How to improve gastronomic social impact measurement: real pricing of the traditional method against the Masterestaurant method

What does it cost today to measure social impact in a food-service MSME?

Between USD 600 and USD 1,400 per unit buys the classic three-year cycle across Latin America, while operational telemetry runs USD 790 to USD 1,730 for the same period.

That first range buys exactly three observations per establishment over thirty-six months. The second delivers thirty-six monthly cuts covering sales, payroll, purchasing and waste, and SATE Institute runs the license on the Masterestaurant S.A.S. platform. Divide either range by its number of observations and the only figure that moves an investment committee shows up: PRICE PER USEFUL DATA POINT drops between 88% and 94%. The monthly invoice looks scarier, and that fright has sunk more than one sound tender at the wrong table. Total cost of knowledge comes out lower, every time the arithmetic gets done properly. Let me break the bands apart, because the average lies. USD 600 to USD 850 per MSME pays for a structured survey applied by a local enumerator, data entry, cleaning and a descriptive report of self-reported employment and sales, with no counterfactual and no documentary verification.

What each investment range includes?

The USD 850 to USD 1,400 band brings in quasi-experimental design with a comparison group, payroll cross-checked against social security records and a senior evaluator's signature, which is precisely what a multilateral fund accepts as evidence of attribution.

On the telemetry side, USD 790 to USD 1,100 covers point-of-sale licensing, a dashboard and monthly series of sales and headcount. Climb to USD 1,730 and you add the supplier purchasing module, waste traceability, electronic invoicing integration and field support for data hygiene. Three line items never show up in the budget and get paid anyway. Start with sample attrition: with business mortality close to 60% at five years in Latin American chamber-of-commerce registries, a cohort of 300 restaurants reaches the final evaluation with 180 or fewer, so you paid for 300 field visits and drew conclusions about barely half of them.

The three costs no technical proposal ever declares

Self-report bias comes next, and in a sector where 52 of every 100 Latin American tourism workers operate informally, according to CEPAL, asking how many people someone employs returns the prudent answer rather than the real one. Lag closes the list: twenty-one months between data and decision turn the report into history. None of the three shows up in a technical proposal, and all three land squarely on the funder. Five variables explain nearly all the spread in fees, and the firm's reputation is not among them. Geographic dispersion rules: serving kitchens scattered across four departments rather than a single urban corridor raises fieldwork between 25% and 40%, since the evaluator bills travel and not analysis. Cohort size pushes the other way, hard: move from 150 to 600 units and unit cost falls between 30% and 45% as the methodological design gets spread thinner. Demanding a counterfactual with a control group adds 35% to 60% over the contract baseline.

Which factors move the price, and by how much?

On the telemetry route, digital maturity weighs: a venue without an electronic point of sale needs an extra USD 120 to USD 280 in installation and training.

Quarterly reporting instead of annual tacks on 15% to 20%. Nothing costs development cooperation more money than this confusion, so I will say it bluntly: an impact evaluation with a control group answers whether the program CAUSED the effect, while an operational dashboard answers what is happening this week across three hundred kitchens. Different questions, sold as one. Buy credibility when you needed timeliness and you get elegant reports about businesses already closed; buy timeliness when the board wanted causal attribution and you get beautiful dashboards the committee dismisses in ten minutes. Diego F. Parra frames it this way inside the Masterestaurant method: telemetry is the nervous system, external evaluation is the audit, and no sensible restaurant swaps one for the other. For years I argued that measuring well once was enough.

Credibility and timeliness are not the same product

I was wrong. A program measuring once a year loses part of its cohort without noticing: the annual survey is not cheaper, it is blinder, and that blindness gets paid in decisions that arrive late. Picture three hundred food-service MSMEs in Barranquilla with formal employment captured every twelve months. Month eight: chicken climbs 22% and eighty kitchens cut shifts. Month eleven: forty of those eighty close. The month-twelve survey reaches the 260 survivors, reports stable employment and the program declares victory, though it lost 13% of its cohort to a cost variable the purchasing dashboard would have flagged in week three. With monthly food cost readings, the manager had nine months to redesign recipe cards before the first closure, and nine months is a long runway in this trade. Split the contract in two, because buying it whole is what inflates the price: continuous telemetry licensing for the entire served universe, and external evaluation with a counterfactual on a 20% to 25% subsample.

How to negotiate and cut the invoice without losing rigor?

That single move trims 30% to 40% off the evaluation component without touching the validity of the causal finding, because the evaluator stops collecting a baseline, reads it from the historical series instead, and bills analysis rather than enumerators.

Negotiate unit price afterwards in volume tiers, with a reduction clause past 400 licenses. From the data provider, demand open-format series and usage rights held by the program, never by the vendor. Installation is paid once; spreading it monthly is money given away. Vendors resist the split, and their resistance tells you where the margin was hiding. The hybrid model wins on total cost and on decision usefulness, however uncomfortable the monthly outlay, and that sentence belongs at the top of the approval memo instead of the fee table. Numbers hold it up: the same order of investment per MSME, twelve times more observations, and a third-party-evaluated subsample that preserves the attribution evidence a multilateral fund requires.

The decision that belongs in the approval memo

Add market context: Mexico's restaurant industry counts 581,530 economic units according to INEGI's 2024 Economic Census, a universe impossible to survey and perfectly readable through telemetry. Your next step is not requesting three more quotes. Open the current tender document and split it into two budget lines: continuous licensing for the universe, causal evaluation for the subsample. Once those two lines exist, the memo writes itself. The traditional method buys CREDIBILITY and continuous measurement buys TIMELINESS, two different goods the market sells as if they were one. An impact evaluation with a control group answers whether the programme caused the effect; an operating dashboard answers what is happening this week across 300 kitchens. Confusing them is expensive in both directions. Unit cost deceives. The USD 600 to USD 1,400 per MSME of the classic cycle buys three observations; the USD 790 to USD 1,730 licence buys thirty-six, plus purchase and waste detail.

Where the two models genuinely diverge?

Price per useful data point drops 88% to 94%, and that is the figure that belongs in the approval note, not the monthly invoice.

Though no budget line names it, sample attrition is a cost. With 38% to 55% effective response among highly informal MSMEs, you pay for the full round and keep half the information, and the missing half is not random, because units in crisis are precisely the ones that never answer. Survivorship bias inflates the programme's reported result. Continuous measurement carries an Achilles heel no vendor mentions: it only sees whoever keeps the software running. Should 20% of the portfolio abandon the platform, that 20% vanishes from the series and survivorship bias walks back in through another door. Hence the hard rule of phone follow-up on platform drop-offs, costed from day one. On food loss and waste the asymmetry is brutal.

Where the two models genuinely diverge — in practice?

Characterising waste the classic way, with weighing, categorising and sampling by shift, costs USD 12,000 to USD 28,000 and yields a two-week snapshot;

operating inventory delivers waste per input, per day, per venue, at zero marginal cost, because the restaurant already needs that number to stay solvent. Methodological independence still belongs to the traditional method. When the programme operator also supplies the measuring system, a structural conflict appears that good intentions will not settle: what settles it is external audit of the series plus a contract giving the funder access to raw data, with explicit separation between who operates and who evaluates. The Twin Ecosystem Model between SATE Institute and Masterestaurant S.A.S. exists for exactly that.