What Should Food Cost Actually Be? An Operator's Answer

Ask ten consultants what your food cost should be and nine will say "28 to 32 percent." That answer is popular because it's safe — and it's the wrong place to start.

I've run kitchens attached to hotels, golf clubs, wedding venues, and standalone restaurants. I've seen a pizza operation thrive at 22% and a steakhouse print money at 38%. Neither was broken. Both understood the only rule that matters: food cost is one half of a conversation, and the other half is labor.

Start With Prime Cost, Not Food Cost

Prime cost is food plus labor, together, as a percentage of sales. For most full-service operations, a healthy prime cost lands between 55% and 65%. That's the number that decides whether you make money — food cost alone can't tell you.

The steakhouse runs 38% food cost because a broiler team can turn out $12,000 nights with six cooks — low labor. The pizza place runs 22% food cost but pays for skilled dough hands and a long prep day — higher labor. Same prime cost, opposite food costs, both profitable.

The Math, So the Number Doesn't Lie

Two honest warnings. First, calculate it weekly with real counts. "Purchases divided by sales" swings wildly with delivery timing and will have you chasing ghosts. Second, one week is weather; four weeks is a trend.

When the Number Is High, It's Rarely the Menu Prices

Operators instinctively reach for the price gun. In my experience, the leak is usually in one of six places: portioning drift (the 6-oz. ladle became a heaping 8), waste and spoilage nobody logs, untracked comps and staff meals, vendor price creep on your top 20 items, theft, or menu mix — guests buying more of the low-margin items than your costing assumed. Price increases mask these for a quarter; they don't fix them.

Find Your Number

Cost your ten best-selling items — actual recipes, actual current invoice prices. Weight them by sales mix. That weighted number, checked against your labor model and a 55–65% prime cost target, is what your food cost should be. If reality runs more than two points above it, you have a leak worth finding.