The operating controls that hold under a Friday night rush

That is the test. Not whether a control is good policy, whether it survives ninety minutes of a full room with one person short. Most of what gets called a control in a restaurant fails that test on the first Friday and then gets blamed on the staff.

Judge a control by what happens when it is inconvenient, because that is the only time it matters. A rule that only works on a Tuesday is a preference.

What makes a control fail exactly when you need it?

Four properties, and they show up together.

It requires a decision at the moment of pressure. "Use judgment on comps" is not a control. During a rush, judgment defaults to whatever makes the next ninety seconds easier, which is a comp.

It requires finding a manager. Any step gated on locating a person who is currently in the weeds will be skipped or approved without being read. Both outcomes are the same outcome.

It requires a computer. If the step means walking to the office and logging in, it happens after service, from memory, if at all.

It is invisible from the station. A rule on a document in a binder does not exist during service. A rule printed on the wall at eye level above the station where the work happens does.

Notice that none of those are about the quality of the rule. A well written control with any of these properties will fail, and a crude control without them will hold. Design beats wording, every time, including when the wording is perfect.

Which controls actually hold?

The ones decided in advance, built into equipment, or visible where the work happens.

Par levels and a prep list produced from a forecast. The decision about how much to make is taken before anyone is under pressure. The rush executes a plan instead of guessing.

Physical portioning. Scoops, ladles, pre portioned proteins, marked containers. The control is in the tool, so it holds at any speed with any crew.

A posted cut order. Written before service, showing who leaves in what sequence. During the rush there is no negotiation, because the decision was already made.

Ticket times visible on the line. Not a report reviewed later. A number the cooks can see while they cook. Visible measurement changes behavior without anyone managing it.

Comps and voids that require a reason code at entry. The control is inside the transaction, not attached to it. The server cannot proceed without categorizing, and you get a usable record instead of a mystery.

One named owner of the pass. Not a rule, a position. Somebody is responsible for what leaves the kitchen, and during a rush the single most valuable control is that this is unambiguous.

A short nightly count of high movement items. Ten items, at close, by the closing position. It survives because it happens after the pressure, not during it.

That list is short on purpose. A restaurant can hold maybe five or six real controls at once. Every additional one dilutes the others, and a crew that is asked to hold fifteen will hold none.

What is a control that holds actually worth?

Enough to argue about, and the argument is easier when you put a number on it.

Start with the ruler. The National Restaurant Association's Restaurant Operations Data Abstract, 2025 edition, built on financial and operating data from more than 900 restaurant operators nationwide, puts full service salaries and wages including benefits at a median of 36.5% of sales for 2024. Operators reporting a pre-tax profit ran 34.2%. Operators reporting a loss ran 42.9%. The distance between profitable and losing is 8.7 points, and a posted cut order is one of the few things that moves it without touching a wage.

Now do the comps arithmetic on a restaurant at $1.2M in sales.

Comps and voids are running at 3% of sales, which is $36,000 a year. Half of that has no reason code attached, because the current control is a manager approving something verbally during a rush. That is $18,000 you cannot categorize and therefore cannot manage.

Requiring a reason code at entry does not reduce comps by itself. It converts $18,000 of mystery into $18,000 of data. If reading that data for two months lets you remove one point of comps, that is $12,000 a year.

Then convert it into the unit that settles the argument. The Association describes a typical independent restaurant as running a pre-tax margin near 5%, so a dollar of recovered profit is worth about twenty dollars of new sales. That $12,000 is the profit equivalent of roughly $240,000 in revenue.

One POS setting change, worth more than most of what an independent restaurant spends on marketing in a year. That is the case for controls, and it is stronger than any speech about standards.

What happens when a control meets the public?

It gets tested harder than any internal process ever will.

Consider the phone during your rush. Maple, a restaurant phone platform, analysed 1.2 million calls across more than 1,000 US locations and found that 68% of all calls arrive during the lunch and dinner rushes, that reservations are the largest call type at 38%, and that 58% of reservations still originate on the phone. Revmo AI, working from 12,091 call recordings, found greeting consistency in fast casual dropping 38 points from lunch to dinner and wait times rising 31% under peak load. Both are vendor datasets. They are also the clearest available measurement of a control degrading under pressure: same greeting, same script, same people, and it falls apart when the room fills.

Then there is what happened when large chains tried to hand a public facing control to software.

McDonald's ran an automated order taking pilot with IBM across more than a hundred US drive-thrus, and confirmed in June 2024 that it was ending the partnership, with the technology switched off by late July. The failures that went viral were long tail order modifications: bacon added to ice cream, hundreds of nuggets on one order, items piling on after the customer said stop.

Taco Bell reached roughly 500 locations and, after two million AI orders, publicly slowed and rethought the deployment in 2025 following viral trolling, most famously a customer who ordered 18,000 cups of water. That failure was not misheard orders. It was adversarial input, a different design problem.

Wendy's is the one still expanding, at more than 500 locations, reporting that 86% of transactions complete without employee intervention. Those are unaudited company figures. But the interesting number is the other one: 14% of orders are deliberately routed to a human. Wendy's is not claiming full automation, and that is precisely why it is still running.

A January 2025 YouGov survey found 55% of Americans prefer a human at the drive-thru and only 4% prefer an AI chatbot.

The transferable lesson is the one this article is about. McDonald's failed on the long tail. Taco Bell failed on the adversarial tail. Neither failed because the rule was badly written. A control that meets the public needs a named failure rate, an escalation path, and a ceiling on what it accepts. The version that survives is the one honest enough to say what it does not handle.

How do you install one so it survives the first real test?

Install it into the shift, not into the binder.

Put it where the hands are. The station, the pass, the walk in door, the POS screen. If somebody has to go somewhere else to comply, they will not.

Make the default correct. The strongest controls do not ask anyone to do anything. Pre portioned product does not require compliance. A par sheet already filled out does not require initiative.

Give it to a position that is present when it matters. A control owned by someone who leaves at nine will not hold at ten.

Name what it does not cover. Every control has an edge case, and the crew will find it on night one. Decide in advance what happens then and who owns it. An undefined edge case becomes an improvised precedent.

Run it through a real Friday before you decide it works. A Tuesday pilot proves nothing. Stand there, watch it break, and fix the break rather than the person.

The last part is the one owners skip. You install the control, the first busy night goes badly, and the conclusion is that the team did not follow it. Usually the team hit a genuine conflict between the control and the guest, chose the guest, and nobody built a path for both.

New hires are where this compounds. Whatever the line does under pressure in someone's first two weeks becomes their permanent definition of the job, which makes the first two weeks of a kitchen hire the cheapest time to install a control and the most expensive time to skip one.

Where does control design run out?

Three places, and each one is a reason to stop building and go fix something else.

When the control is competing with revenue. Given that most reservation calls arrive inside the rushes, a rule that keeps everyone on the floor and nobody on the phone is holding a labor number by giving away a party of eight. Controls have to be priced against what they prevent you doing, not only against what they save.

When the pressure is not the problem. If your Friday is genuinely under control and the money is going somewhere else, adding controls is motion. The Association estimates total expenses for an average restaurant jumped 36% between 2019 and 2026 and reports that 42% of operators said their restaurant was not profitable in 2025. Plenty of restaurants are losing on rent, price, or volume, and no control on the pass reaches those.

When there is nobody to hold it. A control needs a present owner. Installed under an absent owner, with no local ownership, it decays quietly and you find out a quarter later.

What to do this week

List the controls you believe you have. Most operators can name eight or ten. Then mark which of them require a decision, a manager, a computer, or a document.

Whatever is left is what you actually have on a Friday. That number is usually smaller than expected and it is the honest starting point.

Pick one item from the failing list and convert it into something physical, posted, or built into the transaction. One conversion, done properly, is worth more than a policy rewrite.

Price it before you install it. Take what it should recover in a year, then multiply by twenty to see what it would take in sales to earn the same money. That is the sentence that gets a manager to care.

Then work a full Friday close in the position that owns it. Do not manage, watch. You will see the exact moment where the control and the room conflict, and that moment is the design problem.

Questions about what holds on your busiest night?

Email me at eric@seod.com with the list of controls you think you have. I will tell you which ones will survive a Friday and which are policies wearing a control's uniform. Usually about half.

Sixteen years running multi unit restaurants and more than $54M in annual P&L, most of it learned standing in the pass at seven on a Friday watching a good idea fail in real time. The daily habit that supports all of it is knowing the numbers an operator should see before opening, and the monthly one is a review habit that survives a busy month.

There is also more on running the operation here.