Staffing to a forecast instead of to a feeling

That last step is the whole thing. A forecast nobody grades is a guess with better handwriting.

Most restaurants already have what they need. Two years of daily sales sit in the POS, and nobody has put them in a column next to a schedule.

The stakes are visible in published data. The National Restaurant Association's Restaurant Operations Data Abstract, 2025 edition, collected financial and operating data from more than 900 restaurant operators nationwide. Salaries and wages including benefits ran a median of 36.5% of sales for full service restaurants in 2024, with operators reporting a pre-tax profit at 34.2% and operators reporting a loss at 42.9%.

Nearly nine points of labor separate a profitable full service restaurant from a losing one, and a schedule built without a number in front of it is one of the main ways a restaurant travels from one end of that range to the other.

What actually goes into a restaurant forecast?

Four inputs, and only the first one is heavy lifting.

The same weekday, recent weeks. Take the last four or five of that weekday and use the middle of the range rather than the average. One outlier night pulls an average badly.

The trend. Compare those weeks to the same weeks last year and apply the direction. If you are running ahead of last year, the Tuesday forecast should reflect that rather than sitting at last Tuesday's number.

Known events. A game, a concert, a nearby closure, a holiday, road work on your street. This input lives in your head and belongs on paper, because the manager building the schedule does not know the street is closed on the fourteenth.

Weather, but only for shifts it moves. Patio and lunch driven restaurants are weather sensitive. A dinner house with no outdoor seating usually is not, and adjusting where it does not belong adds noise.

That is a forecast. Twenty minutes a week once the sheet exists, and the sheet is a spreadsheet.

A forecast is not a target. Do not forecast what you want to sell. That habit produces overstaffed shifts and a manager who learns to distrust the number.

Why does writing the number down change the schedule?

Because it forces a decision to be made before the pressure arrives, and it makes the decision reviewable afterward.

With no forecast, the schedule is a social document. Who wants hours, who is owed a good shift, who complained last week. All of that is real and none of it should be the primary input.

With a number posted, the question is concrete. We expect this volume Thursday, so this many on the floor and this many in the kitchen. The conversation about hours now has a reference point that is not the manager's mood.

It also fixes the direction of the error. Restaurants that staff by feeling are wrong the same way every week, usually overstaffed early and thin at the peak, because the schedule came from a memory of the busiest moment rather than the shape of the day.

And the forecast feeds more than labor. Prep, pars, and ordering come off the same expected volume, which is what makes usage predictable enough for a real conversation with a distributor. Vendor negotiation for independent operators turns on saying what you will use, and that starts here.

How accurate does a forecast need to be?

Close enough to make a staffing decision, which is a much lower bar than people assume.

You are not predicting sales to the dollar. You are deciding whether Thursday gets four servers or five. A forecast that lands in the right band has done its job even when the exact number is off.

Track the miss anyway. Forecast and actual, same sheet. After six weeks you will see the character of your errors, which is more useful than any single week.

What shows up:

• Consistently low on one weekday, meaning that baseline is stale and the business grew into a new pattern
• Consistently high on one daypart, usually a lunch or late night that quietly declined
• Wild misses on specific dates, almost always events you knew about and did not write down

Fixing the method is the work. A forecast wrong the same direction every Tuesday is not a bad forecast, it is an uncorrected one.

Keep the sheet small enough that a manager maintains it. Same discipline as deciding what a business dashboard should actually show, where the temptation is to add columns until nobody reads any.

What does one bad forecast actually cost?

Here is the arithmetic on a single Thursday, using numbers you can swap for your own.

You forecast $5,200 and staffed to it: five servers at 6.5 hours each, 32.5 labor hours. That plans for $160 of sales per labor hour.

The night does $4,150. Same 32.5 hours. Sales per labor hour drops to $128.

The fifth server was the marginal decision. Six and a half hours at a loaded hourly cost of $19, meaning wages plus payroll taxes plus whatever benefits you carry, is $123.50.

One night, $123.50, and nobody notices.

A forecast wrong once is weather. A forecast wrong the same direction twice a week is a method problem. Twice a week at $123.50 is $247, and across a year $12,844.

The Association puts the pre-tax margin of a typical independent restaurant near 5%. At that margin, $12,844 of profit is what a restaurant doing roughly $257,000 in annual sales produces. Forecasting high on two shifts a week is worth a quarter of a million dollars of sales, and the fix is a spreadsheet column.

Now flip it. Forecast $4,150 when the room does $5,200 and you get a thin night, long ticket times, comped desserts, and a table that does not rebook. That cost never appears on the labor line, which is exactly why operators drift toward overstaffing. One error shows up on a report and the other shows up nowhere. Write both misses down or you will only ever correct one of them.

Why can you not just use an industry number?

Because industry numbers in this category are usually averages across segments that behave nothing alike, and the most famous one in restaurant operations is not research at all.

Here is the trace, because it is worth seeing once. The claim that 62% of business calls go unanswered comes from a blog post published by 411 Locals on 18 January 2016. It says it monitored 85 businesses across 58 industries for thirty days, roughly one and a half businesses per industry, with no country stated and no raw data. The publisher sells virtual receptionist services, and the same post pivots into pricing for them. The figure is now attributed variously to Forbes, to Ruby Receptionists, and to a ServiceTitan analysis of contractor phone lines. None of those attributions hold up. A statistic with three incompatible parents is telling you something about itself.

Now look at measured data. Revmo AI, analyzing 12,091 call recordings, reports answer rates by restaurant segment: QSR 59.9%, fast casual 75.3%, full service 91%, pizza 93.1%. Vendor research, disclosed sample, and the spread is the finding.

The segment, not the industry, determines the number. Below the segment, the individual restaurant determines it again.

Apply that to forecasting. There is no useful industry figure for what your Tuesday should do or where your peak sits. Each is a property of your street, your menu, your hours and your neighbors. The only calibrated instrument is your own last five same weekdays.

If someone hands you a staffing ratio as an industry standard, ask for the sample size and the segment. That is usually the end of the conversation.

What do you do when the forecast is wrong on the night?

You need a decision rule written before the shift, not judgment invented during it.

Two triggers cover most of it. A cut trigger: if sales at a stated hour are below a stated number, one person goes home. A call in trigger: above a stated number, the person on call comes in. Both need a time, a number, and an owner.

Without those, what happens is nothing. The manager notices at eight that the room is empty, thinks about cutting, feels bad, and everyone stays. That hesitation, repeated across a year, is a real number on the labor line.

On call has to be honest to work. If somebody is holding their evening for you, tell them by a stated time and treat it as a commitment on both sides. A dishonest on call list stops functioning within a month.

The morning read closes the loop. Yesterday's forecast, yesterday's actual, and today's forecast belong in the same glance, which is part of the daily numbers an operator should see before opening.

Where does forecasting break down?

Three places, and two of them are invisible from inside the spreadsheet.

When demand arrives somewhere the POS cannot see. Your history records transactions, not interest. Maple, a company selling restaurant phone automation, analyzed 1.2 million calls across more than 1,000 US locations between December 2023 and November 2025 and found 68% of calls landing during the lunch and dinner rushes, and 58% of reservations still originating from phone calls rather than apps or online booking. Vendor research, and Maple states parts of the same report are merchant self reported. Take the measured piece: the phone peaks exactly when the floor is least able to answer it. A forecast built purely on completed sales will keep telling you your ceiling sits where you stopped being reachable.

When the baseline has drifted. A construction project, a neighbor closing, or a schedule change at the office park down the street resets the level rather than moving it. If your misses run one direction for four consecutive weeks, the baseline is stale and averaging will not fix it.

When the forecast becomes a target. Once a number is posted somebody will manage to it, and only one of the available routes involves selling more food. Grade the method, not the person.

What to do this week

Export daily sales for the last twelve months into one sheet, a row per day, with columns for weekday, sales, covers, and notes.

Write next week's forecasts. Fifteen minutes, using the middle of the last five same weekdays, adjusted for what you know.

Post the forecast where the schedule is posted, both numbers visible to the crew.

Build the schedule from the forecast, not from last week's schedule. People skip this step, and skipping it makes the rest decoration.

Next Monday, fill in actuals next to your forecasts. Do that six weeks before judging the method, and expect the first two weeks to be poor.

Questions about building your first forecast?

Email me at eric@seod.com with eight weeks of daily sales, just dates and totals in any format. I will send back a simple forecast sheet built off your own numbers, with the weekday baselines already filled in and the two days I would watch most closely.

Sixteen years running multi unit restaurants and more than $54M in annual P&L, and every location I ran that held its labor line held it because someone wrote a number down on Thursday. If reading the statement those numbers land on is still unfamiliar, reading a restaurant P&L without a finance background is the companion piece.

Forecasting gets harder with more than one location, partly because your own stores start competing for the same guests, which shows up online too in multi location SEO that cannibalizes itself.

Otherwise there is more on running the operation.