The restaurant metrics that actually matter: a KPI guide for owners

You can't manage what you don't measure — but most operators fail the opposite way, drowning in dashboards full of numbers that don't drive a single decision, while the two or three that matter go unwatched until the monthly P&L arrives too late to act on. This is the short list: the metrics that actually change what you do on Monday, what "good" looks like for each, and the cadence that catches a problem while you can still fix it.

The financial vitals

These are the numbers a lender, a buyer, and a sane operator look at first. Benchmarks are as a percentage of sales:

Prime cost is the master number. Food and labor trade off against each other, so managing either alone hides problems — I made that case in the food cost and labor cost guides. Above 65%, there's usually too little left for rent and profit.

One honest flag on labor: the "target" is 25–35%, but the real-world full-service median is around 36.5%, and profitable operators hold it near 34.2%. So the benchmark isn't a participation trophy — hitting it is the work, not the baseline.

The operational vitals

These you watch during and right after service, not at month-end:

• Sales per labor hour (SPLH) — sales ÷ labor hours worked. A productivity gauge with no universal target; trend it against your own baseline by daypart.
• RevPASH — revenue per available seat-hour (revenue ÷ seats ÷ hours open). Combines check size and seat utilization; again, concept-specific, so trend it.
• Average check — sales ÷ covers. Track by daypart and push it up with menu engineering and upselling.
• Table turnover — parties served ÷ tables per service. Rough ranges: fast casual 4–5 turns at peak, family dining ~3 at dinner, fine dining 2–3.
• Void / comp / discount % — comps typically run 3–5% of sales, voids should stay under 1–2%, and industry loss/fraud averages around 4% of sales. These are your theft-and-leak early-warning lights.
• Repeat-guest rate — a "good" repeat rate is roughly 30–40%; it's the retention number the whole loyalty argument rests on.

A word on the context-dependent ones — SPLH, RevPASH, average check, break-even: stop looking for a universal benchmark for these. There isn't one. Their value is the trend line against your own history, not a number you copy from a blog. Anyone who hands you a "good RevPASH" figure without knowing your concept is guessing.

The cadence that actually catches problems

Which metric you watch matters less than how often. The rhythm that works:

• Daily (about five minutes): gross sales, covers, labor hours clocked so far, and what got 86'd (it changes tomorrow's prep). A glance, not a meeting.
• Weekly (about thirty minutes, same morning every week): the flash report — prime cost and its components, food and labor %, average check, voids and comps, top and bottom menu items. Weekly is the cadence that catches drift before the month is gone. A prime-cost variance over 3% of target should trigger a fix within 24 hours, not a shrug at month-end.
• Monthly: the full P&L, cash trend, inventory variance, guest-review themes. The scoreboard — but by the time it lands, the month's cash is already spent, which is why the weekly flash exists.

I break down the monthly statement itself in how to read a restaurant P&L — this post is the operational layer that feeds it.

What your systems should do about this

Here's the test: if pulling last week's prime cost, labor %, and comp rate takes your manager an afternoon of spreadsheet surgery, you'll stop doing it — and the drift wins. The data already lives in your POS; the only question is whether it's assembled for you. A reporting layer that produces the weekly flash automatically, shows labor as a live percentage during service, and flags variance against target turns these metrics from a month-end autopsy into a Monday decision. The analytics guide covers how to pick the dashboard that does it.

Disclosure: I work at Katalyst, and we build that reporting, so read the bias in. But the discipline is vendor-neutral: watch prime cost above all, trend the context-dependent numbers against yourself, and run a weekly flash so problems surface in days, not months. The operators still standing in a hard year are almost always the ones who saw the number move while they could still do something about it.

Related Katalyst products

• See Katalyst analytics