Profitability #7 — Benchmarking across sites: find your best kitchen and clone it

The fix was already in the village

In 1990–91, Jerry and Monique Sternin arrived in Vietnam for Save the Children, working in villages where roughly 65% of children were malnourished. The standard approach would have been to import outside solutions — different food, external aid, new farming methods. Instead, they asked a different question: in these same poor villages, with the same resources, were any children not malnourished? There were. A handful of families, no wealthier than their neighbours, had well-nourished kids. Studying what those families did differently, the Sternins found something specific: they fed children three or four smaller meals a day instead of the village norm of one or two large ones, and added tiny foraged shrimp, crabs and greens to the rice — foods most villagers believed, wrongly, were unsuitable for young children. The fix didn’t need to be imported. It already existed a few houses away. Spread peer-to-peer rather than taught top-down, the practice produced sustained recovery in hundreds of children in the pilot villages and was later scaled nationally. The method got a name: positive deviance — find who’s already solving the problem with the resources everyone else has, and study them instead of designing a solution from scratch.

The same method, applied somewhere completely different, produced one of its most rigorously documented results: starting in 2001, the VA Pittsburgh Healthcare System used positive deviance to attack MRSA infections, cutting cases from roughly 60 a year to 17 and reducing surgical-unit infection rates 70%. The approach was later validated in a peer-reviewed, cluster-randomized trial published in the New England Journal of Medicine in 2011, and expanded to 152 VA facilities nationally — a 65% reduction in MRSA infections across the network. Different industry, same structural insight: somewhere in a system that looks uniform on paper, someone has already solved the problem everyone else is still working on.

Switzerland already runs a version of this measurement, even without calling it positive deviance. ANQ, the country’s national hospital-quality body, publishes per-facility infection-rate data across Swiss hospitals — 2024 figures for colorectal-surgery site infections ranged from 0% at the best-performing facilities to roughly 20% at the worst, against an 11.6% national average. Nobody needs to argue that variance this size exists between facilities that in theory follow the same national protocols — ANQ’s own published numbers prove it every year. The gap between a group’s best and worst kitchen is rarely made that visible. It should be.

Your best kitchen already exists

A multi-site restaurant or catering group has the same structure as that Vietnamese village, minus the excuse. Every location, in theory, runs the same recipes, buys from the same negotiated contracts, and follows the same brand standards. And yet real variance shows up between sites anyway — different chefs, different supplier relationships on the ground, different staff turnover, different local demand. Some restaurant-technology vendors describe this in benchmarking terms: comparing bottom-quartile to top-quartile locations across a group is a genuinely established practice, sometimes called quartile analysis, used specifically to surface which sites are quietly underperforming and which are quietly doing something right. The exact size of a “normal” versus “concerning” spread between locations varies enough by source that it’s not worth quoting a single number as fact — but the practice of measuring the spread, rather than assuming uniformity, is the real finding worth acting on.

Finding the deviant is the easy part — spreading it is the point

A US restaurant chain, working with a consultancy (case study published, company name withheld), rebuilt a single pilot location as a “model restaurant” over 30 days — cutting kitchen labour 40% and front-of-house labour 35% while hitting the pilot’s highest hourly sales on record — then rolled the same model out region by region across 550 locations. It’s a consultancy-published case, not independent research, so treat the specific percentages as one company’s reported result rather than a universal guarantee. But the sequence it describes matches the Vietnam and Pittsburgh pattern exactly: find the site that’s already outperforming, understand precisely what it’s doing differently, then spread that — deliberately, systematically — rather than issuing a general instruction to “do better.”

This only works if every site is telling the truth

Here’s the catch, and it’s the reason this is the last post in the series rather than the first: benchmarking across sites is only as good as the numbers being compared, and every post in this series has been about making one of those numbers real.

• If one site’s theoretical-vs-actual gap (Profitability #1) is being measured honestly and another site’s isn’t, comparing their food cost percentages compares two different things wearing the same label.
• If contribution margin (Profitability #2) is calculated from stale recipe costs at one site and current ones at another, a “Star” dish at one location might be a mislabeled Plowhorse at the other.
• If yield-adjusted cost (Profitability #3) is applied at one site and skipped at another, their food costs aren’t comparable even if every other number matches.
• If waste (Profitability #5) is tracked by weight and cause at one site and estimated from shrinking stock counts at another, one site’s “low waste” might just be a site that isn’t measuring.
• If portion specs (Profitability #6) are enforced at one site and eyeballed at another, the recipe cost each site is reporting against was never the same recipe in practice.

A group that tries to benchmark across sites before these five are consistent isn’t finding its best kitchen. It’s finding its best bookkeeping — which is a different thing, and a genuinely dangerous one to reward.

How CalcMenu makes the comparison real

• The same metric, calculated the same way, at every site — theoretical food cost, contribution margin, yield-adjusted cost, waste value and portion variance all computed from the same rules everywhere, so a comparison across sites is actually comparing performance, not measurement inconsistency.
• Top and bottom performers surfaced automatically — by dish, by category, by site, instead of waiting for someone to notice a spread during a quarterly review.
• The winning recipe or practice, visible and shareable — once a site is genuinely outperforming on a real number, what it’s doing differently should be easy to find and easy to copy, not locked in one head chef’s habits.
• Consistency enforced going forward — once a practice spreads from the best site to the rest, the same system that surfaced it keeps every site honest about whether it actually stuck.

CalcMenu doesn’t find your best chef or decide what to copy — that’s still a judgment call for whoever runs the group. It makes sure that when a site looks like the best one, it actually is, because every number behind that conclusion was measured the same way everywhere else.

Before you benchmark across sites

Four questions worth answering before ranking locations against each other:

1. Are theoretical food cost, contribution margin and yield-adjusted cost calculated identically at every site, or does each location have its own local method?
2. Is waste tracked by weight and cause everywhere, or does “low waste” at one site just mean nobody’s measuring it?
3. Are portion specs actually enforced at every location, or does the recipe card mean something different depending on which kitchen you’re standing in?
4. When you find your best-performing site, do you have a way to actually spread what it’s doing — or does the insight stop at a slide in a quarterly deck?

If the honest answer to the first three is “it varies by site,” the benchmarking exercise isn’t ready yet — the six posts before this one are the prerequisite, not optional background reading.

---

Want every site measured the same way, with the best performers surfaced automatically? Book a free 15-minute call with our team — no commitment: Schedule a call.

Sources

• The Vietnam Story: 25 Years Later — Positive Deviance Collaborative
• Background — Positive Deviance Collaborative
• Positive deviance — Wikipedia
• Targeted Infection Prevention (Positive Deviance) to Reduce MRSA — New England Journal of Medicine, 2011
• PD MRSA Prevention for VA Pittsburgh Healthcare System — Positive Deviance Collaborative
• Food cost variance control — Supy
• Restaurant prime cost benchmarking across multi-unit locations — TRIS
• How to compare performance across multiple restaurant locations — Restaurant365
• Case study: Remaking a restaurant chain — Heidrick & Struggles
• ANQ — Nationaler Verein für Qualitätsentwicklung in Spitälern und Kliniken
• Taux d’infection — Quel Hôpital