Multi-location consistency: the six mistakes that break a restaurant group and the method that holds it

Open the consolidated P&L unit by unit before you touch anything

Step one of this guide is not writing a manual: it is splitting your consolidated P&L into as many columns as you have locations, because the average is the best hiding place a restaurant group owns. One four-unit group arrived with 29.4% average food cost, 11 points of EBITDA and 18% year-over-year growth, and that picture looked reasonable until we opened it by unit: 26.8% at the founding store, 28.1% at the second, 31.9% at the third and 34.6% at the fourth, already above the 32% ceiling that we at Masterestaurant treat as a tolerable maximum rather than a target. All four bought from the same distributor on the same negotiated price list, so the drift was not coming from purchasing. DELIVERABLE: one table with food cost, prime cost and average check per location for the same period, plus the gap in points between best and worst.

Open the consolidated P&L unit by unit before you touch anything — in practice

Past 3 points, you do not run four locations, you run four businesses. Consistency is won BEFORE the second opening, and that is a question of sequence, not of effort. When a group documents after opening, three versions of the same dish already coexist and the corporate chef stops writing procedures and starts arbitrating disputes between kitchens that have done it their own way for months, each with arguments and with regulars who order it that way; that arbitration burns four to seven times the hours it would have taken to write the spec when a single version existed and the debate was worth one afternoon. Flip it around: if your first location closed tomorrow and only the manual survived, could anyone reopen it identically? If not, location two was born degraded. DELIVERABLE: the date your manual closes, set before you sign the second lease, never after. Domino's projects 1,100 net stores a year through 2028 (Quartr, 2025), and that cadence does not survive improvisation.

Write technical specs with grams, not manuals with intentions

A manual that says «sweat the onion until translucent» is not a replicable manual, it is an intention; one that says 180 grams of onion in 4-millimeter brunoise, oil at 140 degrees, six minutes, is. The difference shows up as money. At that group's fourth location, the risotto went out with 40 grams more parmesan than at the founding store because the cook learned it by watching another cook: forty grams per plate, one hundred and twenty plates a week, fifty-two weeks, and up come 249 kilos of parmesan a year that nobody decided to buy. Nobody stole anything, nobody was careless; the system left a decision open and somebody closed it for you. DELIVERABLE: a per-dish technical spec with grams, waste, yield, temperature, time and calculated unit food cost, signed off. Verify it by weighing three random plates per location: acceptable drift is ±5% of the spec weight.

Run the territorial feasibility of location two with numbers, not hunches

Before signing the second lease you must answer whether that territory can carry your cost model, and the answer is arithmetic. Take rent as a share of a conservative sales projection, add the prime cost your technical spec already proved at location one, and compare it against the 55 to 65% of sales that the National Restaurant Association reports as prime cost in multi-unit operations. If reaching break-even demands 22% more traffic than location one hit in its best month, that site is not expensive: it sits outside your model. Remember that payroll, rent and utilities are NOT loaded onto the plate, they live at break-even. DELIVERABLE: a one-page document with monthly break-even sales, required daily traffic, required average check and the month you reach it. Without that page, you are not expanding, you are gambling. Consistency does not hold up through annual conventions or culture speeches, it holds up through weekly measurement per unit and a threshold that forces somebody to move.

Install weekly measurement per unit with a threshold that forces action

Set the corridor: food cost per location within ±1.5 points of the theoretical spec, prime cost inside the band you defined, and weight drift under 5%. When a unit breaks the corridor two weeks running, nobody sends an email: the corporate chef goes, weighs, reviews waste and pulls that unit back to spec in under ten days. McDonald's runs roughly 95% of its restaurants through franchisees (McDonald's, 2025) and holds the standard precisely because measurement does not depend on any operator's goodwill. DELIVERABLE: a weekly board with one row per location, three indicators and a traffic light. If nobody reads it Monday before ten, it does not exist. Four failures explain almost every degraded replica that turns up in a group audit. First: canonizing the dish from whichever location sells most instead of the one the spec supports, which is rewarding noise. Second: letting each head chef negotiate ingredient substitutions without recalculating food cost, then discovering at quarter close that margin walked out on a different cheese brand.

The four errors that break the replica, and how to cut them at the root

Third: training by observation —the cook learns by watching— which is exactly the mechanism that planted the 249 kilos of parmesan in the case above. Fourth: measuring the consolidated average and celebrating it. Cut at the root: one versioned spec with a date, a closed approved-supplier list, training with a weighed practical test, and a ban on reading the consolidated number without opening it by unit. DELIVERABLE: a signed record per location where its head chef declares receipt of the current spec and a passed test. You will know the replica landed when seven things are true at once, and not before. One: a gram-level technical spec exists for 100% of the live menu, not for the star dishes. Two: real food cost per location sits within ±1.5 points of theoretical for four consecutive weeks. Three: three plates weighed at random in each location fall inside ±5%.

Closing checklist: how to know location two really is a copy

Four: prime cost sits inside the 55–65% band (National Restaurant Association). Five: the food cost gap between your best and worst location drops below 3 points. Six: the weekly board was reviewed all four Mondays of the month, with an owner assigned per deviation. Seven: a new cook produces the dish to spec using only the document, asking nobody. Start today with the seventh: hand the manual to someone who was not there, and measure what comes out. The first difference is sequence, not effort: the group that breaks documents AFTER opening, once three versions of the same dish already coexist and someone has to pick which one becomes canon; the group that scales documents earlier, while a single version still exists and the argument is cheap. Documenting late costs four to seven times more in corporate chef hours, because by then you are not writing a procedure, you are arbitrating a dispute between kitchens that have done it differently for months and each has its case.

Where multi-location consistency actually breaks?

The second is granularity. A manual that says «sweat the onion until translucent» is not a replicable operating manual, it is an intention;

a manual that says «180 grams of onion in 4 mm brunoise, oil at 140 degrees, six minutes, no color» qualifies, because it removes the decision from the operator. Multi-location consistency lives exactly there, in how many decisions the night-shift cook at location five is still making on a Tuesday in February while you sit nine hundred kilometers away. The third one hits the balance sheet and almost nobody sees it coming: expansion CapEx in a group without a frozen layout climbs because each kitchen gets redesigned by whichever architect is available, different equipment gets bought, that different equipment produces different times, and those different times produce different plates. The chain starts in a construction decision and ends on the guest's palate eleven months later, when nobody remembers that location three has a different oven brand.

Where multi-location consistency actually breaks — in practice?

Fourth comes instrumentation. Without blind weighing and without unit economics opened by unit, you do not have consistency: you have a belief about your consistency, which is a different thing and usually an optimistic one.

The consolidated average is the best friend of a degrading group, because one excellent location carries two mediocre ones on its back for quite a while, until it cannot. And the fifth, the one I argue about most with owners: consistency does not oppose local adaptation, it opposes local improvisation. A group can run three regional dishes per city and stay consistent, provided those three dishes also carry a card, a gram weight and a calculated food cost. What breaks the system is not that the Monterrey location sells something different; it is that it sells it without a card.