By Matt Wampler, CEO of ClearCOGS
The short version. A labor percentage is a weekly average. It can read 20 percent while the same day holds a dead afternoon block at $31 of sales per labor hour and a dinner block at $135. The number that tells you whether the day was staffed well is sales per labor hour by block, read against the shape the forecast predicted. The number that tells you what it cost is turnover: in September 2026, the quits rate in accommodation and food services jumped to 5.0 percent, from 3.6 percent in August.
This post is about groups that already hit their labor target and still have a scheduling problem they cannot see in their own report.
Plenty of well-run restaurant groups hold labor to a target in the low 20s and hit it most weeks. In many of them, someone in finance builds a daily labor report by hand, pulling worked hours from the point of sale and scheduled hours from the scheduling tool, then emailing it to every store. And many of those groups will tell you, fairly, that their managers know their own sales better than any program, because they know when the local events are.
None of that is wrong. It leads to a fair question: when the percentage is already good, what is there to fix?
The honest answer is that the percentage is the wrong instrument for the problem. Here is why.
What does a labor percentage actually measure?
Labor percentage is total labor dollars divided by total sales over a period, usually a week. It is a good number for finance. It tells you whether the week paid for itself.
It tells you nothing about the shape of the day. Two very different Tuesdays can produce the same 20 percent. On one, the crew is matched to the guests all day. On the other, three people stand around from 1 p.m. to 5 p.m. and then two people drown from 5 p.m. to 7 p.m. The math averages the boredom and the panic together and reports that everything is fine.
Here is a labeled hypothetical to show it. Assume a $4,000 day, 50 labor hours, and a blended wage of $16 an hour.
| Two-hour block | Sales | Labor hours | Sales per labor hour |
|---|---|---|---|
| 11 a.m. to 1 p.m. | $1,100 | 14 | $79 |
| 1 p.m. to 3 p.m. | $400 | 10 | $40 |
| 3 p.m. to 5 p.m. | $250 | 8 | $31 |
| 5 p.m. to 7 p.m. | $1,350 | 10 | $135 |
| 7 p.m. to 9 p.m. | $900 | 8 | $113 |
| Day | $4,000 | 50 | Labor: $800, or 20.0% |
Hypothetical figures for illustration. Sales per labor hour = block sales divided by block labor hours. Labor dollars = 50 hours x $16. Day labor percentage = $800 / $4,000.
The day hits 20 percent. The report goes out in the morning and says good job. Inside that day, sales per labor hour swung from $31 to $135, more than a four-to-one spread. The afternoon crew cost money for very little, and the dinner crew carried a load that, repeated often enough, is how you lose a line cook.
Osa Osarenkhoe, our CTO, puts it this way in demos: target the labor percentage, but read sales per labor hour as a measure of stress. One hour at $1,000 per labor hour costs you employees. A sustained stretch at $50 costs you money and bores the people you kept. The job is a balanced day that still lands at the target.
Paraphrased from how our CTO describes the labor view to prospects.
Why does the daily report not catch this?
Because the report is built from two systems that do not talk to each other, by one person, every morning. In many groups, worked hours come out of the point of sale and scheduled hours come out of the scheduling tool. Someone in finance copies both into a spreadsheet, compares them, and sends the result. It becomes a routine, and the people who run it are usually good at it. What they tend to want is a way for managers to get the information when they need it, without a person in the middle.
That setup has two limits, and neither is about effort.
First, it is a morning-after document. By the time the manager reads it, yesterday’s dead afternoon and slammed dinner are history. The report can only grade the day, so the manager learns the pattern slowly, one autopsy at a time.
Second, it compares schedule to actual at the day level, because that is what fits in a spreadsheet row. The block-level view, where the stress lives, is not in the row. So the report confirms the percentage and misses the shape.
We saw the same mechanism last month in a different part of the operation: a senior cook retyping a count into a vendor portal, twice a week, because two systems had no path between them. We wrote it up as a 90-minute teardown of one manual order cycle. The labor version is quieter, because the person doing it is salaried and skilled, but it is the same shape: a human standing in for an integration.
What does a badly shaped day cost, beyond the hours?
The usual answer is “nothing, we hit 20 percent.” The data say otherwise, in two places.
Turnover, September 2026
5.0%
Quits rate in accommodation and food services, up from 3.6 percent in August and 3.4 percent a year earlier. Separations (883,000) ran slightly ahead of hires (879,000). Source: BLS JOLTS, seasonally adjusted, preliminary.
Cancelled shifts and wellbeing
43% vs 26%
Share of hourly food-service and retail workers reporting they are unhappy, with and without a cancelled shift in the past month. Source: Schneider and Harknett, American Sociological Review, 2019; 27,792 workers at 80 large companies.
Start with turnover. The Bureau of Labor Statistics’ September 2026 Job Openings and Labor Turnover report, released November 3, shows quits in accommodation and food services rising by 197,000 in one month, to 719,000. The quits rate went from 3.6 percent to 5.0 percent. BLS describes the quits rate as a measure of workers’ willingness or ability to leave. In the same month the industry hired 879,000 people and lost 883,000. One month is one month, and the figures are preliminary. But a sector that is replacing staff as fast as it hires them is not a sector where a $135 dinner hour is free.
Now the schedule itself. Daniel Schneider of UC Berkeley and Kristen Harknett of UCSF surveyed 27,792 hourly workers at 80 of the largest U.S. food-service and retail companies between June 2016 and October 2017. They describe the scheduling practices they studied as designed to hold down labor cost by matching staffing closely to demand, which is exactly what a manager does when the afternoon is dead and someone gets sent home. In their sample, 14 percent of workers had a shift cancelled in the past month and 26 percent had worked an on-call shift. Workers who had a shift cancelled were far more likely to report psychological distress (64 percent against under half), poor sleep (82 percent against 72 percent) and unhappiness (43 percent against 26 percent). Their headline finding is the one operators should sit with: schedule stability was more strongly related to worker wellbeing than the hourly wage was.
Put the two together. The cut-at-2 p.m., call-in-at-5 p.m. habit that protects the weekly percentage is the practice the Shift study measured. And the people it lands on are quitting at the highest monthly rate in the release’s 2026 columns.
Can a forecast fix a day that a good manager already runs?
Here is where a finance team and I would usually disagree, and they have a point I want to state fairly first.
Their view: the managers know the market, know the events, and have run these stores for years. A program will not out-guess them on Saturday’s sales. They are right about the total. A good GM’s gut for the day’s number is real information.
Where the gut runs out is the shape. A manager can tell you Tuesday will do about $4,000. Asking the same manager, on Thursday, how the $4,000 will spread across ten half-hour blocks the following Tuesday is a different question, and the answer tends to be “about like last week.” A model that has seen 15-minute sales for two years, plus the weather and the event calendar, answers that question in a way a person cannot, because a person does not carry a hundred Tuesdays at that resolution. That is the whole contribution: the shape, days in advance, so the schedule is written to it rather than corrected on the day.
When the schedule is written to the shape, the manager stops cutting at 2 p.m. and calling in at 5 p.m., because the 2 p.m. shift was never scheduled and the 5 p.m. one was. The day lands near the same percentage. The cancelled shifts and on-call shifts, the ones the Shift study measured, shrink. Alec Swope, the CFO of a 35-location sandwich franchisee, describes the same thing from the finance seat in this video case study: his GMs build schedules around when guests actually arrive and adjust them during the day, and the group’s labor has stabilized.
The budgeting version of this argument, applied to sales rather than labor, is in our post on building a 2027 sales budget store by store. Same principle: the annual number and the daily number have different jobs.
Where do finance and the GM actually disagree?
On less than it looks. Here is the exchange as it usually runs, condensed.
Finance: The percentage is already good. It is the number on the P&L and the number the owners read. Anything that risks it is a step backward.
GM: The percentage is good because I cut people at 2 p.m. A dead afternoon makes the crew feel disposable, a slammed dinner makes them feel abused, and I am the one who gets the two-weeks’ notice.
Finance: So keep the hours the same and move them.
GM: I would, if I knew on Thursday what next Tuesday’s 5 p.m. looks like.
Where it lands: a schedule written to the forecasted shape serves both. The same labor hours are bought, at the right times, so the percentage holds. The bad hours shrink. Finance gives up the comfort of a single number as proof of good scheduling. The GM gives up the right to fix the day by hand at 2 p.m. In my experience both trades are worth making, and the GM’s is the harder one.
Three things to check before the next cut
1. Pull one week of sales and labor hours in two-hour blocks for one store. Find the highest and lowest sales per labor hour. If the spread is more than three to one, the percentage is hiding a shape problem.
2. Count the shifts cut early or called in late last month. Each one is a data point in the Shift study’s cancelled-shift and on-call columns.
3. Ask who builds the daily labor report and how long it takes. If the answer is a person and a spreadsheet, that is the integration you are missing.
If you want to see what a 15-minute forecast does to a schedule for a group your size, Let’s Talk.
What else do operators ask about labor percentage and sales per labor hour?
What is a good sales per labor hour for a restaurant?
There is no benchmark that fits every restaurant, because it depends on check average, service style and wage levels. The useful comparison is inside your own store: the spread between your best and worst blocks on the same day, and how that spread compares to what the forecast predicted.
Is labor percentage or sales per labor hour the better target?
Use both, for different jobs. Labor percentage is the target finance holds the week to. Sales per labor hour by block is the diagnostic that shows whether the schedule matched the day. In my experience, groups that only track the percentage find the shape problem late, through turnover.
Do we need any of this if our labor percentage already holds?
If your schedules go out two weeks ahead, turnover is low, and the percentage holds without midday cuts, you already have most of what a forecast would give you, and a scheduling tool with a good integration will cover the daily report. The forecast earns its keep where the percentage is good but the day is lumpy, and where someone in finance is still the integration between two systems every morning.
How far ahead should the forecast go for scheduling?
At least as far as your schedule is posted. If schedules go out two weeks ahead, the forecast has to be useful two weeks ahead, with a daily refresh so the manager can make small adjustments rather than cuts.
Disclosure and method
ClearCOGS sells restaurant forecasting software, including labor forecasting, so we have a commercial interest in operators scheduling to a forecast. The opening describes a common pattern across multi-unit operators we talk to, not a single company; it is a composite, and no operator is quoted. The hour-by-hour table is a hypothetical, labeled as such. BLS JOLTS figures are seasonally adjusted and preliminary for September 2026. The Schneider and Harknett figures are from the authors’ February 2019 research brief summarizing their American Sociological Review paper; the survey is of hourly workers at large chains and is not restaurant-only. Written November 9, 2026.
Sources
- U.S. Bureau of Labor Statistics, Job Openings and Labor Turnover, September 2026, USDL-26-1630, released November 3, 2026, Tables 2, 3 and 4: bls.gov/news.release/archives/jolts_11032026.htm
- Schneider, D. and Harknett, K., “Consequences of Routine Work-Schedule Instability for Worker Health and Well-Being,” American Sociological Review 84(1), 2019, pp. 82 to 114. Research brief (The Shift Project, February 2019): shift.hks.harvard.edu (PDF)
