a decision archive. Its purpose is to stop the business from re-deciding things it has already decided. Every time a cook asks "how much salsa verde?", the business is paying to make a decision it made four years ago. Multiply by six hundred questions a week across four buildings and you have the actual cost of not writing things down: it is not the wrong plates, it is the interruption.

The write-it-down test

The reason most operations manuals are useless is that nobody applied a filter. Somebody sat down with good intentions, wrote everything they could think of, produced two hundred pages, put it in a binder, and put the binder in the office where it has remained, out of date, ever since. A manual that contains everything contains nothing, because nobody can find anything in it.

Here is the filter. It has two questions.

FIGURE 37.2 — THE WRITE-IT-DOWN TEST                          [constructed teaching example]

  Would two competent people, both genuinely trying to do it right,
  do this DIFFERENTLY?
        │
        ├── NO ──► Don't write it. It is craft, or it is obvious, and
        │          writing it down insults the reader and rots on the shelf.
        │          ("Hold the knife safely." "Be pleasant to guests.")
        │
        └── YES ─► Can you tell, from the FINISHED RESULT, whether
                   it was done correctly?
                        │
                        ├── YES ──► Write a SPECIFICATION.
                        │           Describe the OUTPUT. Leave the method to
                        │           the person doing the work.
                        │           ("2 oz salsa verde. 4 root halves. Chicken
                        │            skin-side up, leg to 6 o'clock. Plate hot.")
                        │
                        └── NO ───► Write a PROCEDURE.
                                    Control the PROCESS, because the outcome
                                    cannot be inspected after the fact.
                                    (Cooling. Sanitizer concentration. Allergen
                                     handling. Cash drops. Voids. ID checks.)

The second question is the one that does the real work, and it is the principle underneath HACCP (Chapter 25). You cannot look at a finished container of soup and see whether it was cooled from 135°F to 41°F inside the required window. You cannot taste a cutting board and detect the allergen. You cannot look at a cash drawer at midnight and see the sale that was never rung. Where the outcome is invisible, you must control the process; where the outcome is visible, control the outcome and leave the process alone.

That single rule answers "how detailed is too detailed" better than any page count. Over-specifying is writing a procedure where a specification would do: three paragraphs on how to plate a salad, when the photograph-in-words of the finished salad and a weight would have been enough. It is insulting to a skilled cook, impossible to enforce, and guaranteed to be obsolete within a season. Under-specifying is "plate attractively," which means nothing, transfers nothing, and produces the salsa verde problem in the Overview.

Three documents that are not the same thing

Confusing these three is the most common structural error in restaurant manuals.

A standard without a number is a wish. "Fast ticket times" is not a standard; "90% of entrée tickets under 19 minutes, measured from fire to pickup on the POS" is. A standard without a consequence is a report; we will handle consequences in §37.7.

⚠️ Where the Money Leaks

The manual that rots — and why a rotten manual is worse than no manual.

Every section of a manual needs three things stamped on it: an owner (a role, not a person), a last-reviewed date, and a review interval. Without them, documents decay at a rate you can almost predict. Cost cards go stale the first time a supplier moves (Chapter 11). Par levels go stale the first time the menu changes seasonally (Chapter 13). Staffing guides go stale when volume moves ten percent (Chapter 19). Wage and hour policy goes stale when the law changes, which it does.

The financial leak is obvious: a stale cost card prices a plate against last year's protein contract, which is how a 29.4% item quietly becomes a 34% item.

The bigger exposure is not financial. A documented standard you do not follow is worse than an undocumented one. If your manual says sanitizer buckets are changed every two hours and your logs show four hours, you have created a written record of your own non-compliance. If your handbook promises a complaint procedure you do not actually run, you have created an expectation you failed. In an inspection, a wage-and-hour matter, or any dispute, the gap between your written standard and your actual practice is the most damaging document in the building — and you wrote it yourself.

The discipline is unglamorous: a review calendar. Four sections a month, thirty minutes each, one manager. Anything that has not been reviewed in twelve months is presumed wrong until someone looks.

What actually has to be in it

Here is the structure. Fifteen sections, which is enough for a group of any size under a dozen units, and few enough that people can hold the map in their head.

Hold that table. In the Business Plan checkpoint at the end of this chapter we are going to walk down it, section by section, against what the Bellwether plan has already produced — and the result is the most useful thing in the chapter.

⚖️ Code and Compliance

The sections that are not optional, and the ones that need a lawyer.

Most of the manual is your choice. Several parts are not, and several more carry enough legal weight that they should not be drafted by an operator alone.

• Food safety (§5). Your jurisdiction's adopted version of the FDA Food Code governs holding temperatures, cooling, employee health and exclusion, handwashing, and allergen awareness. Certain operations require a written HACCP plan or a variance. A Certified Food Protection Manager requirement is common. Chapter 25 covers this; the manual is where it lives day to day.
• Employment policy (§10). Wage and hour rules — overtime, the tip credit where it exists, tip-pooling eligibility, meal and rest breaks, predictive scheduling in the cities that have it — vary enormously by state, county, and city, and several states have no tip credit at all. Your handbook, your complaint procedure, and your anti-harassment policy should be reviewed by an employment attorney before they are distributed, and re-reviewed when you cross a state line. Form I-9 is federal and applies to every hire, at every unit, on day one.
• Alcohol (§6). Licensing, permitted hours, service refusal, ID verification, and dram shop exposure are state and local. A written refusal-of-service procedure and documented server training are among the cheapest risk reductions available to a restaurant. Chapter 8 covers the licensing.
• Accessibility. The Americans with Disabilities Act applies to the building and to service. A second location is a second building and a second obligation; do not assume compliance travels.

Everything above varies by jurisdiction and changes over time. When you open in a second city you are opening under a second set of rules, sometimes a materially different one. Verify locally, with a licensed professional, before you write the policy — and treat the multi-jurisdiction problem as a real cost of expansion, not an afterthought.

37.3 Standards and audits: measuring consistency without micromanaging

Brand consistency is the guest getting the same experience at any of your units on any day they walk in. That is the usual definition and it is subtly wrong in a way that matters operationally.

Consistency is not a statement about your average. It is a statement about your variance. No guest ever eats at the average. A group whose four units score 92, 91, 93, and 44 has an average of 80, which sounds like a mediocre group; in fact it is three good restaurants and a fire, and the fire is generating the reviews that define all four. Consistency work is variance-reduction work, and the first thing it requires is a measurement that can actually detect variance — which means the same instrument, applied the same way, at every unit.

That instrument is an audit.

What an audit is, and the four kinds

An audit is a scheduled, scored, documented inspection against written standards. The written-standards part is not optional: an audit against an unwritten standard is just an opinion with a number on it, and every unit manager can tell the difference immediately.

The peer audit is the most underused of the four in independent groups and the highest-yield. When the manager of Unit Two audits Unit One, two things happen: Unit One gets inspected, and Unit Two's manager spends four hours looking closely at how somebody else solves the same problems. The learning runs backwards up the clipboard, and it is free.

Scoring: the trap of the flat checklist

Here is the design error that ruins most audit programs. Somebody builds a hundred-item checklist where each item is worth one point. Item 34 is "mop sink area clean and organized." Item 61 is "all cold holding at or below 41°F." Both are worth one point.

You have just told every manager in your company that a clean mop sink and a functioning walk-in are equally important. They are not stupid; they will clean the mop sink, because it is cheaper. The score will go up and the risk will not go down.

Two fixes, and you need both:

Weight by consequence. Sections get point allocations proportional to what failure actually costs — a closure, a lawsuit, a sick guest, a lost regular, a point of prime cost. Cosmetic items are worth cosmetic points.

Add critical items that fail the audit outright. A small number of items — cold holding, cooking temperatures, handwashing, allergen procedure, a required certification, a documented cash control, a blocked exit — cap the result regardless of score. This mirrors how health codes themselves work: the FDA Food Code framework distinguishes items directly linked to foodborne illness from items that are not, and inspectors treat them differently. Your audit should too.

🧾 Read the Numbers

```text FIGURE 37.3 — "The 94 that was really a fail" [constructed teaching example] THE ARTIFACT Quarterly leadership operations audit, one unit of a constructed three-unit casual full-service group. Conducted unannounced on a Wednesday, 4:15 p.m. — deliberately during the shift transition. THE CONTEXT Unit 2, open three years, best financial performer in the group. Manager has been in place two years. Last audit, nine months ago, scored 91 with no criticals.

                 SECTION                              POSSIBLE   EARNED
                 ──────────────────────────────────────────────────────
                 Food safety and sanitation               30       26
                 Product and recipe adherence             25       24
                 Service and hospitality standards        20       19
                 Financial controls and cash handling     15       15
                 Facility, safety, and compliance         10       10
                 ──────────────────────────────────────────────────────
                 TOTAL                                   100       94

                 CRITICAL ITEMS                                  RESULT
                 ──────────────────────────────────────────────────────
                 Cold holding at or below 41°F           walk-in at 44°F   FAIL
                 Cooling log complete for 7 days         4 entries blank   FAIL
                 Cook temperatures logged                        complete   pass
                 Handwash sinks stocked and accessible           complete   pass
                 Allergen procedure posted and known             complete   pass
                 Cash control: two-signature deposit             complete   pass
                 ──────────────────────────────────────────────────────
                 AUDIT RESULT: CONDITIONAL — 2 critical failures
                 (Group rule: any critical failure caps the result at
                  Conditional regardless of point score.)

                 The six non-critical points lost: dusty hood return; three
                 unlabeled squeeze bottles; one server could not name two
                 wines by the glass; allergen matrix on the wall is one menu
                 change out of date; two burnt-out bulbs in the entry; ice
                 machine scoop stored in the ice.

WHAT IT SHOWS A 94 reads like an A, and this unit is the group's financial star. Two of the six critical items failed, and they are the only two on the page that could hurt a guest or close the building. Everything else on the audit is real, reversible, and worth a Tuesday. WHAT IT DOESN'T It does not say WHY the walk-in was at 44°F. Three very different diseases produce the same reading: a failing compressor (a capital problem, ~$3,000–$9,000), a door propped during a delivery (a procedure problem, free to fix), or a box that has never held 41°F while the logs said it did — which is a falsification problem and by far the worst of the three, because it means every temperature record in the building is now worthless. An audit identifies the wound. It never identifies the weapon. It also cannot see hospitality. Nineteen of twenty service points tells you the sequence was followed. It does not tell you whether anybody in that room had a good time. THE DECISION Three things before Friday, each with a role and a date. (1) A refrigeration technician on the walk-in with a written findings report — not a phone call. (2) The cooling log moves from an end-of-shift sheet to a time-stamped entry initialed at the point of cooling, and the manager pulls and reviews it daily for four weeks. (3) An unannounced re-audit of the food-safety section only, within fourteen days, by someone other than the original auditor. The six cosmetic items go on the unit's period action plan with a role and a date. None of them go to the owners. THE LESSON An audit that does not distinguish critical from cosmetic will always be optimized toward the cosmetic, because the cosmetic is cheaper. Score for information; act on the criticals. And when your best-performing unit fails two criticals, assume it is not an anomaly — assume it is what the other units look like too. ```

The micromanagement line

The fastest way to destroy an audit program is to fill it with items that measure obedience rather than outcome. Every added line looks free to the person writing it and costs a manager real minutes every week. Two tests keep the instrument honest:

Would a guest, a regulator, or the P&L notice? If the answer is no for all three, cut the item. The color of the pen on the prep list is not a standard. The prep list existing and being accurate is.

Does the item measure a result, or a method? Prefer results. "Walk-in organized by cook date, oldest forward, nothing unlabeled" is a result and a good item. "Walk-in reorganized every Tuesday and Friday" is a method, and it will be performed on Tuesday and Friday regardless of whether the walk-in is organized.

There is also a hard practical ceiling. An audit that takes more than about ninety minutes will be rushed, and an audit that is rushed produces numbers that look like data and are not. Fifty to seventy well-chosen items, weighted, with six to ten criticals, is a real instrument. Two hundred items is a ritual.

Gaming, and what to do about it

Every audit gets gamed. Assume it, design for it, and do not take it personally.

The classic is the announced visit: the district manager comes Wednesday at ten, so Tuesday night is spent cleaning, and what gets audited is a restaurant that has been prepared for an audit. Fixes:

• Announce a window, not a time. "Some time in the next three weeks" preserves fairness and destroys the value of a one-night clean.
• Audit during service or transition, not at ten in the morning. Mise, sanitizer, holding temperatures, and ticket times are only real at 6:40 p.m.
• Weight the metrics that cannot be staged. POS ticket times, void and comp rates, cash over/short, overtime hours, and the text of guest reviews are generated continuously and cannot be tidied up the night before. These belong in the reporting package (§37.6), not the audit, but read them together.
• Audit the audit. Re-score a sample of self-audits. If a unit self-scores 96 and the leadership audit returns 78, the gap is the finding — and it is a bigger finding than the 78.

🤝 Hospitality

Standardize the promise, not the sentence.

There is one place where the logic of this chapter has to stop, and it is the place the book cares most about. You can standardize the timing of a greeting — thirty seconds, every table, measured. You cannot standardize the words, and every operator who has tried has produced the same result: a dining room where every server says the identical warm sentence and every guest can hear that it is not warm. Scripted hospitality is audible. It reads as contempt, which is precisely the opposite of what it was written to produce.

The workable line runs like this. Standardize the obligations; leave the expression free.

• Obligation: every table is acknowledged within 30 seconds of being seated. Free: how.
• Obligation: every server can describe every dish and name three wines by the glass without looking. Free: the description.
• Obligation: when something goes wrong, the manager visits the table before the check drops. Free: what is said and what, if anything, is comped.
• Obligation: nobody waits more than two minutes with an empty water glass. Free: who fills it.

Audit the obligations, because they are observable and they are the structure that makes hospitality possible. Do not audit the sentence. Chapter 23 made the commercial case — the second visit is where the business lives — and a second visit is produced by a person who was allowed to be a person.

37.4 Span of control and the district-manager model

Chapter 34 built Bellwether a control program: roughly \$4,849 a year of controls standing against about \$53,122 of measured exposure, plus an investigation ladder that deliberately puts theft seventh of seven possible explanations for a variance. It is a good program. It has one structural weakness, and Chapter 34 named it and left it open:

Who audits the owners?

In a single unit, that question has an uncomfortable answer: nobody, and mostly it doesn't matter, because the owners are also the ones being protected. The owner counts the walk-in, reviews the void report, signs the checks, and reconciles the bank statement. They are simultaneously the control and the thing being controlled, and the business tolerates this because the owner's incentives point the same direction as the business's.

Growth breaks that arrangement in two distinct ways, and it is worth separating them.

First, the owner stops being able to perform the control. They cannot count two walk-ins on the same Sunday night. They cannot read four void reports with any real attention. At two units, every control the owner personally performs runs at half strength; at four, at a quarter. The owner becomes a bottleneck, and then a rumour — a name invoked in a building they have not walked through in five weeks, attached to standards nobody has read.

Second, the owner's own transactions stop being self-policing. In one unit, an owner taking product home is taking it from themselves. In a group with a partner, an investor, a bank, or an eventual buyer, owner-coded expenses, owner comps, related-party purchases, and inter-unit transfers are exactly the transactions with the least oversight and the most room for honest error. This is not an accusation; it is a design problem. Any control that exempts the person with the most authority is not a control.

Answering it concretely

Four mechanisms, none expensive, all of which should be in place before the second unit opens:

1. Segregate at the top. The partner who signs checks is not the partner who reconciles the bank statement. In a two-partner business this costs nothing but a swap of two tasks, and it converts the largest single control gap in the company into a routine.
2. Write a delegation-of-authority table. Who can approve what, up to what dollar amount, and who is the second signature above it — including the owners. A line that reads "owner-partner: any amount, no second signature" is a decision you should make deliberately, in writing, rather than by default.
3. Report the owners' unit on the same form as everyone else's. When the founding unit's numbers arrive in a different format, on a different day, with a different definition of labor cost, it is not comparable — and it is not being examined. Same form, same day, same rules.
4. Buy an outside look. At two or three units, an annual review by an outside accounting firm — bank reconciliations tested, owner-coded expenses sampled, inter-unit transfers traced — costs a few thousand dollars and is the only genuinely independent audit in the building. It is also the single most useful document to have already been producing when anyone ever asks to see your books.

🧮 Run the Numbers

How many units can one person actually carry?

Span of control is the number of people or units one supervisor can oversee before supervision becomes nominal. You will hear numbers quoted — five, seven, ten — as though there were a law. There isn't one, and anyone quoting you a hard figure is quoting folklore. What there is, is arithmetic.

Budget a district manager's usable week at 45 hours (they will work more; the rest goes to traffic, phone calls, and things that catch fire).

Fixed load, independent of unit count:

Task	Hours/week
Consolidated reporting, the weekly numbers call	4
Vendor, HR, and facilities escalations	4
Own leadership meeting, planning, administration	2
Fixed total	10

Per-unit load, per week, for a stable unit with a competent manager:

Task	Hours/unit/week
One meaningful visit, on site, including part of a service	4
One-on-one with the unit manager	1
Reviewing that unit's numbers before the visit	1
Travel	1
Per-unit total	7

$$\text{Units} = \frac{45 - 10}{7} = \frac{35}{7} = \mathbf{5 \text{ stable units}}$$

Now break the assumption, because "stable" is doing all the work in that sentence:

• A unit open under a year, or one with a manager in their first six months, needs two to three visits a week. It consumes 2–3 slots.
• A unit in trouble — a critical audit failure, a prime cost four points over, a manager who is leaving — consumes 4 slots and sometimes all of them.

So the same district manager carries five stable units, or three that are not, or one that is on fire plus two that are behaving. And the moment you add a sixth unit to a five-unit district, you have not added 20% of load; you have removed the slack that absorbed the next problem.

Span of control is not a headcount. It is a time budget divided by the variance in the units. The way you increase span is not by demanding more of the supervisor — it is by reducing variance, which is what §37.2 and §37.3 were for. Every hour of documentation and every point of audit consistency buys back supervisory capacity somewhere else in the company.

What a district manager's product actually is

The role goes by several names — district manager, area manager, director of operations, multi-unit manager — and in an independent group of three to six units it is often the founding operator wearing a different hat two days a week. Whatever it is called, its product is other managers' performance.

This produces the single most common failure in the role, and it is nearly universal among people newly promoted into it: the former best-GM who keeps running shifts. They arrive at a unit, see that the expo is drowning, step in, fix it, and leave at midnight feeling useful. They were useful — to one service, in one building — and they were absent from four others while doing it. Worse, they taught that unit's manager that when things get hard, someone senior arrives and takes the wheel, which is precisely the lesson that guarantees they will need to do it again.

The tell is simple and you can measure it: what fraction of a district manager's on-site hours are spent doing work a unit employee could have done? Above roughly a quarter, the group has bought a very expensive floating manager rather than a supervisory layer.

The exception is real and worth naming: a genuinely failing unit sometimes needs a leader to take direct control for a defined period. Fine — but say so out loud, put a date on it, and hand it back on that date. An undeclared takeover never ends.

What the layer costs, and the valley you have to cross

Overhead in a multi-unit group is lumpy. Revenue arrives one unit at a time; support functions arrive in whole people. The result is a margin curve that dips every time you add a layer and recovers as you fill it — and the first dip is deep enough that a great many groups never get past it.

FIGURE 37.4 — THE MULTI-UNIT VALLEY                            [constructed teaching example]

  Assumption: a mature unit produces $1,550,000 of revenue and a 12.0% unit-level
  operating margin ($186,000) AFTER paying a full market-rate management team and
  BEFORE any shared overhead. (That is below Bellwether's 16.8% plan margin for
  exactly the reason Chapter 35 identified: the plan's margin includes two owner-
  partners doing two salaried jobs for less than two salaried jobs cost.)

  UNITS  GROUP REVENUE   UNIT PROFIT   SHARED OVERHEAD   GROUP PROFIT   MARGIN
  ──────────────────────────────────────────────────────────────────────────────
    1     $1,550,000      $  186,000      $      0        $  186,000     12.0%
    2     $3,100,000      $  372,000      $ 55,000        $  317,000     10.2%
    3     $4,650,000      $  558,000      $190,000        $  368,000      7.9%  ◄ valley
    4     $6,200,000      $  744,000      $205,000        $  539,000      8.7%
    5     $7,750,000      $  930,000      $220,000        $  710,000      9.2%
    6     $9,300,000      $1,116,000      $310,000        $  806,000      8.7%  ◄ 2nd layer
    8    $12,400,000      $1,488,000      $355,000        $1,133,000      9.1%
   10    $15,500,000      $1,860,000      $395,000        $1,465,000      9.5%

  WHAT THE OVERHEAD BUYS
    2 units   $ 55,000  part-time bookkeeper $35k; shared systems, insurance
                        administration, travel $20k. Owners are still the district.
    3 units   $190,000  bookkeeper full-time $58k all-in; a director of operations
                        $105k all-in with vehicle; systems, legal, travel $27k.
    6 units   $310,000  the director of operations cannot carry six plus the office;
                        a training/QA manager appears at $85k all-in, plus $5k systems.

Three readings, and they are the reason this figure exists.

The valley is at three units, and it is 4.1 points deep. You buy the support layer at three because you cannot run three without it, and you do not have the volume to carry it until roughly five. Groups that stall permanently at two or three units are very often looking at exactly this arithmetic and concluding, correctly, that the next unit makes them less profitable. That is a legitimate place to stop. It is not a legitimate place to be surprised.

Margin never returns to the single-unit level. Ten units, at 9.5%, is still below one unit at 12.0% and far below Bellwether's owner-run 16.8%. Multi-unit growth does not buy margin. It buys dollars — \$1,465,000 against \$186,000, a 7.9× increase in profit on a 10× increase in revenue — and it buys something the single-unit operator can never have, which is a business that produces money while its owner is asleep. Whether that trade is worth it is a genuine question with a genuine answer either way, and Chapter 35 is where you answer it.

The dips repeat. Every new layer — the second field leader at six, a purchasing function, a real controller, a marketing hire — produces another sawtooth. Plan for them. An operator who models a straight line from three units to ten will be startled twice.

🔍 Check Your Understanding

1. A district manager has 45 usable hours, 10 of them fixed. Their district contains three stable units, one unit that opened five months ago, and one whose general manager just resigned. Using the slot arithmetic in this section, is the district staffed correctly?
2. Explain, in one sentence each, the two different ways growth breaks the "owner as the control" arrangement from Chapter 34.
3. Why does group operating margin fall between one unit and three, when nothing about the individual restaurants got worse?

(1: No. Three stable units = 3 slots, a five-month-old unit = 2–3 slots, a unit with no GM = up to 4. That is 9–10 slots against a capacity of 5. The district needs interim support or the two problem units need to be moved temporarily to another district — and if neither happens, the three stable units will stop being stable. 2: The owner can no longer physically perform the controls at more than a fraction of their former strength; and owner-level transactions — comps, expenses, related-party purchases, inter-unit transfers — cease to be self-policing once there is a partner, an investor, or a buyer with an interest in them. 3: Because support overhead arrives in whole people while revenue arrives one unit at a time; the layer that three units require is not paid for until roughly five.)

37.5 Central production and commissaries: when centralizing pays

A commissary — also called central production — is a licensed production kitchen that produces components for multiple units, or for a mobile or small-format operation, separately from the units' own kitchens. Chapter 30 priced one honestly for Bellwether in the context of trucks and ghost kitchens and found it does not pay at the scale Bellwether has or is likely to reach soon. This section generalizes that finding into a decision you can run for any item, at any scale.

At its core the commissary is a fixed-for-variable trade. You take labor and space out of every unit and concentrate it in one cheaper place, and in exchange you take on a large fixed cost that does not care how busy Tuesday was. That is the same structural bet as a fixed labor floor (Chapter 1), and it fails the same way: brilliantly at volume, brutally without it.

The four conditions

Central production pays for a given item only when all four of these hold. Missing one usually kills it.

1. The item is labor-intensive relative to its transport cost. Stocks, sauces, dressings, pickles, brines, spice blends, and butchery all carry a great deal of labor in a small, stable package. A salad does not.
2. The item survives the journey without quality loss. This is the condition that eliminates most candidates in a chef-driven concept. A demi-glace travels perfectly. A brined and portioned chicken travels perfectly. Anything with texture — fried components, dressed greens, baked goods past their first day, anything crisp — degrades on the van, and the guest pays the difference.
3. The volume is high enough to run the concentrated labor near full utilization. A production cook who is busy four hours a day is more expensive than the four hours you removed from each unit, because you now pay for the other four.
4. The route is short and the cold chain is controllable. Transport time, vehicle refrigeration, and documented temperatures at both ends are food-safety obligations, not logistics preferences.

Rank the candidates and the picture is consistent across almost every group:

🧮 Run the Numbers

What a commissary has to save before it earns a dollar.

A modest central production kitchen, constructed but realistic for a group of Bellwether-sized units:

Annual fixed cost	Amount
1,600 sq ft production space at \$15/sq ft all-in | \$24,000
Equipment lease and amortization	\$10,000
Utilities	\$7,500
Licensing, insurance, pest control, waste	\$4,500
Production lead, salaried, fully loaded	\$68,000
Delivery van, fuel, insurance, maintenance	\$9,000
Driver hours, 20/week at \$22 fully loaded | \$22,880
Total annual cost of having a commissary	\$145,880

That is the hurdle. \$145,880 a year, before it saves a nickel.

Now the savings, per unit served, per year:

Per-unit annual saving	Amount
Prep labor removed from the unit: 14 hrs/wk at \$22 loaded | \$16,016
Purchasing gain — whole primals, bulk buying on affected items	\$3,700
Waste and yield improvement from batch production	\$2,400
Less: marginal commissary production labor beyond the lead, 6 hrs/wk at \$21 | −\$6,552
Net saving per unit, per year	\$15,564

$$\text{Break-even units} = \frac{\$145{,}880}{\$15{,}564} = 9.4$$

Nine to ten units. Below that, the commissary is a subsidy you are paying for consistency, and you should be honest about calling it that rather than pretending it is a cost saving.

Three things move that number hard, and you should test all three before accepting it:

• Rent an hourly commissary instead of leasing one. Shared commercial kitchens rent by the hour in most metros. That converts \$145,880 of fixed cost into a variable one and drops the break-even to roughly the point where the labor arithmetic alone works — often three or four units. This is almost always the right first move, and almost nobody does it because owning the space feels more serious.
• Use space you already have. A unit with a large kitchen and a soft Monday can be the commissary for the group, and the incremental cost is close to zero. It creates a real internal accounting problem — that unit's labor line now carries production for units it does not own, so its P&L is no longer comparable (§37.6) — but it is solvable with a transfer entry, and it is far cheaper than a lease.
• Density. Six units in one metro with 20-minute routes is a different business from six units across three cities. Route time is the hidden variable in every commissary model.

⚠️ Where the Money Leaks

The prep labor you "saved" that never left the schedule.

Here is how a commissary loses money while its spreadsheet says it is winning. The model assumed 14 hours a week come out of each unit's prep schedule. The sauces now arrive by van. And the schedule at each unit does not change, because the prep cook is scheduled 8 to 4 on Tuesday and always has been, and now they are doing something else during those hours.

Centralization only produces savings if the hours actually come off the schedule. That means rewriting the staffing guide (Chapter 19), which means someone has to have the conversation, which is a conversation nobody enjoys. Skip it and the group pays for the same labor twice — once in the commissary and once in the unit — while congratulating itself on efficiency.

The discipline: before the first delivery, publish the new staffing guide with the reduced prep hours and a date. Measure it in week four with actual hours, not intentions. If the hours did not come out, the commissary is not saving money; stop and find out why before you sign anything longer.

And centralizing production centralizes risk. One cooling failure in a unit kitchen affects one restaurant. One cooling failure at the commissary is in every building by Thursday. The more you centralize, the more you need genuine process control — HACCP-style monitoring, batch codes, traceability from the commissary to the unit to the day — because your ability to contain a problem now depends entirely on your ability to say which units got which batch. Chapter 25 built the plan; a commissary makes it mandatory rather than advisable.

For a small-format operator, none of the above is optional. Most jurisdictions require a mobile food unit to operate from an approved commissary or base of operations, with documented servicing, water and waste handling, and food storage. That is a licensing condition, not a margin decision, and Chapter 30 covered its economics. Verify the requirement locally before you buy a truck; it varies by county and it is not negotiable.

37.6 Comparable-store sales and the multi-unit reporting package

Here is the number that ruins more multi-unit operators than any other: total sales growth.

A group with three restaurants opens a fourth. Total sales rise 33%. Everyone is delighted. Meanwhile each of the original three is down four percent on last year, the guests are leaving, and nobody will notice for eighteen months because the consolidated line goes up every single period. Growth is an extraordinarily effective anesthetic. It hides operating decay for exactly as long as you keep opening restaurants, and it stops hiding it on the day you stop.

The instrument that prevents this is comparable-store sales.

What comp sales are and why the exclusion matters

Comparable-store sales — universally shortened to comp sales or just comps — measure the change in sales at units that have been open long enough for a like-for-like comparison against the same period last year. Units that have not been open that long are excluded from both sides of the calculation until they qualify.

The exclusion is the entire point. A new restaurant's first months are not representative of anything: there is an opening spike driven by novelty and press, a subsequent decline that is normal and alarming to people who have not seen it before, and a slow ramp toward a real run rate. Including that pattern in a comparison would tell you about your construction schedule, not your business. Public restaurant companies typically require a unit to have been open somewhere between twelve and eighteen months before it enters the comp base; the specific threshold varies by company, but the convention is real and near-universal. Pick one, write it down, and never change it opportunistically — a comp base that expands when the news is good is not a measurement, it is a press release.

Note carefully what a comp is not. It is not a measure of profit. A unit can post +8% comps and lose money. It is not a measure of guest satisfaction. And it is not a same-store measure of volume unless you decompose it, which brings us to the most important arithmetic in this section.

Decomposing a comp: traffic, price, and mix

A comp sales number is a product of three things, and the three tell completely different stories:

$$\text{comp sales} \approx \text{traffic} \times \text{price} \times \text{mix}$$

• Traffic is covers. Are more people coming?
• Price is your menu price change. Did you raise prices?
• Mix is what those people ordered. Did they trade up or down, add a drink, skip dessert?

Price and mix together produce the change in average check, which is the number most operators watch. But price and mix are not the same thing at all, and pooling them hides the story.

Consider two groups both reporting +4.0% comps:

Group A is winning: more people are coming, and they are spending slightly more on their own. Group B is not winning; it is harvesting. It took a six percent price increase, lost two and a half percent of its guests as a result, and netted a number that looks identical on the consolidated line. Group B can do this for perhaps two years. Then it cannot raise prices again without accelerating the traffic loss, and the traffic that left is not coming back on request.

A comp built entirely on price is a loan against next year's traffic. That sentence is worth more than most of what you will read about multi-unit reporting.

🧾 Read the Numbers

```text FIGURE 37.5 — "Forty percent growth and a shrinking business" [constructed teaching example] THE ARTIFACT Period 7 multi-unit sales and comp report, four weeks, for a constructed three-unit casual full-service group. Comp base requires 12 months open at period start. THE CONTEXT Unit 1 open six years. Unit 2 open three years. Unit 3 opened eleven months ago and is NOT comp-eligible. The group took a 3.0% menu price increase in Period 3.

               SALES
                                 THIS YEAR    LAST YEAR      CHANGE
               ─────────────────────────────────────────────────────
               Unit 1             $198,400     $205,600       -3.5%
               Unit 2             $186,200     $174,900       +6.5%
               ─────────────────────────────────────────────────────
               COMP BASE          $384,600     $380,500       +1.1%
               Unit 3 (new)       $151,300          n/a          n/a
               ─────────────────────────────────────────────────────
               TOTAL GROUP        $535,900     $380,500      +40.8%

               THE DECOMPOSITION
                                 COVERS TY   COVERS LY   TRAFFIC   CHECK
               ──────────────────────────────────────────────────────────
               Unit 1                4,120       4,410     -6.6%   +3.3%
                                                             ($48.16 vs $46.62)
               Unit 2                3,930       3,800     +3.4%   +2.9%
                                                             ($47.38 vs $46.03)
               ──────────────────────────────────────────────────────────
               COMP BASE             8,050       8,210     -1.9%   +3.1%
                                                             ($47.78 vs $46.35)

               COMP +1.1%  =  traffic -1.9%  ×  price +3.0%  ×  mix +0.1%

WHAT IT SHOWS Two numbers on one page that point in opposite directions. Total sales are up 40.8%; comparable sales are up 1.1%; comparable TRAFFIC is down 1.9%. Every dollar of comp growth and then some is the Period 3 price increase — mix contributed essentially nothing, meaning guests did not trade up, they simply paid more for the same order. Unit 1 is the problem: down 290 covers on the period, a 6.6% traffic loss, partly masked by a $1.54 higher check. Unit 2 is genuinely growing on both traffic and check and is the only unit in the group that is actually winning. WHAT IT DOESN'T It does not say WHERE Unit 1's 290 covers went. Four candidates, and they require four different responses: Unit 3 opened eleven months ago and may be cannibalizing; a competitor may have opened; the price increase may have hit Unit 1's guest harder than Unit 2's; or Unit 1 has an operating problem — a manager who left, a service decay, a review trend. The report cannot distinguish them. It also shows no profit. Unit 3 could be the highest-volume-per- seat unit in the group and still be losing money at month eleven, and this page would look identical. And one period is four weeks of a fifty-two week year; treat a single period as a signal to investigate, never as a trend. THE DECISION One question first, because it determines everything else: did Unit 3 take Unit 1's guests? Compare Unit 1's traffic by day of week and daypart against Unit 3's opening date; check reservation or loyalty overlap if the systems can see it (Chapter 26); look at drive-time between the two sites. If it is cannibalization, the group's real comp is better than it looks but the site-selection model is broken and must be fixed before Unit 4 is signed. If it is not cannibalization, Unit 1 has an operating problem, and the last two audit scores and the last ninety days of review text are where to look. Either way: no further price increase this year until comparable traffic stops falling. THE LESSON Total sales measure your construction schedule. Comparable sales measure your business. Comparable traffic measures whether people still want it. Report all three on the same page, every period, forever — and put traffic first, because it is the only one of the three that cannot be manufactured. ```

Consolidated numbers hide unit problems; unit numbers hide group problems

The other half of the reporting package is the profit view, and it has to be presented both ways at once. Same group, same period:

(Unit controllable profit is unit-level profit before shared overhead, corporate general and administrative cost, depreciation, interest, and taxes — what a multi-unit operator usually calls restaurant-level profit. It is the number a unit manager can actually be held to, which is exactly why it is drawn where it is.)

Read the group column alone and you see a 61.3% prime cost: a point over target, mildly concerning, the kind of thing you would put on a list. Read across and the picture is entirely different. One unit is exactly on plan at 60.0%. One is excellent at 59.0%. One is at 66.0%, six points out, and dragging the consolidated number 1.3 points on its own. The group's mild problem is one unit's serious one.

And the reverse trap is just as common: three units all at a healthy 59% prime, a group that is losing money because shared overhead was never allocated and nobody looked at the consolidated statement at all. You need both views on the same page, every period, or you will see exactly one of the two problems you have.

Whether Unit 3's 66.0% is a disaster or a normal month-eleven ramp is not answerable from this table — which is why the reporting package includes months-open next to every unit, always.

The comparability problem

Everything above depends on a condition that is quietly very hard: the units have to be counted the same way. A single-unit flash report only has to be consistent with itself; if you always count the walk-in on Sunday night and always exclude the salaried sous from hourly labor, your trend is honest even if your definitions are eccentric. A multi-unit report has no such luxury. If Unit 1 counts inventory Sunday night and Unit 2 counts Monday morning, if Unit 1 puts the salaried sous in kitchen labor and Unit 2 puts them in management, if Unit 1's POS maps non-alcoholic beverages to food and Unit 2's maps them to beverage — then the variance you are looking at is accounting, not operations, and you will spend a quarter investigating a difference that does not exist.

Six things must be identical before comparison means anything:

1. The count date and time. Same day, same point in the week, every unit, every period.
2. The item list and units of measure. One master inventory list. A case is a case everywhere.
3. The POS category map. Same categories, same items in them, maintained centrally, and never edited at the unit.
4. The labor classification. Which roles are hourly, which are management, and where salaried supervisors sit — decided once, in writing.
5. The period calendar. Four-week periods or calendar months, chosen deliberately. Four-week periods are common in multi-unit operations precisely because they contain the same number of Fridays and Saturdays every time; calendar months do not, and a month with five Saturdays will look like a hero.
6. The accrual treatment. Invoices received but unpaid, and prepaid items, handled the same way everywhere.

None of this is interesting and all of it is load-bearing. It is also the single cheapest thing to fix before the second unit opens and one of the most expensive to fix afterward, because by then two buildings have habits.

The package and its cadence

The weekly flash report is the spine, and it is already built — Chapter 31 constructed it for a single unit. The multi-unit version adds exactly three things: a per-unit column, a consolidated column, and a comparability footnote naming any definition that differs and why. That third one is where honesty lives; a group that cannot produce it does not have comparable numbers and should say so out loud rather than presenting a variance it cannot explain.

37.7 Management by exception: dashboards, thresholds, and the visit that matters

At one unit, you look at everything, because everything is in front of you. At five units, "look at everything" means five flash reports, five schedules, five void reports, five inventory variances, five sets of review text, and five managers who need an hour a week. Something has to give, and what normally gives is attention: the reports are received, skimmed, and filed, and nobody notices anything until it appears in a period close.

Management by exception is the discipline that fixes this. Performance inside pre-agreed thresholds requires no attention and generates no conversation. Only deviations outside them trigger review. The scarce resource — supervisory attention — flows to where the variance is, automatically, without anyone having to decide where to look.

A threshold needs three things

Most "dashboards" are decorations because they contain metrics without thresholds, or thresholds without consequences. A working exception threshold has three parts:

1. A metric with an unambiguous definition. "Labor cost" is not a metric until you have said whether it includes payroll taxes, whether it includes salaried management, and whether it is measured against net or gross sales.
2. A band, with a number. Not "high labor" — "above 32.5% of net sales for the week."
3. A named consequence. What happens when it trips: who is called, by whom, within what time, and what document results. Without the consequence it is not a control, it is a color.

Setting the band is a judgment call with a useful rule of thumb behind it: set the band so that roughly one unit in five trips it in a normal week. Too tight and every cell is red, which produces alert fatigue and teaches everyone to ignore the page. Too loose and you find out at the period close, which is what you built the dashboard to avoid. If everything is red, the threshold is wrong before the units are.

FIGURE 37.6 — THE ONE-PAGE GROUP EXCEPTION DASHBOARD, WEEK 29   [constructed teaching example]

  LEGEND    ·  inside threshold — no action, no conversation
            ◄  exception — requires a named owner and a date by Monday noon
            ✱  critical — escalates to the owners same day, regardless of anything else

  METRIC                       THRESHOLD     UNIT 1     UNIT 2     UNIT 3
  ────────────────────────────────────────────────────────────────────────────
  Sales vs. forecast            ± 5.0%      -1.2%  ·   +3.8%  ·  -11.4%  ◄
  Comp sales vs. last year      ≥  0.0%     -3.1%  ◄   +6.9%  ·      n/a
  Prime cost                    ≤ 61.0%     60.3%  ·   59.1%  ·   66.4%  ◄
     food cost                  ≤ 30.5%     28.4%  ·   27.1%  ·   30.2%  ·
     labor cost                 ≤ 32.5%     31.9%  ·   32.0%  ·   36.2%  ◄
  Ticket time, 90th percentile  ≤ 19 min     17.2  ·    16.4  ·     24.6  ◄
  Comps + voids, % of sales     ≤  1.2%      0.9%  ·    0.7%  ·    2.4%  ◄
  Cash over/short, week         ± $40         -$12  ·     +$8  ·     -$96  ◄
  Overtime hours                ≤ 12 hrs        6  ·       4  ·       31  ◄
  Turnover, trailing 90 days    ≤ 65%         48%  ·     41%  ·     104%  ◄
  Food-safety criticals, open   0 (any = ✱)     0  ·       0  ·        1  ✱
  Open audit items > 30 days    0                0  ·       2  ◄        5  ◄
  ────────────────────────────────────────────────────────────────────────────
  EXCEPTIONS      Unit 1: 1        Unit 2: 1        Unit 3: 9 + 1 CRITICAL

Now read it the way a district manager should, which takes about forty seconds.

Unit 1 has one exception: comp sales down 3.1%. That is a conversation, not an emergency — and it is the same conversation Figure 37.5 already started. Unit 2 has one: two audit items open past thirty days, which is a discipline problem in a well-run building and gets a sentence in the Monday call.

Unit 3 has nine exceptions and a critical. Nine exceptions on one unit is not nine problems. It is one problem with nine symptoms, and the shape of it is legible from the page. Food cost is fine at 30.2%. Labor is 3.7 points over, overtime is 31 hours, turnover is 104% on a trailing ninety days, and ticket times at the ninetieth percentile are 24.6 minutes against a 19-minute standard. That is a kitchen that is short-staffed, covering with overtime, running slow because it is short, and losing people because it is running slow — which makes it shorter. This is Chapter 21's warning arriving as a table. The reliable people are leaving first, exactly as predicted, and every week that passes makes the next hire harder.

Notice also that the critical does not wait its turn. One open food-safety critical goes to the owners the day it appears, whatever else is on the page, and it does not queue behind the labor conversation.

The escalation ladder

Thresholds without a routing table produce a dashboard where everything goes to the owners, which is the bottleneck from §37.4 wearing a nicer interface.

FIGURE 37.7 — WHO SEES WHAT                                  [constructed teaching example]

  LEVEL 1  ── UNIT MANAGER ──────────────────────────────────────────────
     Sees:      everything about their own unit, daily.
     Owns:      every exception at their unit, with a written action and a date.
     Escalates: anything they cannot close within 14 days.
                       │
  LEVEL 2  ── DISTRICT MANAGER ──────────────────────────────────────────
     Sees:      exceptions only, across all units, weekly. Not the full reports.
     Owns:      any exception open past 14 days; any pattern across two or more units.
     Escalates: anything open past 30 days; anything needing capital.
                       │
  LEVEL 3  ── OWNERS ────────────────────────────────────────────────────
     Sees:      exceptions open past 30 days; the period package; the quarterly review.
     Owns:      standards, capital, people decisions, and the things nobody else can fix.

  ═══ THE BYPASS ═══  These go to Level 3 the same day, from anyone, always:
        food-safety critical · suspected theft or cash irregularity (Ch. 34)
        injury · allergen incident · alcohol incident · harassment complaint
        law-enforcement or regulator contact · anything a guest posted publicly
        and anything at all that a manager is afraid to put in writing

That last line is deliberate and it is not sentimental. The exceptions that destroy companies are not the ones on the dashboard; they are the ones somebody decided not to report. A bypass that explicitly includes "anything you are afraid to write down" — and a leadership team that has never once punished its use — is the only mechanism that reaches them.

⚠️ Where the Money Leaks

Every metric you manage to becomes a metric someone manages.

This is the oldest problem in measurement, and the economist Charles Goodhart's observation is the usual shorthand for it: when a measure becomes a target, it stops being a good measure. In a restaurant group it is not theoretical. Every line on Figure 37.6 has a known workaround, and if you do not name them in advance you will be surprised by all of them:

Metric	How it gets managed	What to watch instead
Ticket times	fire tickets late so the clock starts later	table-to-entrée time from the reservation system; guest complaints about waits
Comps and voids	stop comping tables that should be comped	review text, and the ratio of manager visits to comps
Cash over/short	do not ring the sale at all	inventory variance and pour cost (Ch. 34) — a perfect drawer with a bad variance is a signal
Food cost %	buy nothing in the last week of the period	purchases-to-usage gap; on-hand inventory value trend
Labor cost %	ask people to clock out and keep working	this one is wage theft and it is illegal; audit punch edits, and treat any manager who does it as a firing matter (Ch. 20)
Turnover	reclassify separations, or never terminate anyone	average tenure, and open shifts covered by overtime
Audit score	clean for the visit	unannounced windows, service-hour visits, self-audit vs. leadership-audit gap

Two structural defenses. First, never build a bonus on a single metric. A bonus on food cost alone buys you a bad walk-in and a starved menu; a bonus on prime cost with an audit-score and turnover gate is much harder to game because the workarounds for one show up in another. Second, pair every efficiency metric with a quality metric on the same page, so that the cost of gaming is visible in the next row down. Figure 37.6 does this on purpose: labor sits directly above ticket time and turnover.

The visit that matters

The dashboard tells you where to look. It never tells you what you will find, and there is no substitute for standing in the building. But most multi-unit visits are worthless, and the way they are worthless is remarkably consistent.

👨‍🍳 On the Line

The visit that is worth the drive, and the one that isn't.

The worthless visit. Announced. Ten a.m. on a Wednesday. The district manager walks the walk-in, which was scrubbed last night, notes that the labels look good, has a coffee with the general manager, discusses the numbers that were already in the report, admires the new patio furniture, and leaves at 11:30. Nothing was learned. Everyone was polite. Four hours and a tank of fuel.

The visit that matters. Six things, and they are not expensive:

1. Arrive unannounced within a published window. "Some time in the next three weeks" is fair and unstageable. Surprise is not a trick; it is the only way to see the restaurant that exists.
2. Be there for the transition. Four-thirty to six-thirty. The shift change is when every system in the building either holds or doesn't: the prep handoff, the pre-shift, the first tickets, the moment the reservation book stops being theoretical. A restaurant at ten a.m. tells you about cleaning. A restaurant at six-fifteen tells you about management.
3. Start with the last visit's list. Before anything else, walk the previous visit's open items and close them or explain them. A visit that does not begin here teaches the unit, permanently, that the list does not matter — and after that no list ever will.
4. Talk to three hourly employees without the manager present. Not an interrogation. Three questions: what is going well, what is stupid, and what would you fix. The third one gets real answers surprisingly often, and the answers are usually about a system, not a person.
5. Eat in the dining room, as a guest, at least once a quarter. Pay. Wait for a table. Order what a guest would order. Nothing on any report tells you what this tells you.
6. Leave with three written items, each with a role and a date. Three. Not fourteen. A list of fourteen is a list of zero, because the manager cannot do fourteen things and will do none of them rather than choose.

And use the one-question test. Pick a standard — the salsa verde, the greeting window, the cooling procedure — and ask a line cook or a server what it is. Three possible answers, three different diseases:

• They answer, and it matches the manual. The system is real. Move on.
• They answer confidently, and it is different from the manual. You have drift, and probably a local decision that someone made for a good reason and never routed back. Find out why before you correct it; sometimes their version is better and your document is wrong.
• They say "I'd have to ask the chef." You have a documentation problem, and it does not matter how good your manual is, because it is not reaching the station.

Three answers, three treatments. It takes ninety seconds and it is the highest-yield ninety seconds in multi-unit management.

37.8 Scaling culture: the part that does not travel by manual

Everything in this chapter so far can be written down. This section cannot, and pretending otherwise is the characteristic error of growing restaurant groups.

Scaling culture is the deliberate transfer of a restaurant's operating norms — what gets praised, what gets tolerated, how people are treated when it is inconvenient — into units the founders do not work in. Chapter 21 made the case that culture is a line item rather than a poster, and it issued the warning that governs this entire chapter: culture does not travel by manual, and when it fails, the reliable people leave first.

That last clause deserves unpacking because it is the mechanism, not the moral. Turnover in a deteriorating culture is not random. The people who leave first are the ones with options, the ones who notice soonest, and — critically — the ones who had been holding the standard voluntarily, without being asked, because they cared. Their departure is therefore doubly expensive: you lose a good employee, and you lose the unwritten enforcement they were providing for free. Chapter 17 taught you to cost a single departure. In a second unit, the cost is higher than the formula says, because the person leaving was also part of the system.

What actually travels, in order of reliability

1. People. This is the mechanism. The single most reliable way to put your culture in a new building is to move a human being who has it in their body into that building. A general manager, a chef de cuisine, and ideally a third or more of the key hourly positions — a bartender, a lead server, a sauté cook — transferred from the original unit and backfilled at home.

This is why Chapter 35's "no management bench" finding is not an HR technicality. The bench is the culture strategy. A group that opens unit two with a leadership team hired entirely from outside has opened a different restaurant with the same name, and it will discover this in about six months, in the reviews.

The backfill is easier than it sounds, and this is the part operators miss: the original unit is the easy one to hire into, because it has systems, a reputation, and people who can train. The new unit is the hard one. So you send the experienced people to the hard building and hire the new people into the easy one — which is the exact opposite of what most operators do, because sending your best people away feels like weakening the thing that works.

2. What leadership walks past. Culture is the average of what a leader tolerates without comment. In one unit the founder sets that continuously by being present. In two units the standard at each building is set by whoever is present there — which means the unit's manager is the culture, whatever your manual says. You do not get to choose this. You only get to choose who that person is, and to accept that you have delegated your culture to them the day you hand over the keys.

3. Structures that encode values in numbers. These do travel by document, and they matter more than any values statement:

• Whether schedules are posted two weeks out or on Thursday for Friday.
• Whether time-off requests are honored or theoretically honored.
• Whether the tip pool is fair, transparent, and lawful in that jurisdiction (Chapter 20).
• Whether managers get two consecutive days off, and whether that survives a busy December.
• Whether the complaint procedure has ever produced an outcome anyone can name.
• Whether people get paid correctly, on time, every time, without having to ask.

A group can copy every one of these across units precisely because they are policies. They are also, collectively, a better predictor of a unit's turnover than anything you would find in a culture deck.

4. Rituals. Pre-shift, family meal, the way a mistake is handled at the pass, what happens on someone's last night. Cheap to replicate, durable once established — but they only take root in a new unit if the founders personally model them for the first several months. A ritual introduced by memorandum is a chore.

What does not travel, ever

The founder's palate. Their threshold for "good enough." Their tolerance for a certain kind of difficult, talented person. Their willingness to work a double when someone calls out at four p.m. on a Saturday — which the whole staff has watched them do, and which is doing more work in your culture than any policy you have written.

None of that transfers. It can only be replaced, by a written standard plus a person who believes in it. That is a worse instrument than your presence, and it is the only instrument that scales.

Measuring the thing everyone says cannot be measured

Culture is a line item, so measure it. Six proxies, all of which you already have or can get cheaply:

That last one is the most interesting and almost nobody tracks it. A group where staff move between units to cover shifts, pick up hours, and train has a shared culture almost by definition — people carry norms in their hands. A group where nobody has ever worked at a second location has three separate cultures and a common logo, and will discover it the first time a unit gets into trouble and there is nobody who can be sent.

The founder's job changes, and you may not want the new one

One honest thing to close on, because Chapter 35 asked whether Bellwether should grow and this is the half of the answer that is not financial.

Growth does not scale your job. It replaces it. The chef who opens a second restaurant stops cooking — not partly, not eventually, but within about a year — and starts doing hiring, reporting, standards, and other people's problems. The front-of-house partner stops running a room, which is the thing they are excellent at and love, and starts running a schedule of schedules and a set of P&Ls. Both of them spend materially more time in a car and materially less time in a restaurant.

A great many operators discover at unit two that they have promoted themselves into a job they do not want, and the industry treats this as a failure of nerve. It isn't. The correct answer is sometimes one excellent restaurant, run well, for twenty years, by people who are still doing the work they are good at. That is a complete and honorable career, it produces a real living, and it is chosen far less often than it should be — usually because growth is the only story the industry knows how to praise. Chapter 40 comes back to this. Decide it deliberately, with the arithmetic in §37.4 and this paragraph both on the table.

🔍 Check Your Understanding

1. A group opens a second unit and staffs its entire leadership team with experienced hires from outside. What specifically has failed to transfer, and roughly when will it become visible?
2. Why is a comp of +4.0% built on +6.0% price and −2.5% traffic a worse result than a comp of +4.0% built on +3.2% traffic and +0.5% price?
3. A unit's dashboard shows food cost inside threshold, labor 3.7 points over, overtime at 31 hours, and trailing turnover at 104%. Name the single underlying problem and say which two chapters you would go to first.

(1: The culture — specifically the unwritten standards and the norms that were transmitted by the founders' presence and by experienced staff. It typically becomes visible in six to twelve months, in turnover and in review text, well after the opening period masks it. 2: Because price is a lever you can only pull a limited number of times and the guests it drives away do not return on request; the traffic-led comp reflects demand, which compounds, while the price-led comp borrows from next year. 3: One problem — the kitchen is understaffed and losing people faster than it can hire, so it covers with overtime, runs slow, and loses more people. Chapter 21 for the retention mechanism and Chapter 19 for the staffing guide that should have caught it.)

🍽️ The Business Plan

Checkpoint 37 of 40 — multi-unit readiness: the manual you already wrote.

Chapter 35 tested whether Bellwether should open a second location and returned not yet. Chapter 36 tested whether the concept could be franchised and returned not yet, for the related reason that you cannot sell a system that has not been written down. Both verdicts stand. This checkpoint does not argue with them. It asks the more useful question: if the answer is systems, how much of the system does this plan already have?

The answer is more than you would guess, and the reason is worth stating plainly. Documentation is a by-product of measurement. You could not cost the Hearth Chicken to \$8.52 (Chapter 11) without first standardizing the recipe (Chapter 10). You could not set par levels (Chapter 13) without writing product specifications. You could not build a staffing guide (Chapter 19) without defining every position and what it does. You could not write a food-safety plan (Chapter 25) without writing procedures. Thirty-six chapters of financial rigor have produced, as a side effect, most of an operations manual — by people who thought they were doing arithmetic.

Here is the audit, section by section, against the fifteen-section structure in §37.2.

The tally: eight sections essentially written, five partial, two missing. Call it roughly two-thirds to three-quarters of an operations manual, produced by a business plan whose author was trying to satisfy a lender rather than run a second building. That is a genuinely useful discovery, and it reframes what the multi-unit conversation is about. Bellwether's documentation problem is not that the manual does not exist. It is that the manual exists in fifteen pieces, in fourteen chapters, in three formats, under no index, with no owner and no review date, and nobody has ever called it a manual.

Assembling it is a weekend of work, and it is the highest-return weekend available to this business.

What is genuinely missing, and why it is always the same list

Look at what the financial work did not produce, because it is not random.

Service standards (§7). The largest gap by far. Nothing in a business plan forces anyone to write down what a greeting is, how long a table waits, or what happens when a dish goes back. Bellwether's service is very good and entirely undocumented, because the front-of-house partner is the standard. That is precisely the arrangement §37.1 says does not survive a second building. This is the single highest-value document the plan does not have.

Training (§9). There are job descriptions, because the staffing guide required them. There is no training path, no checklist, and no certification-of-competence step. Training is what converts a manual into behavior; without it, documentation is a shelf.

Facilities and preventive maintenance (§14). Nothing in the financial plan required it, so nobody wrote it, and the hearth — the one piece of equipment the whole concept depends on, and the one that caps the kitchen at roughly 132 covers — has no written maintenance schedule attached to it.

The management layer itself. There is no written job description for a general manager who is not an owner, because there is no such person. That is Chapter 35's "no management bench" restated as a documentation gap, and it is the same finding twice.

The audit. Chapter 34 built a control program: about \$4,849 a year of controls against roughly \$53,122 of measured exposure, plus a seven-step investigation ladder that puts theft last for good reasons. It is an excellent program and it is not a document. It lives in two owners' habits, which means it has a span of control of exactly one building. Converting it into a scored, dated, signed quarterly audit with critical items — using §37.3's structure — is the single highest-leverage conversion available to this plan right now, and it costs nothing but the writing.

The readiness verdict

Three things to do in the next twelve months, none of which require a second unit

1. Write the service standards and the training path. Cost: the front-of-house partner's Monday mornings for a quarter. Use §37.2's write-it-down test so it comes out as one binder of obligations rather than four hundred pages of script. This is the document that makes everything else portable.

2. Convert the Chapter 34 control program into a scored quarterly audit — and audit the owners' unit first. Weight the sections. Define six to ten critical items. Set the owner-partners' segregation of duties (one signs, the other reconciles) and write the delegation-of-authority table, with the owners on it. Cost: a weekend and a conversation neither partner will enjoy.

3. Hire or promote one non-owner general manager, and then actually leave the building. This is the test, and it is the only one that produces real evidence. Not whether the numbers hold while you are there — whether they hold for eight consecutive weeks with both owners absent two services a week, measured on the same flash report, against the same prime-cost target of 60.0%, with the audit run at week four by someone else. One salary buys the only honest answer to the second-location question that exists, and it buys it inside one building, at a fraction of the risk of buying it inside two.

What this checkpoint does not settle. It does not settle whether Bellwether should ever have a second location; Chapter 35 answered not yet and nothing here changes that. It does not settle whether the concept travels — the Rivermill District's eight-year gentrification is a market condition, not a portable asset, and a second neighborhood is a second market analysis. And it does not settle the hardest question, which is whether two owners who are excellent at running one restaurant want the job that comes after.

Open questions carried forward:

1. Can the room hold 60.0% prime cost and its service standard for eight consecutive weeks with a non-owner general manager and the owners out two services a week? (the test above; revisited in Ch. 40)
2. Which of the two owner-partners' jobs is the one that must be replaced first — and can it be? (Ch. 40)
3. If the answer to growth is "not yet" for several years, what is the plan for the owners' own sustainability in a business that currently requires 113 hours a week from two people? (Ch. 21, Ch. 39, Ch. 40)
4. Assembled into one indexed document with owners and review dates, does the plan read as a business somebody else could operate? (Ch. 40, and Appendix C)

Conclusion

The thing that makes a great single-unit operator great — presence, taste, a fast eye, and the willingness to fix it themselves — is exactly the thing that does not scale, because all of it lives in a body that can only be in one building at a time. Multi-unit management is the discipline of converting those instincts into artifacts: a specification, a procedure, a standard, an audit, a threshold, a report, and a person whose job is other people's performance.

The conversion is not free and it is not neutral. A written standard is a worse instrument than your attention — less nuanced, slower to adapt, blind to the thing you would have noticed in half a second. It has exactly one advantage, and the advantage is decisive: it works in a building you are not standing in. Everything in this chapter is an attempt to make that trade as small as possible: the write-it-down test so you only document what varies, the critical-item audit so the score does not distract from the risk, the comp decomposition so growth cannot hide decay, and management by exception so that scarce attention lands where the variance is.

Two arithmetic facts should stay with you. Span of control is a time budget divided by the variance in the units — which means every hour spent reducing variance buys supervisory capacity somewhere else. And multi-unit growth is a dollars strategy, not a margin strategy: the group margin never returns to what a well-run single unit produces, and the compensation for that is scale and a business that runs without you. Whether that is a good trade is a real question, and "one excellent restaurant for twenty years" is a legitimate answer to it.

For Bellwether, the finding is the useful one. The plan has quietly written two-thirds of an operations manual while thinking it was doing financial work — and the pieces it is missing are the same pieces every independent is missing, because no lender ever asked for them. Service standards. A training path. A maintenance schedule. A written audit. A general manager who is not an owner. None of those require a second building, and all of them make the first one better.

Chapter 38 turns to a different kind of system, and one where the arithmetic runs the same direction as the ethics more often than people expect: sustainability. Food waste is simultaneously the largest environmental impact a restaurant has and one of the largest recoverable costs on its P&L, which means the measures that survive in a four-point-margin business are the ones that also pay. We are going to count the dumpster.

Key Terms

Multi-unit leadership — the practice of producing restaurant performance through other people in buildings you are not standing in; its output is standards, instructions, feedback, and hiring decisions rather than services worked. (Ch. 37)

Operations manual — the written record of a restaurant's standards, specifications, procedures, and already-made decisions, organized so a competent stranger can find an answer without asking the owner. A decision archive, not a rulebook. (Ch. 37)

Commissary (also central production) — a licensed production kitchen that produces components for multiple units or for a mobile operation, separate from the units' own kitchens; a fixed-for-variable cost trade that pays only at volume, and a licensing requirement for most mobile operators. (Ch. 37)

Span of control — the number of units or people one supervisor can oversee before supervision becomes nominal. Not a fixed number: a usable-hours budget, less the fixed load of the job, divided by the per-unit load — which rises sharply for new, unstable, or troubled units. (Ch. 37)

Comparable-store sales (also comp sales, comps) — the change in sales at units open long enough for a like-for-like year-over-year comparison, typically twelve to eighteen months, excluding newer units from both sides. Separates operating performance from the effect of opening restaurants. (Ch. 37)

Brand consistency — a guest receiving the same experience at any unit on any day. Operationally a statement about variance rather than average, because no guest ever eats at the average. (Ch. 37)

Management by exception — a control discipline in which performance inside pre-agreed thresholds requires no attention and only deviations trigger review, so supervisory attention flows automatically to where the variance is. Requires a defined metric, a numeric band, and a named consequence. (Ch. 37)

Scaling culture — the deliberate transfer of a restaurant's operating norms into units the founders do not work in; achieved primarily by moving experienced people, secondarily by policies that encode values in numbers, and never by a manual alone. (Ch. 37)

Spaced Review

1. Without looking back: what are the two questions in the write-it-down test, and what does each answer produce — a specification or a procedure?
2. A group's Period 4 report shows total sales +22%, comp sales +0.4%, and comp traffic −3.1%. The group took a 4% price increase in Period 1. Write the two sentences you would put at the top of the report for the owners.
3. From Chapter 31 and Chapter 1: a unit's flash report shows COGS of \$26,800 and labor of \$31,200 on weekly sales of \$96,000. Compute prime cost. If the group's exception threshold is 61.0%, does this trip it, and what would you want to see next to that number before deciding anything?
4. From Chapter 34: the control program costs about \$4,849 a year and stands against roughly \$53,122 of exposure. Explain what changes about both figures at three units, and why the ratio is not the thing that changes most.
5. The recurring question: an operator proposes moving all stock, sauce, and pickle production to a rented hourly commissary kitchen for a three-unit group. Does this decision move prime cost, and in which direction? What would you have to measure in week four to find out whether it actually worked?