April 4, 2026
AI Demand Forecasting for Restaurants: Which Day, Which Item, How Much?
Friday night, 8 p.m., the dining room is full and dozens of orders have stacked up on the kitchen display. But the chicken pizza dough is gone, the sausage stock has run out, and the cook cannot fill half of the orders coming in over the next thirty minutes. Earlier in the week it was the opposite: too much meat was prepped, it did not sell, and it went in the bin by closing. Both scenes come from the same restaurant in the same week. The problem is forecasting, or more precisely, forecasting by gut.
AI demand forecasting, which RestApp is developing, is built for exactly this moment. It looks at your past orders, your day and hour patterns, and the effect of weather and the calendar together, and tries to answer one question: which hours will be busy tomorrow, roughly how much of each item will sell, and based on that, how many portions of prep and how much stock do you need? The goal is to help you build the morning opening plan on data rather than a hunch.
An important note: this feature is currently in early access development and is not yet live on all accounts. Every forecast it produces is a suggestion; the decision on how much to prep and which shift to reinforce is always yours. A forecast is not a guarantee. It shows likely demand by looking at history and patterns. In this article we explain how demand forecasting works, what it can add to your restaurant, and how it combines with the RestApp features you already use, with a concrete example.
Why Is Demand Forecasting Hard in a Restaurant?
Demand forecasting is one of the oldest pains in the restaurant business. Prep too little and you run out, the customer asks for an item on the menu and cannot get it, and you lose both that sale and their goodwill. Prep too much and product sits on the shelf, fresh items spoil, and you write it off as waste at the end of the day. Hitting the middle takes experience, but even experience cannot always keep in mind the difference between Friday and Monday, between the start of the month and the end of it, between a rainy day and a sunny one.
Once variables like promotions, holidays, match days, and school breaks enter the picture, forecasting gets even harder. Most operators manage it with the head chef's intuition or by looking at last week's numbers. That method is not bad, but it depends on one person's memory, and when that person goes on leave, the system goes on leave too.
What AI demand forecasting tries to do is put that intuition on paper and factor in the entire history fresh every day. A person remembers the last few weeks; the model sees months back at the same time and surfaces the patterns that repeat.
What Data Does AI Demand Forecasting Look At?
Forecasts are not produced in a vacuum; they come from your restaurant's own data. On the RestApp side this data is already accumulating: how many order tickets opened on which day and hour, how many portions of each item sold, what the average ticket was, and which channel (dine-in, delivery, online) it came from. The cloud reports you already use today to see these patterns are the core feed for the forecast.
On top of that comes the effect of the calendar and the seasons: the difference between weekdays and weekends, public holidays, the rush at the start of the month, the split between summer and winter menus. Over time, outside signals like the weather can also be added to the model; on a rainy evening, delivery orders may rise while dine-in traffic drops, and that pattern becomes visible in the data.
There is also the structure of the menu itself. If a recipe and cost setup defines which ingredients make up an item, the sales forecast can be turned directly into a stock requirement. So it becomes possible to calculate not just how many pizzas you will sell, but also how many kilos of flour and how many packs of mozzarella you will need for them.
A Concrete Example: Mario's Pizzeria's Friday Plan
Say you run a mid-sized pizzeria called Mario's Pizzeria. Looking at the last three months of data, the demand forecast produces a picture like this for next Friday: 7 p.m. to 10 p.m. is the busiest window, with an estimated 120 order tickets. The three items expected to sell the most are the mixed pizza (around 70 portions), the chicken pizza (45 portions), and the fries (90 portions).
Combined with recipes, these numbers turn into a prep list: roughly 18 kilos of pizza dough, 6 kilos of mozzarella, 5 kilos of potatoes. If last Friday you ran out of chicken pizza after 14 portions and had to refuse orders, the system sees that too and suggests pushing the stock a bit higher this week. The suggested number, say 200 grams of extra prep on chicken, drops onto the screen as a note.
The decision is still yours. You can look at the 120-ticket forecast and say, 'there is an event at the mall next door this week, let me prep for 140.' The AI does not apply its number on its own; it does not order stock or change prices. It only puts a concrete starting point on the table, and your local knowledge rides on top of it. If an item is selling faster than expected, the kitchen display system (KDS) shows it in real time, and you reprioritize during the shift.
Cutting Waste and Stockouts at the Same Time
The most concrete payoff of demand forecasting is that it tightens both ends at once. On one side there is waste: fresh product prepped in excess, ingredients headed for the bin at closing, spoiled stock. On the other side there are lost sales: orders turned away because an item ran out, and the impression that 'sorry, that one is finished' leaves on the customer. Operators usually solve one while growing the other; they prep little and run out, or prep a lot and throw it away.
Because the forecast tries to bring prep closer to expected demand, it aims to narrow both ends. To be realistic: no forecast is right one hundred percent of the time, and there will be surprise crowds and empty evenings alike. But forecasting based on data, rather than by gut, shrinks the margin of error over time and makes a clear difference especially on regular, repeating days.
You can track this saving in your reports. Put the forecast sales side by side with the actual sales and you see how far off you were on which days. Because the model is updated with new data every week, as your menu or customer profile changes, the forecast tries to adapt with it.
Working Together With Loyalty and Promotions
Demand forecasting is not separate from your campaigns; on the contrary, it becomes more meaningful alongside them. When you plan a campaign on the loyalty and promotion side, for example 'second pizza half price on Wednesdays,' how that campaign affected demand in the past sits in the data. The forecast tries to factor in the likely demand bump on that day when you set up a similar campaign again.
The reverse also holds. If the forecast shows which days are naturally quiet, you can think about building a special offer for your loyalty members to fill those empty windows. So the forecast becomes an input not only for 'prepping the busy day' but also for 'reviving the quiet day.'
The line here is the same: the AI can give you a draft idea about which campaign might work on which day, but you are the one who builds the campaign, sets the discount rate, and pushes it live. The system does not launch a promotion on its own.
Is This Feature Live Now? An Early Access Note
Let us be clear: AI demand forecasting is not a live product right now; it is being developed in early access inside RestApp. It is not switched on automatically across all accounts, and the example numbers you see are there to explain how the feature will work. We are not saying 'it is ready in your account tomorrow'; it will roll out gradually and after testing, as it becomes available.
The core principle does not change in this process. The AI is the side that suggests and prepares drafts; the decision and the approval are always with the human. A forecast is advice, not an instruction. The system does not order stock on your behalf, does not play with prices, and does not launch campaigns. You decide how much prep to do and which suggestion to apply and which to skip.
Let us also be clear on the data side. Your restaurant's data belongs to you and is processed in line with GDPR. Shared models that are sold or handed to other businesses are not trained on your data. The forecast learns from your own history, and the result stays with you. If you are interested in early access, you can start building the reporting and recipe foundation today by trying RestApp for free, so that when the feature opens you are ready with data already in place.
Key takeaways
- AI demand forecasting starts from your past order data and tries to predict which days and hours will be busy and how much of each item might sell.
- When the sales forecast is combined with recipes, it turns into a prep and stock plan; it aims to cut over-prepping and waste and running out and losing sales at the same time.
- A forecast is not a guarantee; it shows likely demand and is updated with new data every week. The decision, the approval, and the final word are always with the operator.
- The feature is not live yet; it is being developed in early access. Setting up your cloud reports and recipe and cost foundation today means the feature finds you ready when it opens.
- Your data belongs to you, is processed in line with GDPR, and is not used to train models shared with other businesses.
Frequently asked questions
Is AI demand forecasting live now, and when can I use it?+
It is not live yet; it is being developed in early access inside RestApp and is not switched on automatically across all accounts. It will roll out gradually and after testing. In the meantime, if you set up your cloud reports and recipe and cost foundation, you will start with data already in place when the feature ships.
Is the forecast a guarantee, and what happens when it is off?+
A forecast is not a guarantee. It shows likely demand by looking at past data and repeating patterns. A surprise crowd or an unexpectedly quiet evening can always happen. The goal is to shrink the margin of error compared to forecasting by gut, not to zero it out. You can put the forecast and the actual side by side in your reports and track the gap.
Does the AI make the prep and stock decision?+
No. The AI only produces suggestions and drafts; the decision on how much to prep and which shift to reinforce is yours. The system does not order stock on your behalf, change prices, or launch campaigns. You put your own local knowledge on top of the suggested number.
Is my data safe, and is it shared with other restaurants?+
Your restaurant's data belongs to you and is processed in line with GDPR. The forecast learns from your own history. Shared models opened up to or handed to other businesses are not trained on your data.
What data does the forecast look at?+
Your past order ticket and order activity, day and hour intensity, item-level sales, channel split (dine-in, delivery, online), and the effect of the calendar and seasons are the core inputs. Over time, outside signals like the weather can also be added. If recipes are defined, the sales forecast can be turned directly into a stock requirement.
How does it combine with the RestApp features I already have?+
The forecast is fed from cloud reports, the stock requirement is calculated with recipes and costs, you see item movement in real time during the shift on the kitchen display system (KDS), and it works together with loyalty and promotion campaigns to revive quiet days.
Stop paying commission on your orders
Launch your own commission-free online ordering system with RestApp.