Why the new menu comes out different at every location.
It is almost never the recipe.
A rollout arrives as a specification and a launch date. Everything between the two gets resolved locally, by whoever is standing there, differently in each building.
Which station owns the dish. What happens when the prep list runs long. Which substitution is acceptable at eight o'clock on a Saturday. None of that is in the spec, all of it gets decided in week one, and by the time the variance shows up in your numbers it has already been trained in. We settle those questions before the date is fixed, and hand over the standard your team uses to hold it.
Anyone can be inconsistentat work that never repeats.
Every method built to remove variance needs a task that repeats often enough to be measured and held. Engineered labor standards, time studies, continuous improvement. Point them at a group where the menu, the crew and the calendar all reset before anyone has settled in, and there is nothing for the method to grip. That is not an immature business. It is the business. The hard part is consistency across work that never comes back the same way twice.
The five places it actually drifts
None of them are the recipe, and four of them are decided in the first week by people who were never told they were deciding anything.
Nobody owns the station
The spec says what the dish is. It rarely says which station builds it, who plates it when that station is buried, and what moves off that station to make room. Each location answers this differently in week one and never revisits it.
The training assumed the wrong crew
Training is built and demonstrated by people who already know the dish. It then goes to a crew who has never made it, on a shift, between covers. What survives that transfer is not what was demonstrated.
The throughput math was never done
A dish is tested at the first cover, not the four-hundredth. Ticket times, hold windows and the pass all behave differently at volume, and the first Saturday is where that gets discovered rather than where it gets planned for.
Substitutions were left undefined
A product will be short somewhere. If the acceptable substitution is not written down, every location invents one, and the invented ones are reasonable, different, and permanent.
The revision trail has no current version
By launch there are four versions of the spec in circulation across email, a shared drive and a printed copy in a binder. Locations execute whichever one they were handed, which is usually not the last one.
What you receive
A menu consultant hands back a menu. This ends in the four things that decide whether it survives contact with your locations.
The operating standard
How it is executed, station by station, written for the crew who will actually run it rather than for the person who designed it. The licensed artifact, not a slide deck.
The throughput model
Ticket times, station load, hold windows and the pass, modeled at volume before the date is fixed. This is the part that kills dishes early, which is the cheapest place to kill them.
The training design
Built for someone who has never run it, on a shift, and measured on whether it holds after the person who trained it has gone home.
The sign-off chain
One document, checked by every discipline that would have to deliver it: culinary, operations, purchasing, food safety, training and the staffing partner. Each one competent to deliver the part they sign, and paid the same whether they approve it or not.
How the work gets made
Applied AI, and it is method rather than product. The rollout is modeled before anyone spends: the throughput, the station load, the cost, the failure cases. That is what makes it affordable to kill a dish in week two instead of finding out in month four across twelve locations.
The judgment stays with people who have run a service. What is new is the reach: at every point where the work varies, the experience of many specialists is available at once rather than whoever happens to be in the building that night. A surgeon who hits something unexpected mid-procedure, with the best surgeons in the world reachable in that moment, makes a better call than a surgeon alone.
Nobody signs off on this who would not have to build it, and a reviewer is paid the same whether they say yes or no. If approval improved their position they would be a vendor rather than a reviewer, and your team would see it instantly.
When this is the wrong call
If your locations run the same menu the same way every day, the established industrial engineering methods are cheaper and better than this, and you should use them. They need recurrence, and you have it.
If you already carry an internal culinary and training function built for rollouts, you have this capability and the honest answer is to use it. This is built for groups in the middle, with all of the need and none of the standing infrastructure, who rebuild the capability from scratch every cycle using the same people who are also running the business.