Demand forecasting for multi-outlet F&B: start with the data you have

Restaurant demand forecasting for a multi-outlet group can start with the POS data you already hold. Clean it so stock-outs, promotions and closures don’t distort it, build a baseline from the same weekday over recent weeks, adjust for events you can see coming, forecast each item by outlet and daypart, and check the result against actual sales every week. That is enough to plan central kitchen production on evidence rather than habit.

It won’t be perfect. It doesn’t need to be. It needs to beat “roughly what we did last week”, and be honest about where it misses.

What is restaurant demand forecasting?

Restaurant demand forecasting is estimating how many of each item each outlet will sell in a given period, so the central kitchen can produce, and purchasing can buy, before the orders arrive. It differs from sales forecasting in F&B finance, which predicts revenue for a budget. The kitchen needs units: plates of nasi lemak, trays of sambal, kilos of chicken.

It also differs from a plan. The forecast says what customers are likely to buy. The production plan decides what to make, after stock on hand, shelf life and kitchen capacity. Keeping the two separate is how you tell a bad forecast from a bad plan.

We’ve written about the wider signals that shape demand. This piece is narrower: a method you can start next week with records you already keep.

The data you already have

Data you already have	What it tells you	How to use it
POS sales by item, outlet and time	What sold, where and when	The baseline, by item, outlet and daypart
Sold-out times	Where demand ran ahead of supply	Flag those days; don’t treat them as true demand
Wastage and leftover logs	Where production ran ahead of sales	A bias check: steady waste means a forecast running high
Outlet orders to the central kitchen	What outlets asked for, against what they sold	Shows padding in outlet orders
Promotion calendar	Which spikes were bought	Tag them, and measure the uplift for next time
Closures, short hours and renovations	Days that weren’t normal	Exclude them from the baseline
Delivery platform orders	How the channel mix moves by daypart	Forecast delivery separately if it is a large share
Stock counts	Whether “on hand” is real	Turns a forecast into a production quantity

The last row matters more than it looks. A forecast minus a wrong stock figure is a wrong order, which is why a clean month-end stock take is part of forecasting, not separate from it.

How do you clean POS data before you forecast?

Stock-outs hide true demand. If the ayam masak merah sold out at 7pm on Friday, Friday’s POS figure records what you had, not what customers wanted. Flag any item-day that sold out before closing. Leave it out of the average, or, if you record the sell-out time, estimate the missing hours from a normal Friday’s pattern. Left in, these days teach the forecast to under-produce, which causes more stock-outs.

Promotions inflate it. A buy-one-free-one week is real demand at that price, not at the usual one. Tag promotion days and keep them out of the baseline.

Closures and short days deflate it. Renovations, power cuts, mall closures and floods all produce low numbers that say nothing about next week.

Code changes break the history. If an item was renamed or re-coded, its sales are split across two records and both look weak. Clean item master data is what keeps a year of history usable.

New outlets need a stand-in. The first weeks after opening are rarely typical. Borrow the pattern of a similar sister outlet until the new one has its own history.

A starter method in five steps

1. Pick the level. Forecast each item by outlet and by daypart, using the dayparts that match your dispatch runs. This is what multi-outlet forecasting means in practice: outlet totals hide the swings that cause waste, and a central kitchen that dispatches twice a day needs to know which half of the day the demand falls in.
2. Build the baseline. For each item, outlet and daypart, average the same weekday over the last four to six clean weeks. Use the median if one odd week keeps skewing the result. A Tuesday and a Saturday are different businesses, so never average across them.
3. Adjust for what you can see coming. Keep an event calendar: paydays, school holidays, public holidays, local events, promotions, Ramadan. Measure each event’s effect from your own history. If, hypothetically, the weekend after payday has run about 15% above baseline at a mall outlet, apply that, not a guess. For Ramadan, align last year’s data by day of Ramadan rather than calendar date, and expect demand to shift between dayparts as much as change in total.
4. Turn the forecast into production. Forecast, minus stock on hand at the outlet, plus a buffer sized to the item’s usual error, gives each outlet’s requirement. Add the outlets together for central kitchen production planning, then convert through recipes into ingredient quantities for purchasing. If outlets still order from par levels set when they opened, compare the two.
5. Record the actuals. Every forecast gets compared with what sold. Without this step nothing improves.

How do you know if the forecast is any good?

Two numbers, in plain words.

How far off, on average. For each item, take the daily gap between forecast and actual sales, ignore whether it was over or under, and express it as a percentage of what sold. That is the average error, and it tells you how much buffer the item needs.

Which way, consistently. Now keep the sign. If an item is over-forecast four days out of five, the forecast is biased, and bias is what fills the bin. A forecast that misses in both directions is noisy. One that misses in one direction is wrong in a way you can fix.

Then compare against the lazy alternative: same as last week. If your method doesn’t beat it on your top items, simplify before you add anything.

A weekly review keeps it honest. Thirty minutes, the same day each week, with the central kitchen manager, operations and someone who knows the outlets: the ten biggest misses, the reason for each (event, stock-out, promotion, data error, genuine surprise), and one change to the event calendar or the cleaning rules. Record the reasons. In three months they are your list of what to fix.

Where RMS and the AI-Kitchen Command Center fit

A spreadsheet can carry this method a long way. What it struggles with is connecting the forecast to everything downstream, every day, across every outlet.

Codemax RMS (Resource Management System) runs on live POS sync, so sales by item and outlet arrive without exports. It plans production against rolling outlet demand, uses demand- and supply-based job scheduling, and turns the plan into pick lists and batch tickets. On the buying side it recommends purchase-order quantities from par levels and POS data, and because stock-on-hand is live, “forecast minus on hand” uses a real number. Outlets order from the central kitchen on mobile, so orders and sales sit side by side for the padding check.

The AI-Kitchen Command Center watches every outlet and product against its own baseline and flags when one drifts: an item’s sales stepping outside their usual pattern at one outlet, or waste climbing while sales hold flat. It doesn’t replace the weekly review. It makes sure the review starts with the right items. To see how that would map to your outlets, talk to our team.

Where does your forecasting sit today?

1. Habit. Last week copied forward and adjusted by feel.
2. Baseline. Same-weekday averages from cleaned POS data, in a spreadsheet.
3. Adjusted. An event calendar, item-by-outlet-by-daypart forecasts, and error and bias reviewed weekly.
4. Connected. Forecasts feed production plans and purchase orders directly, stock on hand is live, and drift is flagged automatically.

Three questions will place you. Can you list last Friday’s sold-out items and when each ran out? Do you know your top ten items’ average error, and which way it leans? Does production start from a forecast, or from whatever the outlets ordered?

If the answers are no, no and outlet orders, you are on the first rung. The second needs only a spreadsheet, a few weeks of clean data and someone who owns the weekly review.

Frequently asked questions

How can a restaurant forecast demand without special software?

A restaurant can forecast demand from its own POS history using a spreadsheet. Export sales by item, outlet and hour, remove days distorted by stock-outs, promotions or closures, and average the same weekday over the last four to six clean weeks. Adjust for known events such as paydays and public holidays, then compare each forecast with actual sales every week.

How accurate should a restaurant demand forecast be?

There is no single accuracy target for a restaurant demand forecast, because volatility differs by item, outlet and daypart. Measure your own average error and whether the forecast leans consistently high or low, then compare it with simply repeating last week's sales. A useful forecast beats that simple method on your top items, and buffer stock is sized to the error that remains.

How do stock-outs affect restaurant sales forecasting?

Stock-outs make sales data understate true demand, because POS records only what was available to sell. If those days feed the forecast unadjusted, the forecast learns to under-produce, which causes more stock-outs. Flag every item that sold out before closing, then exclude those days from the baseline or estimate the lost hours from a normal day's pattern.

Should a central kitchen forecast by outlet or for the whole network?

A central kitchen should forecast each item by outlet, and ideally by daypart, then add the outlet forecasts together for production. A network total hides the outlets that are rising or falling, and errors at outlet level are where waste and stock-outs actually happen. Purchasing can then convert the combined production plan into ingredient quantities through recipes.

How do you forecast restaurant demand during Ramadan in Malaysia?

To forecast Ramadan demand in Malaysia, align last year's sales by day of Ramadan rather than by calendar date, because Ramadan moves earlier each year. Expect demand to shift between dayparts, towards buka puasa and sahur at many outlets, rather than simply rising or falling. Review each outlet separately, since office, mall and neighbourhood locations behave differently.