How to measure gastronomy and local development: the errors that ruin the number and the method that survives an audit

The 640 permits that were not 640 businesses

A mid-sized city government closed 2025 celebrating 640 new operating permits along its restaurant corridor, and eighteen months later the cross-check against social security payroll showed that 231 of those establishments no longer reported a single contributor: 36% silent mortality inside a program whose reports still said 640. The registry measured the entrance door and nobody budgeted for the exit door, which is the costliest methodological flaw in local economic development applied to this sector. Context did not help either: according to ACODRES, Colombian restaurants raised prices 9,8% from February 2025 just to sustain 98.000 jobs, with sales dropping 44% in some segments. A permit is paperwork. A contributor still active twelve months later is a job. The decision this cross-check triggers is cheap and simple: tie your program indicator to the payroll file, never to the application form. Food service is an unusually efficient vehicle for local economic development because it chains backward into family agriculture, forward into tourism, and sideways into youth and female first-job employment.

Why restaurants chain better than almost any sector?

More than 67% of U.S. adults have worked in the industry at some point, according to the National Restaurant Association in 2025, a figure no other service sector reaches and one that explains its real role:

it is the country's labor school, not another line in GDP. That same chaining density, however, is precisely what makes it hard to measure, because the impact scatters across four separate statistical classifications —retail, services, agriculture, tourism— and no administrative registry consolidates it. Whoever measures gastronomy as a closed sector underestimates its contribution by a third to a half. The operational consequence: track local supplier purchases over total purchases, a number already sitting in the restaurant's purchase ledger. The indicator an investment committee can defend without blushing is sustained formal employment per 100,000 USD of recorded sales, not the count of establishments served. Counting establishments blends a three-table grill house with a forty-cover restaurant running a central kitchen; internal variance destroys any average and produces reports nobody can audit.

First layer: formal jobs per 100,000 USD of sales

A formal job sustained twelve months, by contrast, can be verified against a registry external to the program, and that detail —external— is what makes it fundable before an IDB Group officer. Add the cost of losing it: each avoided departure saves 150% of the salary in replacement costs, according to StaffedUp in 2025, so turnover stops being a human resources topic and becomes a line in the income statement. Start by measuring contributors per 100,000 USD billed, quarter against quarter. The second layer is monthly food cost variance, and here it pays to be blunt: the percentage is not what informs you, the deviation is. A business holding food cost steady at 31% is healthier than one swinging between 24% and 36%, even when the latter shows a prettier average. The operating ceiling is 32% per dish as a MAXIMUM, never as a target, and above that line purchasing eats the margin.

Second layer: food cost variance exposes real health

Variance measures three things at once —portion control, supplier negotiating power and inventory quality— without a single survey. AI-driven scheduling delivers 8-12% labor savings with forecast accuracy above 90%, according to TimeForge in 2025, yet those savings evaporate when purchasing swings twelve points. The decision: require three consecutive monthly closings with deviation under three points before scaling credit. Kilos of waste per guest served is the third layer, and also the cheapest to capture because it comes from the same point of sale already issuing tickets. It serves two different purposes at once: it measures operational efficiency —high waste means miscalibrated purchasing or missing recipe cards— and it measures environmental impact with a number a city government can publish unretouched. To size the volume this sector moves, US Foods donated nearly 7 million pounds of food, roughly 6 million meals, during 2024. A corridor of 400 venues cutting waste by half a kilo per hundred guests frees tons a year and recovers cash at the same time.

Third layer: waste kilos per guest, the number nobody asks for

Despite that, almost no local development program requests it, because it demands walking into the kitchen. Install a scale and a daily waste sheet: fourteen days give you a baseline. Small restaurant mortality does not spread evenly over time, it clusters between month 14 and month 30, right after the soft loan ends its grace period and right before mid-term evaluations get published. Measuring at six months produces beautiful, useless figures, because at that point hardly anyone has closed yet and the program reports 95% survival. What would happen if that mid-term evaluation moved to month 30? Survival numbers would drop, certainly, but the instrument design would move with them: grace periods would stretch, coaching would concentrate in year two, and working capital would stop draining into initial equipment. An uncomfortable figure on time beats a comfortable one out of time. Set your measurement cuts at months 14, 24 and 30, not at the semester mark.

Where the data is born: the ticket, not the questionnaire?

At SATE Institute we approach the problem backwards from the usual framing:

instead of designing a questionnaire and going into the field, we take the data the small restaurant already produces every day —tickets, purchases, payroll, waste— and turn it into a comparable series. Masterestaurant, technology partner of the model and owner of the platform, contributes the technical layer that normalizes that flow; we define what gets measured, against which baseline, and how often it gets audited. Diego F. Parra keeps pressing one point the public sector resists: a perception survey measures what the owner believes he sold, and owners are almost always wrong on the high side. Digital channels confirm it —37% of adults order delivery at least once a week and more than 40% do so three to five times a month, according to UpMenu in 2024—, so the ticket already sits on a server. Pull it. Three numbers and no more, because a fifteen-indicator dashboard never gets reviewed.

The 3 numbers you should tattoo on yourself

One: active contributors per 100,000 USD of sales, verified against external payroll rather than the program's own form —if it falls two quarters running, suspend disbursements and audit payroll before approving the next tranche—. Two: monthly food cost deviation in percentage points, with a hard ceiling of 32% per dish —if the swing exceeds three points, freeze menu expansion and renegotiate the five highest-spend items—. Three: kilos of waste per hundred guests, captured with a scale and a daily sheet —if it climbs two months, review recipe cards and portioning before blaming the supplier—. That third number separates a program that accompanies from one that merely disburses. And the odd part is that all three come from the same place: the point of sale the business already runs. The first break is about UNIT. Counting establishments served mixes a three-table grill with a forty-cover restaurant running a central kitchen; internal variance destroys any average.

Where the measurement breaks?

The useful unit is the formal job sustained for twelve months, because it is the only one verifiable against a registry outside the program and the only one an IDB Group investment officer can defend before a committee.

The second break is about TIME. Gastronomic MSME mortality is not evenly distributed: it clusters between month 14 and month 30, right after the soft loan's grace period ends and right before most mid-term evaluations get published. Measuring at six months yields beautiful, useless numbers. If your cohort has no 36-month cut, you have no evidence of development, you have a snapshot of initial enthusiasm. The third break is about SOURCE. Self-declared data systematically overstates what the operator wants to believe and understates what embarrasses him: survey-declared waste hovers around 6% while kitchen-weighed waste settles between 12% and 16%. Those eight points are not statistical noise, they are the entire margin of an average restaurant in the region.

Where the measurement breaks — in practice?

The fourth break is COUNTERFACTUAL. Suppose a program reports 61% three-year survival and celebrates it. What would happen if untreated venues in the same corridor and the same period survived at 58%?

Then the program moved three percentage points and everything else was the business cycle taking credit. Without a comparison group —even one built by statistical matching rather than random assignment— the number is not attributable and should never enter a results report. The fifth break is COST. An in-person survey runs around 41 USD per establishment per wave, which pushes teams to collect once a year; point-of-sale extraction costs 2.4 USD monthly and allows a continuous series. The gap is not budgetary, it is about the nature of the data: a yearly photograph versus a film. I got this wrong for years, defending the rigor of the long survey over operational telemetry, until comparing both against payroll showed which one tracked the verifiable fact.