Restaurant Cost Control: Where Margin Actually Leaks
Most margin doesn't disappear in one dramatic loss. It leaks in small amounts, in places month-end reports rarely look closely enough to catch.
Theoretical against actual
Every cost control conversation is really one subtraction. What you should have used, given what you sold, minus what you actually used. The first number comes from recipes; the second comes from counting. The gap between them is the entire subject, and most venues cannot calculate it because they have never written the first one down.
That gap has a small number of causes, and they are boring rather than dramatic. Nobody loses a month's margin in one event. They lose it in eight grams at a time, five hundred times a week, in places month-end reporting was never going to show.
Over-portioning, quietly
A dish specced at 180g of protein and plated at 200 is an eleven per cent overrun on the most expensive line on the plate. It is invisible on any individual plate, invisible to the guest, and invisible in a monthly food cost percentage that moved two points for reasons nobody could name.
The fix is not a scale on every station, which nobody sustains. It is knowing the theoretical usage well enough that the gap shows up, and then looking at the two or three dishes it concentrates in. Almost always it is a handful of items and one or two people, and it is a conversation rather than a policy.
Waste that never gets logged
Waste is the most under-recorded number in hospitality, and the reason is entirely practical: logging it during service is a form, and forms lose to a busy Saturday every single time. So the bin gets it and nobody knows.
Whatever system you use, the test is how many taps it takes at the moment somebody is holding something they are about to throw away. If it is more than about three, it will not be used, and the number you end up with will be a fiction that makes everything downstream of it wrong.
The reason to bother is that waste categorised by cause is directly actionable in a way the total never is. Spoiled means you are over-ordering. Prep error means a training problem. Returned means a recipe or a portion problem. Same weight, three completely different responses.
When stock should come off
A detail that sounds like a technicality and is not: at what moment does selling something reduce your stock? There are two defensible answers and they suit different kitchens.
When it starts being made
Right for kitchens where beginning the dish commits the ingredients — anything portioned, fired or assembled to order. Stock reflects reality even if the plate is later cancelled, which is honest, because the ingredients are gone.
When it is served
Right where a cancellation genuinely returns the goods — bottled drinks, packaged items, anything that goes back on the shelf. Deducting earlier would slowly understate what you have.
Pick the wrong one for your kitchen and your counts drift in a direction you cannot explain, which is worse than not counting at all, because now you have a number you believe.
A count that tells you something
Counting stock is the least popular job in the building and most of the reason is that it produces a number nobody does anything with. A count is only informative next to what the system expected you to have. The variance is the finding; the count on its own is inventory theatre.
Which changes how you should count. Counting everything monthly produces one large variance you cannot attribute to anything, because thirty days happened in between. Counting the ten most expensive and fastest-moving items weekly produces small variances you can still trace to a week, a shift or a dish.
Count what matters, often
Protein, spirits, anything with a high unit cost or a fast turn. These carry most of the money and most of the drift.
Count the rest rarely
Dry goods and low-value stock monthly is fine. Trying to count everything weekly is how counting stops happening at all.
Count at the same point in the cycle
Before a delivery, after a service — pick one and keep it. A count taken at a different point each time produces a variance that is mostly timing.
Write down what you found, not what you expected
Obvious, and the single most common way a count is quietly rendered useless.
Slow-moving stock and menu drag
Cash sitting in a storeroom is cash not doing anything, and it goes off. The specialist ingredient bought for one dish that sells twice a week is a classic — it ties up money, it occupies space, and eventually a portion of it is thrown away, which means the real cost of that dish is much higher than its recipe says.
This is where inventory and menu analysis have to be read together. A dish with a respectable margin on paper and a slow-moving exclusive ingredient behind it may be losing money once spoilage is counted, and neither report shows that on its own.
Seeing it before month-end
A monthly food cost percentage is a post-mortem. By the time it arrives, the month it describes is over and nobody can remember what happened in week two. Everything above only works if it is visible while you can still act on it.
Usage that updates as you sell
If recipes deduct ingredients when dishes sell, running low is a screen rather than a discovery at seven o'clock.
Counts that compare against expected
A count is only informative next to what the system thought you had. The variance is the finding; the count on its own is just a number.
Waste in the same place as stock
Separate systems mean nobody reconciles them. Same place means the variance already has the waste subtracted.
Ordering built from what actually moved
A shopping list generated from usage rather than from habit is the single easiest way to stop over-ordering the thing you always over-order.
Inventory covers how usage and waste are recorded here, and Reporting covers reading margin against sales once the costs are real.
Common questions
What food cost percentage should we be aiming at?
It varies too much by format to be a useful target from the outside. The number that matters is your own trend and the gap between theoretical and actual usage.
Do we have to build recipes for everything?
Start with the dishes that sell most and the ingredients that cost most. Twenty per cent of the menu usually explains most of the variance.
How often should we count?
High-value and fast-moving items weekly; everything else monthly. Counting everything every week is how counting stops happening at all.
Is waste tracking worth the effort during service?
Only if it takes seconds. Categorised waste is one of the most actionable numbers in the building, but a form nobody fills in is worth less than nothing.