Impact monitoring and evaluation (M&E) guide: the errors that inflate reporting and the method that survives an audit

Where should a gastronomy program's impact indicator come from?

A valid indicator is read from the restaurant's operating system (payroll, sales, purchasing, turnover), never from an exit survey the beneficiary answers while the trainer watches.

The distinction is budgetary before it is methodological. Small firms account for 99% of Latin American companies and 61% of formal employment, yet only 25% of output, according to CEPAL, so a program promising productivity has to prove it inside that gap instead of in an attendee satisfaction box. Ask for «people trained» and you get a signature sheet; ask for «cooks under active contract at month nine per social security filings» and you get a figure any third party can rebuild without phoning a single participant again. One takes twenty minutes to sign. The other survives an audit.

Participation is not a result: the error that fails external evaluations

Call participation a result and the external evaluator will send the report straight back; in employability programs that single confusion produces more rejections than any methodological flaw. A cook swears in good faith that he respects portion weights while month-end closes food cost variance 4.8 points above theoretical, against a 32% per-dish ceiling, and that second number is one anybody can recompute from purchase invoices and closing inventory. Attendance captures intention. Transactions capture effect. Underneath sits an aggravating structural fact: the ILO puts regional youth unemployment at 20.4% versus a little over 7% among adults, so a program reporting only workshops delivered cannot tell whether it moved that gap or merely filled six Saturdays for two hundred kids.

The measurement window: month nine, not day ninety

Between month four and month nine is where food-service job retention actually settles, so a reading taken at day ninety photographs the start and files it as a trajectory. The mistake comes from the calendar rather than from statistics: teams measure when the tranche has to be disbursed. Put the counterfactual on the table. A program reports 78% placement at day ninety; high season ends, the operator trims shifts, and what is left by month nine? Half of it, with the closing report already signed. Confecámaras finds that barely 34 of every 100 companies created in Colombia reach their fifth year, and with that mortality underneath, any trajectory indicator read at three months measures opening enthusiasm, not staying power.

Without a comparison group there is coincidence, not attribution

If you cannot name the cohort you are comparing against, what you have is an activity report with charts, not an impact evaluation. That is the rule I put in front of investment officers, and it does not bend: when the quarter after the program lands on high season, the indicator climbs by itself and the agency ends up paying for the calendar. Now, a word about budgets. A randomized trial rarely fits an MSME operation; a cohort of neighboring restaurants of the same format and average check does fit, and it is enough to pull the effect out of seasonal noise. INEGI documents that 96 of every 100 Mexican restaurant units are micro firms employing 70 of every 100 workers in the sector, so the comparison universe sits four blocks away.

Whoever measures must not be whoever sells the software

Whoever measures cannot also be paid for the software under measurement, and that boundary is a credibility condition, not a governance formality. SATE Institute runs its gastronomy employability programs under a twin-ecosystem model with Masterestaurant S.A.S. as technology partner: the institute sets the development agenda and evaluates, the platform supplies the instrumentation (MTIE, Restaurant Model Canvas, meseros.ai and its dashboard) that turns daily operations into a data series. According to Diego F. Parra, restaurant consultant and founder of Masterestaurant, an indicator's data has to originate in a transaction and not in a form, so the vendor holds no lever over the needle that grades it. CEPAL reports over 70% of Latin American small firms with no internet presence, and over 60% of those online never transact: instrumenting means existing digitally first, measuring second.

How to read these numbers in YOUR operation?

A single-shift venue does not need the dashboard a five-location group needs, so scale these benchmarks down before arguing methodology. In the small place two numbers do the work:

food cost variance against theoretical, and crew turnover. Close above 32% two months running and the program failed, however many certificates went out. With two or three locations, add prime cost and month-nine retention site by site, because the spread between venues tells you whether the effect came from training or from one particular head chef. From five up, the number that carries a banking conversation is the productivity gap, that 61% of employment against 25% of output CEPAL reports. Every point you close there is credit risk your loan officer stops carrying blind.

Measuring badly gets paid out of the loan book

Two files reach the committee with identical amounts in identical sectors: one shows 340 people trained, the other verifiable payroll plus nine months of purchasing variance. The first supports no calculation. The second feeds the model, and the gap between them is paid out of the loan book, because the margin that predicts delinquency lives in food cost and prime cost, never in workshop headcount. One committee objection deserves an early answer, the claim that this sector offers no career: the National Restaurant Association measured in 2026 that 9 of every 10 managers and 8 of every 10 owners started at entry level. If that mobility is real, M&E has to capture it through payroll and enrollment records, not testimony.

Where these figures come from and how far they reach?

ILO for regional youth unemployment, CEPAL for small-firm weight and the digital gap, INEGI 2022 for the structure of Mexico's restaurant sector, Confecámaras for five-year business mortality, National Restaurant Association 2026 for internal mobility:

that is where this guide's benchmarks come from, all public and checkable. None of it is a primary Masterestaurant study, and the limit deserves plain language: these are national or regional averages capturing neither format nor average check nor seasonality of YOUR operation, and several blend informality definitions that differ country by country. Use them for order of magnitude and as an argument before a financier, never as a stand-in for your own series. The figure governing a cash decision is yours, measured across two consecutive periods with one method. Start this week: export nine months of purchasing and payroll, then compute prime cost by site.

The five differences that decide whether the report clears

Self-reported numbers record intention; instrumented ones record transaction. In perfect good faith a cook will swear he respects portion weights, and then closing inventory returns a 4.8-point gap between theoretical and actual food cost. Only that second number can be recomputed by an evaluator holding invoices, which is why it is the only one that belongs in the logframe. Nobody shortens the window for methodological reasons: the disbursement calendar shortens it. The effect matures between month four and month nine, by which time the tranche has been wired and the report signed. Three months buy a photograph of the launch, never a trajectory. With no mirror there is no attribution, only coincidence.

The five differences that decide whether the report clears — in practice

Let high season fall right after the program and the operator will report an 18% sales jump the calendar handed him. A difference-in-differences estimator against eligible restaurants that missed the cut separates one thing from the other. Formality has the rare virtue of being binary: a job with social security enrollment gets checked against a public registry, while a job declared on an operator's spreadsheet gets checked against nothing. Local economic development stakes its credibility on that difference. Well-built M&E data has a second owner nobody planned for, the credit analyst. Twelve months of sales, prime cost and staff turnover is precisely what an MSME scoring model needs before it places working capital, and for the beneficiary that by-product frequently outlives the closing report.