Prime Cost from 68.4% to 62.3% in five months: the cost stress scenario simulation that exposed the leak, using MTIE and the Standard Recipe Generator

Three straight overdrafts with sales growing 6%

A 14-table trattoria with 11 employees entered the MSME portfolio showing the symptom that fools a development bank most often: rising sales and a bank account in the red. Seven years of operation, a 21 USD average check, 68% of revenue from the dining room and 32% from aggregator delivery, annual billing under 500 thousand USD, 6% year-over-year growth, and three consecutive months of overdraft with two deferred loan payments. The income statement flagged nothing, because a monthly P&L folds cost of sales into a single line and that line always reconciles against purchases. What never reconciles is the recipe. The owner summed up his own diagnosis in a sentence heard across the whole portfolio: the money comes in fine, and then it evaporates in production. That is where the work started, not at the menu price. It measures the exact point where an operation stops covering payroll when an input rises, and that point almost never matches the owner's intuition.

What does a cost stress-scenario simulation actually measure?

Simulation does not extend last year's trend: it deliberately breaks the cost structure, input by input, until the breaking threshold appears.

Here the trattoria withstood an 8% protein increase — tight, but current — broke at 18%, where contribution margin no longer covered fixed costs, and lost cash flow entirely at 24%. The owner had estimated his tolerance at 20%. He was off by twelve points, and in a market where food waste occupies the equivalent of nearly 30% of the world's farmland (UNEP, Food Waste Index 2024), protein price pressure is not a rare event but the baseline condition. Purchase price explained 2.3 of the 9.7 points of variance between the recipe's theoretical cost and the plate's real cost. The remaining 7.4 came from three sources no supplier can negotiate for you: free-hand portioning on the hot line, unrecorded waste at expediting, and uncontrolled comps in the dining room.

9.7 points of variance: where they really came from

Do the proportion yourself: an owner haggling over cents with a supplier is working on less than a quarter of the problem while the other three quarters walk out through the pass every single day. An honest concession belongs here, because for years I pushed purchasing negotiation first myself: it is the most visible lever and the most comfortable one, which is precisely why it moves the least cash while the production gap stays open. Measuring the plate comes before calling the supplier. The recipe costing module of the Masterestaurant ecosystem was used, the same one that anchors the food cost framework with its 32% ceiling per plate, loaded with the 26 active menu recipes and checked against 21 days of real service. The procedure was easy to describe and hard to execute: weigh the portioning of the eight dishes carrying 61% of sales, log waste per shift on a two-column sheet, and close the cycle with weekly inventory counts instead of the monthly ones already in place.

The tool applied and how it was applied

Diego F. Parra led the reading of results alongside the SATE Institute team, and the kitchen received a single instruction: nobody plates a portion that has not been weighed at least once. With that base you can finally simulate, because simulating without measuring is inventing. Under the original structure, an 18% protein increase would have eaten contribution margin in eleven weeks and forced the owner to choose between payroll and rent by the fourth month. This is not a laboratory hypothesis: in a region where 52 out of every 100 tourism workers are informal (CEPAL, Panorama del turismo en México y América Latina 2024), the easy exit under that squeeze is to informalize the team, and that shortcut destroys the operation faster than the cost itself. Once the 7.4-point operating gap closed, the same 18% scenario stopped being terminal: the follow-up simulation moved the breaking threshold from 18% to 29% without raising a single menu price.

What would have happened at 18% with nothing fixed?

The trade's paradox resolves right there, because lowering real cost did not require buying cheaper — it required serving what the recipe said.

The gap between theoretical and real cost fell from 9.7 to 2.8 points in sixteen weeks, and the overdraft disappeared in the third month. Consolidated food cost went from 38.4% to 31.1%, inside the method's ceiling; recorded waste dropped 44% against the first month of measurement, and comps — which nobody had been counting — turned out to be worth 1,180 USD a month, roughly a cook's salary and a half. Sales did not move: same 6% year-over-year growth, same 21 USD check, same channel mix. That is the uncomfortable part of the case and it deserves plain language: not one extra dollar of revenue was won, the operation simply stopped losing the dollars already coming in. An industry that sold 1.5 trillion USD in the United States in 2025 (National Restaurant Association, 2025) runs on single-digit margins, and there seven points decide survival.

Transferable lessons

Below 500 thousand USD a year, this week weigh the portioning of the five dishes that carry more than half your sales and write down shift waste on paper: no software, no consultant, two weeks of data already expose the gap. Between 500 thousand and 1 million, close inventory weekly instead of monthly and simulate your menu against a 15% rise in your dominant input before signing any contract. Above 1 million, demand real cost per plate and per location in the monthly report, because the group average hides the worst store. Above 5 million — the media-chef archetype with two high-volume formats and brand licensing — the risk moves elsewhere: the gap no longer sits on the hot line but in the absence of one shared recipe across venues, and the first step is sealing the master recipe book with gram tolerances.

Transferable lessons — in practice

Over 10 million, in a group or chain, the simulation runs by input category and by business unit, and the first step is naming who signs off on the accepted breaking threshold; with no owner for that number, the model becomes one more report. I would not expect these seven points in three contexts, and saying so protects the reader from survivorship bias. First, in an operation already weighing portions and counting inventory weekly: there the gap usually sits between 2 and 4 points, the exercise returns far less, and the effort only pays off across several venues. Second, in short-menu, high-rotation formats — quick service with fewer than ten SKUs — where equipment already standardizes portioning and margin is decided by labor cost, not by the recipe. Third, in businesses with a genuine demand problem: if the dining room does not fill, closing the production gap improves break-even but solves nothing, and in the United States, where 22% of sector workers were born abroad (Independent Restaurant Coalition, 2024), a staffing regulation change can move cost more than any recipe.

Limits of this case

Weigh the plate before you believe the model. The traditional method watches purchase price; the Masterestaurant method watches the GAP between what the recipe says a dish costs and what the kitchen actually spent. Purchase price explained 2.3 of the 9.7 variance points here, and the remaining 7.4 came from free-hand portioning, unrecorded waste and uncontrolled comps. An owner negotiating with suppliers while leaving recipes untouched is fighting a third of the problem. Simulating is not forecasting. A forecast extends a trend; a cost stress scenario simulation deliberately breaks the operation to locate the exact point where it stops covering payroll. That point sat at +8% on protein, a level any market in the region crosses in one bad quarter, and the owner had no idea it existed. Labor cost does not fall by firing people. It fell here by rebuilding shifts against the demand curve: Labor Cost percentage actually ROSE by a point and a half, and the operation still gained seven EBITDA points, because 173 monthly overtime hours disappeared and with them the errors of an exhausted brigade.

Where the two methods genuinely diverge?

Confusing Labor Cost with cheap payroll is the most expensive mistake in MSME portfolios. Food loss and waste is not an environmental matter separate from the cash register.

According to UNEP (Food Waste Index 2024), food waste occupies the equivalent of nearly 30% of the world's agricultural land; at this trattoria's scale that meant 11.4 estimated waste points coming straight out of EBITDA. SDG target 12.3 and restaurant profit point at the same kilo of food. For lenders, the difference is one of measuring instrument. A score reading only sales and credit bureau cannot tell this trattoria apart from an identical one with a 2-point variance, and both get priced the same. Theoretical-to-real variance is the operating variable separating a solvent gastronomic MSME from one heading into arrears, and two inventory counts are enough to read it.