Pour cost tells you what the product you used cost against what you sold. It does not tell you whether everything you used was actually sold. That second question is the variance report, and most bars never run it. The healthy range is 1 to 2% of inventory. Industry reporting puts many bars at 5 to 25% without knowing it.
Two numbers, two jobs
Pour cost is a ratio: product cost over beverage revenue. It moves slowly, it blends everything together, and a bar can hold a respectable 22% while a case of gin walks out the door, because pour cost only sees what the POS recorded. If a drink was over-poured, spilled, comped off the record, or never rung in, pour cost does not flag it. It just drifts up half a point and nobody can say why.
Variance is a count: what physically left the shelves against what the POS says was sold. It is the audit behind the ratio. I covered the ratio in Pour Cost 101. This is the other half.
The formula
Two lines, run per product, in a consistent unit (milliliters, not bottles, since bottle sizes vary):
- Usage = (previous count + deliveries) − current count
- Variance = usage − POS sales
Worked example. Last week's count showed 40 bottles of the house gin. Ten more arrived. This week's count shows 32. Usage is 18 bottles. The POS sold the equivalent of 15. Variance: 3 bottles.
Now the money. A 700 ml bottle at a 45 ml pour is 15 serves. At $14 a cocktail, each bottle should produce about $210 in revenue. Three unaccounted bottles is $630 in a week, on one SKU. Held for a year, that is over $32,000 in revenue the product was bought for and never generated.
Benchmarks
From current industry reporting (BarSociety's July 2026 guide):
- Below 1% of inventory: excellent
- 1 to 2%: healthy, the working target
- 2 to 4%: needs review
- Above 5%: investigate now
For scale: bars without dedicated variance and waste controls lose an estimated 15 to 20% of beverage inventory to shrinkage, and the National Retail Federation's latest benchmark put retail shrink at 1.68% of revenue, the highest in over a decade. Alcohol is one of the highest-margin products in the building. Every unexplained milliliter costs more here than anywhere else in the operation.
Where the gap comes from
Four buckets, in rough order of frequency:
- Counting errors. The most common cause of a sudden spike. Count again before concluding anything.
- Over-pouring and spills. The highest-volume operational cause. A bartender free-pouring 60 ml into a 45 ml spec gives away a bottle every 30 drinks.
- Paper problems. A delivery that never got logged, a recipe mapped wrong in the POS, a transfer to the kitchen nobody recorded.
- Unrecorded product. Comps that were poured but never rung, staff drinks, theft. This is the bucket everyone jumps to first. Check the other three before you get here.
Direction matters too. Negative variance (product missing against sales) is the loss direction. Positive variance (sales higher than usage) usually means a recordkeeping error, a missing invoice or a miscount, and it deserves the same investigation. It is not good news, it is a broken number.
The weekly routine
Monthly counts hide problems inside 30 days of noise. Weekly is the working cadence, per the same actual-versus-theoretical logic chains run on food cost. Same day each week, same person owning it, same units:
- Count, compute usage, pull POS sales, compute variance per product
- Recount anything that spiked before investigating it
- Check invoices and comp records for the products still off
- If one premium SKU repeats, run a before-and-after count across a single shift to isolate it
- Chase patterns, not incidents. A one-off is a miscount. The same SKU three weeks running is a process problem
What one point is worth
Bar at $60,000 a month in beverage sales, 22% pour cost
Monthly usage at cost: about $13,200
4% variance: $528 of product a month unaccounted
At the same pour cost, that product should have been $2,400 in revenue
Held for a year: roughly $29,000, recovered by a weekly count
The variance report is one page of arithmetic a week. What it usually reveals is not a thief, it is a training gap, a paper gap, or a spec nobody is holding. Fixing those is program work: specs, batch logs, pars, and a POS that matches what the bar actually pours. That is the audit we run inside a program refresh. If your pour cost has been drifting and nobody can say where, fifteen minutes will usually locate it.