Weekly COGS Variance Reporting: What to Actually Do With the Number

Every restaurant accounting package can produce a weekly COGS variance report. Almost none of the operators we meet do anything with it. The report lands in the manager's inbox on Tuesday afternoon, somebody glances at it, somebody else files it, and by Thursday nobody remembers what the number was.

The variance is one of the highest-leverage signals on the P&L. It is the difference between what your menu mix says you should have spent on food and what you actually spent. A 1.5-point variance at a $2.4M restaurant is roughly $36,000 a year. A 3-point variance at the same restaurant is $72,000. That is one operator salary or three full kitchen renovations every five years, depending on how you read it.

This article is the operating rhythm we install when an operations or financial engineering engagement starts. The mechanics are simple. The discipline is not.

What the variance is actually telling you

Variance is theoretical food cost minus actual food cost. Theoretical comes from the POS — your mix multiplied by your recipe cost. Actual comes from inventory — opening inventory plus purchases minus closing inventory, divided by sales.

When the two numbers diverge, exactly four things can be true:

1. Your recipes are wrong (the theoretical is incorrect)
2. Your inventory count is wrong (the actual is incorrect)
3. Your portion control is broken (real waste exists)
4. Something is leaving the building outside the till (theft or unrecorded comps)

Variance only matters when you can rule the first two out. That is the work most operators skip — and the reason most weekly variance reports are noise.

The first 60 days of every COGS variance program are spent fixing the inputs. The variance number is meaningless until the recipe deck and the count cadence are clean.

Recipe accuracy

If your POS has a recipe attached to every menu item, the theoretical cost is only as good as the recipes. The two most common recipe errors:

• Yield assumptions. A protein costed at 100% yield when the actual yield off the trim is 78%. This shows up as a permanent 2-point under-report on theoretical cost.
• Stale costs. A recipe built three years ago when chicken breast was $2.40/lb, still in the system at $2.40/lb when the current invoice price is $4.10/lb.

Fixing the recipe deck is 90% of the work of getting a clean variance. Most independent DMV operators we audit have at least 30% of their menu items running on stale or inaccurate recipes. Until those are fixed, the variance number says nothing.

Inventory accuracy

The actual side of the variance is only as good as the count. Three rules:

• Count the same way every week, by the same person, on the same day
• Count at the SKU level for the top 20 cost items; bucket-count the rest
• Sheet-to-shelf order matters — count in the order the shelf is organized, not the order the spreadsheet was built

If two different managers count the walk-in, the variance will swing 1.5–2 points week to week with no underlying change. That noise destroys the signal.

The weekly rhythm

Once recipes and counts are clean, the variance report becomes a Tuesday-morning decision tool. The rhythm has four steps.

Tuesday: Read the variance

The report should land by 10am Tuesday with last week's actual COGS, theoretical COGS, variance percent, and variance dollars. The dollar number matters more than the percent — a 2% variance at a small concept and a 2% variance at a big one are very different conversations.

Read the variance against a four-week trailing average, not against zero. A clean operation has a 0.8–1.5% variance baseline that never goes to zero. The week-over-week comparison matters: when this week is 1.0% higher than the four-week trail, that is a signal worth chasing.

Tuesday afternoon: Decompose by category

The variance report should break out by COGS category — proteins, dairy, produce, dry goods, beverage. Variance almost never spreads evenly. A 2.4% total variance is usually one category contributing 4% and the rest contributing 0.5%. The diagnostic work is finding the category, not chasing the total.

The category-level decomposition only works if your purchasing GL is mapped cleanly. If invoices are dumped into one "food cost" bucket, you cannot decompose. That mapping work is one of the highest-ROI line items in any data integration engagement we run.

Wednesday: Run the line-item walk

For the category driving the variance, walk the line items. Two questions:

• Which SKUs in that category had a unit-cost change last week (real cost inflation)?
• Which SKUs in that category had a usage spike against mix (waste, over-portion, or theft)?

The first question is invoice work. The second is theoretical-versus-actual usage. Most POS systems can produce an "ideal versus actual usage" report at the SKU level. Few operators have ever asked for one.

A 4-point variance in proteins almost always decomposes to one or two SKUs. Find them, and the conversation with the kitchen lead becomes a real conversation instead of a generic "we need to do better on food cost."

Thursday: Set the corrective action

Corrective actions should be specific, dated, and assignable. "Watch food cost" is not a corrective action. "Re-portion the bowl protein from 6oz to 5.5oz starting Monday, validate with the kitchen lead, and reassess the variance in two weeks" is.

The list of corrective actions over the trailing eight weeks is the most useful single document the management team can keep. It records what was tried, what worked, and what kept reappearing. That document is also the first thing we ask to see when an engagement starts — it tells us more about the operation than three weeks of P&L analysis.

Common failure modes

Treating variance as a single number

The total variance line is the wrong place to start the conversation. Operators who only look at the total either feel fine when the categories underneath are masking each other, or feel panicked when one category is doing all the damage and the rest are clean.

Comparing to budget instead of theoretical

Budget variance and COGS variance are different reports. Budget variance compares actual spend to your forecast. COGS variance compares actual spend to what your real mix should have cost. Operators conflate the two and end up congratulating themselves for hitting budget while running 3 points of structural waste.

Reporting monthly

The variance is a weekly tool. A monthly variance is a record, not a signal — by the time the month closes, the waste is sunk and the corrective opportunity is gone. The full P&L lies post lays out why month-end reporting is the wrong cadence for any operating-control number.

Letting variance drift without recomputing

Recipe costs need to be refreshed every 90 days. Every 90 days, run the top 30 SKUs against current vendor prices and update the recipe deck. If you do this rigorously, the variance signal stays clean. If you skip it, the variance number quietly stops meaning anything within six months.

The compounding effect of a real variance discipline

Operators who install a weekly variance discipline and hold it for 12 months typically pull 1.5–2.5 points out of COGS. That is true even when nothing about the menu, the vendors, or the recipes changes from the outside — the discipline alone, applied consistently, finds enough waste and drift to move the line.

The reason it compounds is that the conversation changes. Once the kitchen knows that Tuesday morning will produce a category-level variance number and Wednesday afternoon will produce a SKU-level walk, the day-to-day decisions change. Portions hold. Receiving tightens. The "I'll deal with it later" instinct on a vendor short-ship goes away because everyone knows the short-ship will show up in the variance.

This is the difference between a financial engineering discipline and a cost-cutting exercise. Cost-cutting is one-time. A variance rhythm is permanent — and the margin it produces compounds for years.

A practical first 30 days

If you want to install this rhythm without a consultant, do this in order:

Week 1: Audit the top 30 SKUs by spend. Validate that each one has a current invoice price loaded into the recipe deck. Refresh anything older than 60 days.

Week 2: Pick one designated inventory counter. Build a sheet-to-shelf count sheet for them. Count once with that sheet, then audit the count by re-counting two random sections yourself.

Week 3: Run the first clean variance report. Decompose by category. Do not act on it — just read it. The first clean variance is calibration, not action.

Week 4: Run the second clean variance report. Compare to week three. Begin the Tuesday→Wednesday→Thursday rhythm.

By week eight, the variance is part of the operating rhythm and not a "report we generate." That is the goal. The number itself is not the deliverable — the rhythm around the number is the deliverable.

If you want help building the recipe deck refresh or wiring a clean category-level variance report off your POS, book a discovery call. The first call is always free and we will tell you whether the rhythm we run is worth installing at your specific operation.