How to measure gastronomic social impact: traditional method vs Masterestaurant method

Chapter 1 — Why does the expost survey arrive too late to measure gastronomic social impact?

The expost survey arrives too late because it asks, six or twelve months out, about facts the restaurant already recorded the very day they happened, and the beneficiary's memory competes against a dated tax receipt.

When the FAO documents that 181.9 million people in Latin America and the Caribbean cannot afford a healthy diet (SOFI 2024), that figure lands in the logical framework of nearly every gastronomic project in the region, yet the program closes without being able to say how many formal contracts it signed or how much it bought from family farming, which the FAO itself puts at 81% of the continent's agricultural holdings. The observation unit is badly chosen. A cook forgets whether his March contract carried social security enrollment; the payroll ledger, with its filing number, forgets nothing. Change the subject you observe and the memory problem disappears at a stroke. My thesis is uncomfortable for the evaluation industry: the most reliable decent-work indicator of a gastronomic program comes out of the payroll ledger, not a form.

Chapter 2 — The payroll ledger is the SDG 8 instrument, not the questionnaire

The restaurant industry employs 10% of the United States workforce (National Restaurant Association 2024) and in Mexico sustains around 2 million jobs across 581,530 establishments (INEGI/CANIRAC 2022), a universe where turnover is measured in weeks and an annual survey loses half its sample along the way. Payroll, by contrast, captures every hire, every exit, every overtime hour and every social contribution with a date and a receipt behind it. We work this angle through the Twin Ecosystem Model with SATE Institute: the institute sets the agenda and runs the programs, Masterestaurant S.A.S. instruments the operation as technology ally. Diego F. Parra puts it this way at investment tables: if the number is not in the books, it is not evidence, it is literature. Measuring month thirteen costs practically the same as not measuring it, and that asymmetry is what opens the program to small operations.

Chapter 3 — Instrumenting costs marginal; surveying costs fixed, and eligibility is decided right there

A baseline survey with treatment and control carries a fixed cost per round, so the evaluator imposes an eligibility floor to keep measurement from eating the subsidy; with instrumentation, the data was already generated to run the business and all that remains is classifying it. Drop the floor to operations under 500 thousand dollars a year and the bulk of the Latin American MSME universe walks in, which is exactly where the youth employment that SDG 8 chases actually lives: 36.9% of 16-to-19-year-olds were in the labor force in 2023 (BLS, via the National Restaurant Association), and their first payroll is almost always a restaurant's. I got this wrong for years, defending pilots with large samples when what was missing was continuity of the data. Breaking results down by annual revenue band matters more than counting locations, because the available instrument changes completely at each tier.

Chapter 4 — Five revenue bands, five different ways the indicator turns out right or wrong

Below 500 thousand USD the accounting tends to be outsourced and monthly, so the only clean record is the social security filing and the recommendation is to measure formalization and nothing else. Between 500 thousand and 1 million a point of sale with inventory appears, and waste can already be derived, no small matter when ReFED 2025 attributes 70% of waste to food left uneaten on the plate. Above 1 million the ERP arrives and purchases get traced by supplier, which is the doorway to short supply chains. Over 5 million the group already has a people function, with internal promotion records. And past 10 million the problem stops being capture and becomes reconciling criteria across separate legal entities. The celebrity restaurant or the large-format themed venue, above 5 million dollars in revenue, holds the best data in the market and the worst incentive to publish it.

Chapter 5 — The high end: the celebrity-chef restaurant and its own measurement costs

Its payroll is 100% banked, its purchasing goes through tender and its inventory is counted daily, so instrumentation is nearly free; the real cost is reputational and legal, because the same system that proves family-farming purchases also proves the gender gap in command. The number is blunt: women hold 38% of executive posts in United States restaurants against 63% at entry level (Restaurant Business 2024), and across hospitality and catering they make up 60% to 70% of workers according to the ILO. A large-format group that instruments properly will see that inverted pyramid on its own dashboard. Publishing it anyway, with a dated correction plan, is what separates an auditable report from a brochure. The same purchase ledger that feeds the food cost variance calculation feeds the short-supply-chain indicator, and that overlap is the operational finding that makes measurement sustainable. No manager fills out an IDB form with enthusiasm; every one of them reviews food cost variance because their bonus comes out of it.

Chapter 6 — The social indicator as a byproduct of management control, not an annex

When the share of local-producer purchasing is computed from that same supplier master file, the social indicator updates on its own, on the rhythm of the accounting close rather than the donor report. Add the waste front, where ReFED 2024 attributes more than 43% of United States foodservice surplus to full-service restaurants, and you have three environmental indicators derived from inventory you already keep. Measurement stops being an administrative burden and becomes one more view of the same cash data. Suppose IDB Lab tied the second disbursement to a dashboard with payroll, purchasing and inventory connected, rather than to a closing report. The first effect would be a sharp drop in eligible projects, because artificial intelligence penetration in Latin American and Caribbean firms sits under 4% against more than 20% in Europe (ECLAC 2024), and without a digital layer there is no data to connect. The second effect would land around month nine: operators who did connect would start reporting monthly and their reporting cost would fall close to zero, while the rest kept paying consultants to reconstruct a year backwards.

Chapter 7 — What would happen if a fund demanded instrumented data instead of a narrative report?

The third one is the one I care about, which is that the fund would stop comparing narratives and start comparing time series. The paradox resolves there:

demanding more technology up front makes inclusion cheaper later, because the eligibility floor drops instead of rising. Pick labor formalization, connect this month's social security filing and publish the number: contracts with active enrollment over total people in operation, cut on the 30th. Nothing else. A single indicator with an auditable source is worth more to a program officer than twelve estimates from a perception survey, and by month two you already have variation, which is the only thing an evaluator can read as effect. Context pushes that way: public school feeding programs now reach 466 million children worldwide, 80 million more than in 2020, a 20% increase according to the WFP (State of School Feeding Worldwide 2024), and that expansion of public procurement will reward operators who can prove their chain.

Chapter 8 — Start with one indicator and one source, this week

And if by March you want a second indicator, make it local-producer purchasing from the supplier master file, not a new survey. The unit of observation shifts. Traditional evaluation observes the BENEFICIARY and asks; instrumentation observes the TRANSACTION and classifies it. A cook may forget whether his March contract was formal; the social security enrollment record does not. Cost becomes marginal rather than fixed. Measuring month thirteen costs practically the same as not measuring it, because the data was already generated to run the business. That is what lets a program lower its eligibility floor to operations below 500 thousand USD a year, which are most of the MSME universe. The social indicator stops being an annex and becomes a by-product of management control. The same purchasing record that feeds food cost variance feeds the short food supply chain indicator. One data point, two uses, zero extra administrative burden on the operator.

Chapter 9 — The five differences an investment committee actually decides on

The counterfactual is recovered through administrative records instead of a surveyed control group. Where tax and social security registries exist per establishment, matching runs against untreated outlets in the same territory and revenue band, with no new fieldwork. And a new risk appears that deserves naming: instrumentation measures what is formalized and goes blind to what is informal. In a sector where the ILO documents female participation between 60% and 70% in hotels, catering and tourism, a good share of the work that most needs measuring sits outside the payroll ledger. That hole is covered with targeted qualitative fieldwork, not by pretending administrative data sees everything.