Impact measurement in gastronomy programs: from anecdotes to a verifiable dashboard for urban food security in municipal markets

Chapter 1 — An attendance sheet is not an impact indicator

Counting trained people measures logistics, not economic development, and that confusion sustains much of the spending on urban food programs. A municipal market in Bogotá or Guayaquil with 240 stalls can host nine workshops, collect 900 signatures and close the year with exactly the same number of live business units it had in January, because nobody measured the business: they measured the room. The size of the universe at stake comes from the World Bank, which puts SMEs at roughly 90% of firms and more than 50% of employment worldwide (World Bank, SME Finance), while in Mexico INEGI counted 95,4% microenterprises out of all economic units in 2023, holding barely 41,4% of employed staff (INEGI, Economic Census 2024). With that structure, allocating capital against an exit survey means allocating it blind. The right indicator is binary and verifiable: how many stalls are still invoicing twelve months later. A dashboard with daily capture catches the deviation in week two; a baseline and a final evaluation twenty months apart catch it once the budget is already spent.

Chapter 2 — Latency: why twenty months between measurements wreck the program

That is the real difference between monitoring and an autopsy. Follow the counterfactual all the way: if the market's purchasing terms push food cost from 31% to 38% in March, daily measurement lets the field team renegotiate the supplier in April and the quarter closes in the black; without it, the number shows up in the November closing report, when between fourteen and twenty stalls have already pulled the shutter down and their jobs moved into informality, which the ILO places at 57,8% of employed people worldwide (ILO, WESO May 2024 update). Waste runs alongside: UNEP calculates that 19% of available food ends in the bin (UNEP, Food Waste Index Report 2024). Every blind month carries a cash price. The very record an owner uses to know whether the day closed in the black — sales, average ticket, purchases, waste — is the cleanest SDG 8 indicator a program can obtain, and it arrives with no surveyor attached.

Chapter 3 — Daily operating data is the best local development indicator available

SATE Institute works this thesis with Masterestaurant S.A.S. as technology partner under the Twin Ecosystem Model: the institute sets the development agenda and runs monitoring and evaluation, the platform captures data at the point of sale. My judgment, after twenty years and more than 8.400 restaurants served, is that resistance never came from the merchants but from the reporting formats. The weight of the sector justifies the effort: in Colombia the gastronomic sector contributes 8% of national employment (ANDI, Gastronomic Sector Chamber 2024), and in Mexico restaurants and bars held 23,2% of tourism employment in 2024 (INEGI 2024). That employment gets counted stall by stall or it is not counted. Breaking results down by annual revenue band changes the design of the program, and most tenders still segment by number of locations, which explains nothing. Below 500 thousand USD — the bulk of the municipal market, in line with the 95,4% of microenterprises reported by INEGI (Economic Census 2024) — the lever is the purchase log: without it there is no food cost, and a three-point deviation eats the entire margin.

Chapter 4 — Revenue bands: one program, five different effects

Between 500 thousand and one million the second kitchen or second shift appears and the bottleneck shifts to labor scheduled against hourly sales. From one to five million the operator already has data and lacks reading: the problem is menu engineering and inventory turnover. Above five million the owner buys decisions, not training. A single instrument for all five bands fails in four of them. A celebrity-chef restaurant or a large-format themed venue invoicing more than five or ten million USD does not have the market stall's problem: it has the inverse one, fixed costs that forgive no slow month. Corner rent, the chef's fee, an in-house marketing team and a payroll of 90 to 160 people set a break-even point that demands filling service after service, and tolerable food cost rarely exceeds 32% per dish because the structure already consumed the rest.

Chapter 5 — The high end: above five million, cost changes its nature

These operators matter to a municipal program for one concrete reason: they buy volume and can anchor demand for family farming, which according to FAO accounts for 81% of agricultural holdings in Latin America and the Caribbean (FAO, State of Food and Agriculture 2024). Bringing them in as buyers rather than beneficiaries is the profitable move. The food security indicator that matters in a municipal market is how many households in the neighborhood can afford a healthy diet, and the region sits in the red: FAO reports 181,9 million people in Latin America and the Caribbean who cannot afford one (FAO, SOFI 2024), with a regional undernourishment prevalence of 5,1%, equal to 34 million people (FAO, SOFI 2025). Worldwide, between 638 and 720 million went hungry in 2024 (FAO/WHO/UNICEF/WFP/IFAD, SOFI 2025), and Africa passes 20% of its population, with 307 million people. The municipal market is the infrastructure that sets the final price of that basket in the neighborhood, not a folkloric attraction.

Chapter 6 — Urban food security measured in access, not in tonnes

When the dashboard shows stall margin and basket price on the same screen, the program stops choosing between merchant and buyer. Here sits the trade paradox no tender ever resolves: the health authority asks for records the merchant reads as red tape, and the program asks for impact data the merchant reads as surveillance, so we fill in two forms and use neither. The bridge is technical and cheap. One single goods-received log with supplier, batch, temperature and price feeds the sanitary file and the M&E dashboard in the same keystroke, and that record is what allows a batch recall in hours instead of weeks. For years I defended the paper sanitary file out of respect for market custom; I was wrong, because paper cannot trace backwards once an outbreak shows up. UNEP puts global waste at 19% of available food (UNEP, Food Waste Index Report 2024), and a good share of it is lost to an undocumented cold chain.

Chapter 7 — What to demand in the next tender

Write into the tender that the second tranche is released against the twelve-month survival rate of business units, measured with transactional data, and the program changes behavior before it even starts. It is one clause and it reorders everything else: the operator stops chasing attendees and starts chasing stalls that keep invoicing. Add three mandatory fields per stall — daily sales, purchases logged by supplier, and staff employed — broken down by the five revenue bands described above, and you get comparability across markets in three countries without hiring an extra consultancy. Context backs it up: the National Restaurant Association notes that 67% of Generation Z had their first work experience in restaurants (National Restaurant Association, 2025). That first job is protected with dashboards, not with workshop minutes. The first is the unit of analysis. Conventional M&E counts trained people because that is what logistics allows counting; the dashboard counts surviving economic units, which is what SDG 8 actually measures.

Chapter 8 — The five differences that change the disbursement decision

A program can train 900 people and move formal employment by zero, and that result surfaces only if you measure the business rather than the attendance. Second comes latency. With two measurements twenty months apart you discover failure after the budget is spent; with daily capture, a food cost deviation shows up in week two and the field team rewrites the market's joint purchasing terms before the quarter closes red. The ILO puts global informal employment at 57.8% of all workers (ILO, WESO May 2024 update), and in that context every month of misdirected program carries measurable social cost. Third is portability of the data into the financial system. A market stall with six months of recorded cash flow stops being an invisible credit subject: its series partially substitutes the real collateral commercial banks demand. At that point the program stops being expenditure and becomes financial inclusion infrastructure, which is precisely the argument an IDB Group officer can carry into committee.

Chapter 9 — The five differences that change the disbursement decision — in practice

Fourth is attribution. Without a comparison group, any improvement can be explained by seasonality or by the territory's economic cycle; with non-intervened stalls from the same market, difference-in-differences becomes computable and the reported impact survives external audit. It costs less than expected, because the comparison group is simply the program's own second wave. Fifth, and the one that triggers the most resistance, is that the dashboard exposes the program operator. A public panel showing food cost, waste and survival by cohort makes result-polishing impossible. I prefer that discomfort: a program that cannot withstand its own dashboard should not receive a second tranche, and saying so out loud saves years of decorative cooperation.