Restaurant Sales Forecast: Planning Next Week From Your Till
8 September 2026 · MidaOne
Most café owners forecast every week without calling it that. You order milk on a feeling about how Tuesday went, you put an extra person on because it seems like it'll be busy, and by Sunday you either threw away four litres or ran out at eleven. The feeling isn't wrong, exactly — it's just built from the last two days you happened to be standing in the shop, when the till has been recording every hour of every week since you opened.
What a forecast is actually for
Not for knowing next week's revenue. You will not predict that, nobody does, and a number accurate to the dirham would change nothing you do on Monday morning. A café forecast exists to answer two much narrower questions: how much do I order, and who do I put on.
That's a useful distinction, because both questions have an asymmetric cost. Over-ordering fresh milk costs you the milk. Under-ordering costs you the sales you couldn't make and a regular who now knows you run out. Over-staffing costs you a few hours of wages. Under-staffing costs you the queue that walks. Once you know which mistake is more expensive on which line, you stop trying to be accurate and start trying to be wrong in the cheaper direction.
Start with the four weeks you already have
Pull sales by day for the last four weeks. Not one week — one week is a story, four weeks is a pattern, and the difference between them is a public holiday you'd forgotten about. Line the same weekdays up next to each other and read across.
What you're looking for is the shape, not the total. Cafés are astonishingly repetitive: the same days are strong, the same days are quiet, and the gap between them stays roughly proportional even as the overall level moves up or down. Once you can say "Wednesday is usually about three-quarters of Saturday" you can forecast any day from any other day, which is most of what a small café needs.
Then do the same with sales by hour, because a day total is the wrong unit for a rota. Two days with identical revenue can need completely different staffing — one arrived in a ninety-minute crush, the other trickled in from ten until six. That's the same reading behind getting people through the peak, and it's why the hourly view is the one that changes decisions.
Match the horizon to the decision
Different decisions need different distances ahead, and forecasting everything a month out is how the whole exercise becomes a chore you abandon in week three.
| What you're deciding | How far ahead | What actually drives it |
|---|---|---|
| Milk, pastries, anything fresh | One to three days | The same weekday's recent pattern |
| Beans, dry goods, packaging | One to two weeks | Four-week average plus supplier lead time |
| The rota | One to two weeks | Sales by hour, not sales by day |
| Extra cover or a closed day | A few days | Whatever is on the calendar that isn't normal |
The dry goods row is the one that quietly connects to everything else, because a forecast on its own doesn't tell you what to order — it tells you what you'll use. Turning that into an order needs the level you keep on the shelf, which is what par levels are for. Forecast the usage, order back up to par, and the two together stop the Friday morning phone call to a supplier who doesn't deliver on Fridays.
Adjust for what the numbers can't see
The history gives you a baseline. You then adjust it for the things you know and it doesn't, and in the UAE that list is longer than the arithmetic. School terms move families around. The summer takes a visible bite out of anywhere that depends on walk-in trade, and hands some of it to anywhere air-conditioned and indoors. Ramadan reshapes the entire day, not just the total. A long weekend empties an office district and fills a residential one.
Your own street matters as much as the calendar. A neighbouring unit under fit-out, a road closure, a big tenant moving floors, a mall running an event — these show up in your numbers with no warning and no explanation unless somebody wrote them down. Keep a line of notes against each week. Six months later it's the only thing that tells you why one Thursday was double the one before it.
Be careful with weekends specifically. The UAE doesn't have one universal working week across every sector and emirate, so "the weekend" for your café is whatever your own by-day report says it is — the days when the people near you are free, which for an office-district café and a residential one may not be the same days at all.
Check it on Monday, or you never get better
This is the step that makes it a method rather than a guess with extra steps. At the start of each week, write the forecast down — a number per day, and the rough hourly shape for the busy ones. Then at the end of the week, put the actuals next to it and look at the gap.
You're not marking yourself. You're looking for a bias, which is the only thing you can act on. If you're consistently ten per cent under on Saturdays, your Saturday orders have been short every week and you've been quietly capping your own sales. If you're always over on Mondays, you've been throwing milk away every Monday since spring. A bias in one direction is worth fixing; a scatter around the middle means the forecast is doing its job and the rest is weather.
Fifteen minutes, once a week, in the same slot you already read the reports worth checking weekly. The rota built on it belongs in the same session — staff scheduling for a small café is the other half of this, because a forecast nobody staffs against is just a number on a page.
Where MidaOne fits
MidaOne doesn't forecast for you, and there's no prediction screen to point at. What it has is the history the method above runs on: sales by day and sales by hour as selectable reports, a busiest-hours chart, and sales by item and by category when you need to know which lines drove a week rather than just how big it was. Stock updates live as you sell, so the usage you're forecasting and the stock you're ordering against sit in the same record.
For a group, the multi-branch reporting gives you consolidated figures alongside per-branch detail — which matters here because branches rarely share a shape, and forecasting a quiet mall kiosk off a busy street café's pattern is how you end up over-ordering for both. It's AED 200 a month, flat, on devices you already own.
Sales by day, by hour and by item — the history a forecast needs. Free for 14 days.
Start your free trialFrequently asked questions
How much sales history does a café need to forecast a week?
Four weeks of by-day data is enough to start, because it shows you the weekday shape rather than a single week's accidents. A year is better, since it also carries the seasonal swing and the holidays. But don't wait for a year — start with what you have and the forecast improves every week you keep score.
How far ahead can a small café usefully forecast?
It depends entirely on the decision. Fresh items are a one-to-three-day question, dry goods and rotas sit comfortably at one to two weeks, and beyond about a month you're planning rather than forecasting. Trying to predict a specific Tuesday six weeks out is effort spent on a number you'll replace before you use it.
Should I forecast revenue or the number of orders?
Order counts are usually the more useful of the two for ordering and staffing, because they map directly onto cups made and people needed. Revenue can move without volume moving at all — a price change or a shift in what people are buying. Many owners track both and use the count for operational decisions.
Is a spreadsheet good enough, or do I need forecasting software?
A spreadsheet is genuinely fine for one or two sites. The value is in the routine — writing the number down, then checking it against what happened — not in the tool. What you do need is till data you can pull by day and by hour without retyping it, because a method that takes an hour of data entry gets abandoned by week three.
What do I do when a week comes in nothing like the forecast?
Write down why, next to the week. One odd week is usually an event, a closure, weather or a holiday, and it should be excluded from your baseline rather than allowed to drag it. It's the repeated misses in the same direction that mean something — those are telling you your baseline is wrong, not that the week was.
None of this requires you to become a numbers person. It's one printout, fifteen minutes, and the discipline of writing next week's guess down where you'll have to look at it again — which is the only part that's actually hard, and the only part that makes the guess better.