Restaurant sales forecasting: how to predict next week without guessing

Most restaurants schedule, prep, and order against a guess. The manager writing Thursday’s prep sheet is already doing a sales forecast — they just never write the number down, so it can’t be checked, and it never gets better. Restaurant sales forecasting sounds like head-office work, but the useful version is a 20-minute weekly habit: project next week’s sales day by day, split each day by daypart, adjust for what you know is coming, then let that number drive the prep list and the schedule. This guide walks through the whole routine, including the step most guides skip — measuring how wrong you were, so the forecast improves every week.

What a forecast actually buys you

A written forecast turns three weekly arguments into arithmetic. How many servers on Friday stops being a debate about who worked hardest last week and becomes forecasted sales times your target labor percentage. How much chicken to prep stops being “what we did last time” and becomes forecasted covers times the item’s share of the mix. Whether you can afford the extra dishwasher becomes a number you can point at. The forecast also disciplines ordering — vendors deliver against next week’s expected volume, not last month’s habit — and it feeds staff scheduling directly, because a schedule built without a sales number is just a list of people who like working Fridays.

The trailing four-week method (start here)

The workhorse of restaurant sales forecasting is almost embarrassingly simple: for each day of next week, average the same weekday over the last four weeks. Next Friday’s baseline is the mean of your last four Fridays. Same-weekday matters — a Friday and a Tuesday are different restaurants — and four weeks is enough to smooth one odd day without burying a real trend.

• Pull daily net sales for the last four weeks from your POS. Net, not gross — comps and refunds are not demand.
• Average the four values for each weekday. That is your baseline for the matching day next week.
• Drop an outlier only if you can name its cause — a buyout, a blizzard, a closed patio — and average the remaining three. Never drop a number just because you don’t like it.
• If the four weeks slope steadily in one direction, weight recent weeks heavier (for example 40/30/20/10) instead of using a flat average, so the forecast follows the trend rather than trailing it.

That is the whole engine. Everything after this is refinement — and the refinements only pay off once the weekly habit exists.

Split it by daypart, not just by day

A single day number hides the shape of the day, and the shape is what you actually staff and prep against. A $6,000 Saturday that is one-third lunch is a different operation from a $6,000 Saturday that is all dinner. Compute each daypart’s average share of the day over the same four weeks, then apply those shares to the day forecast. Do the same for channel where it matters: a night that is heavy on direct online ordering and delivery loads the kitchen and the pack station very differently from a full dining room at identical sales.

Adjusting for events, weather, and holidays

The baseline assumes next week looks like the last four. It usually does — until it doesn’t. Layer adjustments on top, and keep them explicit: write “plus 15% — home game” next to the number rather than silently padding it, because an adjustment you can see is an adjustment you can grade later. The usual suspects: local events and games, weather that opens or closes a patio, holidays that move a weekday’s behavior (a Thursday before a long weekend trades like a Friday), school calendars, and your own promotions. If you run more than one location, adjust per location — a street festival that lifts one store can starve the one three blocks away, which is why multi-location operators forecast store by store and roll up, never the reverse.

Forecasting sales for a new restaurant

With no history, you forecast from capacity instead: seats, times realistic turns per daypart, times average check, discounted by a ramp-up assumption for the first months. Sanity-check the result against comparable restaurants in your market and treat every version as a range, not a promise — pre-opening forecasts exist to size the loan and the labor plan, not to predict a specific Tuesday. The honest advice is to hold the capacity model loosely and switch to the trailing four-week method the moment you have four real weeks of data, because even a short run of actual history beats a well-argued projection. Our guide to choosing a POS for a new restaurant covers setting up the reporting so that history is clean from day one.

Turn the forecast into prep and labor

A forecast that stops at a dollar figure is trivia. The value is in the conversions:

The prep conversion is where forecasting quietly kills waste: prep to the forecast plus a small safety margin instead of prepping to fear, and the walk-in stops eating your margin. Prep lists and par levels covers that half of the routine in detail.

Track your forecast error — the step everyone skips

Every Monday, put last week’s forecast next to last week’s actuals and compute the miss per day as a percentage. For an independent with four clean weeks of history, landing within about ten percent on most days is a realistic working standard; dayparts will be noisier than full days. The point is not grading yourself — it is pattern-finding. Sustained misses in one direction mean a stale baseline; misses concentrated on adjusted days mean your event bumps are too optimistic or too timid. Managers who do this for a month develop an almost unfair feel for their own building, because they are the only people in the market whose guesses get corrected weekly.

One practical note: this whole routine is a 20-minute job when daily sales, daypart splits, covers, and labor live in one system, and a Sunday-night spreadsheet archaeology project when they don’t. Clean, connected history is the quiet argument for an all-in-one platform — worth keeping in mind when you weigh up POS pricing.