How to calculate PAR levels

There are three PAR formulas in circulation. If you've read any other article on this topic, you've seen all of them. The ones that look different are usually the same formula with variables renamed. Here's what they actually say, worked through with numbers you'd see in a real bagel shop or coffee bar.

Formula 1 — Delivery cycle

PAR = (weekly usage + safety stock) / deliveries per week

Most cited version. If you use 60 units a week, keep 15 units as buffer, and get two deliveries a week:

PAR = (60 + 15) / 2 = 37.5

Round to 38.

This formula answers: how much should be on the shelf right after each delivery?

When it's useful: multi-delivery-per-week setups where you care about the post-delivery state. Bars, high-volume coffee shops, anywhere the vendor drops twice or more weekly.

When it's wrong: if your PAR sheet is for counting, not for post-delivery. Most QSR operators count once per cycle (weekly) and order based on that count. You don't want a post-delivery PAR — you want a pre-delivery PAR.

PAR = (average daily usage × days of coverage) + safety stock

Days of coverage = days between orders + supplier lead time.

Example: a bagel shop orders oat milk once a week. Supplier ships next-day. Daily usage averages 6 half-gallons over the last 30 days. Safety stock: the operator decides on 4 units — enough to cover a weekend bump.

PAR = (6 × 8) + 4 = 52

52 half-gallons of oat milk on the shelf post-count = enough to cover 8 days of use with a 4-unit buffer.

Why it's the one we recommend: it's a PAR you can actually count against. Count on Monday, need to cover until next Tuesday's delivery + Wednesday's first shift = 8 days. PAR says 52; you counted 34; order 18; round to cases.

When it's useful: every weekly-ordering QSR. This is the formula to use.

Formula 3 — Demand-based

PAR = forecasted peak demand × safety multiplier

Used by enterprise inventory systems that run demand forecasts. The number comes out of a model, not arithmetic.

When it's useful: large chains with enough data to fit a real forecast. The model can catch day-of-week, seasonality, promo effects, weather — things a hand-calculated PAR can't.

When it's wrong: for a single-store operator with six months of data and no time to maintain a model, this is overkill. You'd end up with an opaque number you can't defend to your team.

Which formula to pick

For most single-store QSR operators: Formula 2. It matches a weekly count cadence, it uses inputs you can pull from any POS, and it produces a number you can explain on a whiteboard.

If you're multi-unit with a data team and weird ordering cadences, consider Formula 3 — but that's not a DIY exercise. Use a real inventory system or have someone build the forecast.

The safety stock problem

Every PAR formula has a safety stock term. Most articles tell you "25% of average usage" and move on. This is lazy.

25% is a terrible default for every item the same way. A stable, fast-moving item — plain bagels on a Tuesday — doesn't need 25%. Your worst Tuesday is probably 10% off your average Tuesday. A volatile item — a new drink on a weekend — might need 50% because you legitimately don't know what demand looks like yet.

Better approach: look at the last 30 days of daily sales for the item. Compute the standard deviation. Safety stock = 1× standard deviation, sometimes 2× for high-volatility items.

If that sounds like a lot of math: your POS can usually show you a sales-per-day chart per item. Look at the worst three days in the last 30. How far below average were they? That's the buffer you need.

Example

Oat milk sales for the last 30 days — daily units sold:

Mon: 5, 6, 4, 5, 5
Tue: 5, 6, 7, 6, 5
Wed: 6, 7, 6, 7, 7
Thu: 7, 8, 6, 8, 7
Fri: 8, 9, 7, 8, 9
Sat: 10, 12, 9, 11, 10
Sun: 8, 9, 7, 8, 9

Average daily usage: ~7.3. Worst day: a Monday at 4 (-3.3 below average). Best day: a Saturday at 12 (+4.7 above average). A reasonable safety stock is 4–5 units.

PAR (using Formula 2, 8 days of coverage): (7.3 × 8) + 4 = 62.4 → round to 63.

Notice what this captures that "average × 8 days + 25%" misses: the Saturday spike. 25% of 58.4 = 15, so a blanket PAR would be 73 — 10 units too high. You'd throw out 4–5 oat milks every week on average.

What you actually need to run the math

A non-negotiable: daily sales per item for at least 30 days. Not weekly totals, not monthly summaries — daily. Without daily data, you can't see day-of-week patterns, and day-of-week is the biggest driver of within-week volatility for most QSRs.

Every POS has the underlying data. On Toast, you pull it from Item Selection Details and Item Modifier Selection Details (transaction-level exports under Reports → Sales). See Item Selection and Modifier Selection exports for the specifics.

When this math breaks

The manual version vs the automated version

If you do this by hand, it's a 30-minute-per-item exercise. For a 40-item menu, that's a full day, and you'll do it maybe once a quarter. By the time you finish item 40, item 1's PAR is already stale.

Par Inventory runs Formula 2 for every item every time you click a button, using the real daily sales history from your Toast POS. The math is exactly the same math in this article — there's no secret sauce. The value is in not spending a day doing it and not letting the PARs drift for three months between recomputes.

Published 2026-04-21← All articles