Forecasting covers from the history already in your POS

Every scheduling and prep decision in a restaurant rests on one guess: how busy is it going to be.

Most kitchens make that guess from last week. Some make it from a feeling. A few pay for software that makes it for them and then quietly stop believing it after the first holiday it gets wrong.

There is a middle option, and you already own it. Your POS has been writing down how busy you were, by the hour, for as long as you have had it.

Last week is the wrong week

Last week is the easiest comparison and usually the worst one.

A Thursday in the second week of the month is not the Thursday before it. Payday lands differently. A school calendar moves. In the valley, a Thursday in late September is a different animal from one in July, because it is not 112 outside and people want to sit on a patio again.

The better comparison is the same week last year, and the one before that. Not averaged. Side by side.

Why three weeks beat an average

Take the last three times this week happened and put the covers next to each other, by daypart.

An average of 310, 295 and 460 is 355, and that number describes a Thursday that never happened. The three numbers tell you something useful instead: two normal weeks and one where something unusual went on. A GM looking at those three can usually name what it was in about four seconds.

That is the whole method. It is not clever. It is a person who knows the store reading three real weeks instead of one invented one.

If you already pulled six weeks of exports and found them useless, this is why: six weeks never contains the same week last year. You need a year, exported by day, by daypart, by store.

The list your history cannot keep

Here is the part that makes it work, and it is not in any system you own.

Write down the things that moved covers that your POS only saw the result of. The tournament at the park. The road closure. The week the building next door was a film set. Monsoon evenings when the sky turns and the patio empties in ten minutes. The first week snowbirds come back.

Keep it as a calendar, by store. One line each. Date, what happened, roughly what it did to the night.

Six months in, that list is worth more than the forecast. It is the difference between a manager who knows why last October was strange and a new manager who schedules against it as if it were normal.

What the number is for

A cover forecast is not trying to be right. It is trying to be useful to two people: whoever writes next week's schedule, and whoever writes tomorrow's prep list.

Both of them will adjust it. That is correct. The forecast is a starting point with the arithmetic already done, so the argument happens at four in the afternoon over a real draft instead of at nine at night over a blank grid.

Check it on Monday

Write the forecast down before the week starts. On Monday, put the actual covers next to it.

Do not grade it. Look at where it missed and ask why. Almost every miss falls into one of three buckets: something on the exceptions list you forgot to check, something that belongs on the list and is not there yet, or a trend that has genuinely changed and last year no longer describes this year.

The first two fix themselves. The third is the one worth a conversation, because it usually means a menu change, a competitor opening, or a daypart that has quietly moved.

Start with one store and one daypart

Not the group. Pick your busiest store and its most expensive daypart to get wrong, which is usually dinner.

Four weeks of forecast-then-check will tell you whether this is worth rolling out. More usefully, it will produce the first version of the exceptions calendar, and that is the part you keep when you change POS, change scheduling software, or change GMs.

---

Not sure forecasting is where your week is going? Take the free check — eight questions, no email, and it names the two things worth fixing first in your building.