Christmas is approaching, and the latest retail forecasts point to an interesting problem for food businesses.

According to IGD's latest Christmas predictions, UK grocery sales values are expected to increase this festive season, but the actual volume of products sold could remain broadly flat.

Meanwhile, 51% of shoppers say they plan to buy more premium own-label products. Consumers are increasingly looking to save on everyday essentials while spending selectively on higher-quality food, desserts and entertaining.

For retailers, restaurants and their supply chains, this creates a problem that traditional forecasting models can struggle to solve.

A business can predict its total sales almost perfectly and still order completely the wrong stock.

The problem with forecasting at the wrong level

Imagine a food retailer that typically sells 200 units across two prepared-food ranges each day.

Historically, that demand might be split between 160 standard products and 40 premium alternatives.

The retailer's forecasting system predicts another 200 units for tomorrow. Based on historical sales, it orders roughly the same product mix.

But consumer behaviour has shifted. Actual demand is now 125 standard products and 75 premium products.

The overall forecast was correct. Total demand was exactly 200 units.

Yet the retailer could still end up with 35 unsold standard products and 35 missed premium sales.

For perishable food, this is particularly damaging. One side of the equation creates waste, while the other creates lost revenue and disappointed customers.

At enterprise scale, these errors can multiply across thousands of stores, products and replenishment cycles.

And the problem becomes harder when demand shifts between sales channels, promotions, products and individual locations.

Why historical sales data isn't enough

Most businesses have years of transaction data available through their point-of-sale systems.

But historical sales represent what customers purchased, not necessarily what they wanted.

That distinction matters.

If a store ran out of a premium sandwich at lunchtime, its sales data might show 20 units sold. It cannot automatically tell us whether demand was actually 20, 30 or 50.

This is known as censored demand. When products are unavailable, recorded sales underestimate potential demand.

Promotions create another challenge.

A discounted product might sell significantly more units during a promotional period, but some of those sales may have been brought forward from future days. Others may have displaced sales of a competing product.

A forecasting model that treats every additional promotional sale as new demand risks overestimating future requirements.

Weather, local events, school holidays, delivery availability and changing customer preferences introduce further complexity.

Christmas brings all of these variables together.

Building a more intelligent demand forecast

A more effective approach is to separate total demand from the distribution of that demand across products.

Rather than predicting only how much a store will sell, forecasting systems should also estimate which products customers are most likely to choose, and when.

This can involve several layers of modelling.

Hierarchical forecasting allows demand to be modelled at different levels, such as region, store, category and individual SKU. Forecasts can then be reconciled to maintain consistency between the different levels.

Contextual demand modelling incorporates external variables such as weather forecasts, promotions, local events and trading patterns. These inputs help distinguish genuine changes in demand from normal historical variation.

Probabilistic forecasting estimates a range of possible outcomes rather than producing a single number. For example, a store might have an expected demand of 40 units, but a 90th percentile forecast of 55.

That uncertainty matters when deciding how much stock to hold.

An item with a long shelf life and a high cost of running out may justify additional safety stock. A freshly prepared product with only a few hours of useful life needs a different approach.

The goal is not simply to achieve the lowest possible forecast error.

It is to produce better operational decisions.

A forecast is not an order

Even a technically accurate forecast doesn't automatically translate into the correct replenishment quantity.

A useful ordering decision must account for existing stock, confirmed incoming deliveries, supplier lead times, case-pack sizes, minimum order quantities and product shelf life.

For prepared food, there is another layer of complexity: recipes.

A forecast for 100 sandwiches might translate into demand for bread, fillings, packaging and other ingredients. Those ingredients may also be shared across several products.

An increase in demand for one menu item can therefore affect the stock required for several others.

This is where connecting demand forecasting with inventory management and recipe data becomes important.

The system needs to understand not just what is likely to sell, but what is already available, what can be produced and what needs to be ordered.

And critically, it needs to recognise when additional stock would be more likely to become waste than generate sales.

Optimising for outcomes, not just accuracy

Forecast accuracy is a useful measure, but it should not be the only measure of success.

Consider two ordering strategies.

One consistently holds excess inventory to protect availability. It may achieve excellent customer service levels but generate significant food waste.

Another keeps stock exceptionally lean. Waste falls, but customers frequently encounter unavailable products.

Neither represents an ideal outcome.

A more intelligent approach considers the expected cost of both overstocking and understocking, alongside margin, availability requirements and product life.

This is an optimisation problem, not simply a forecasting problem.

For multi-site operators, it also means understanding how decisions should differ between locations.

A city-centre convenience store, a motorway service station and a suburban supermarket can experience very different festive demand patterns, even when selling similar products.

A single national forecast cannot capture all of those differences.

Moving from prediction to action

There is one final problem with even the most sophisticated forecasting systems.

A forecast that nobody acts upon has very little operational value.

For food businesses, the real opportunity comes from connecting data, predictions and operational workflows.

At Orderly, this principle shapes how we approach store operations.

Our technology brings together capabilities including predictive ordering, stock control, recipe management and integrations with existing business systems.

The objective is to help businesses make better decisions using the data they already generate, without forcing store teams to interpret increasingly complicated reports.

That becomes especially important during peak trading periods, when managers have less time to manually investigate forecasts, stock discrepancies and changing demand.

The most useful intelligence is the intelligence that helps somebody make the right decision at the right time.

Christmas 2026 will test more than demand forecasting

This Christmas, the challenge for food retailers and restaurant operators may not be a dramatic increase in the total number of products customers want.

It may be the speed at which customers change their preferences.

Value-focused essentials, premium treats, promotions, prepared food and changing shopping habits create a complicated demand picture.

Businesses that rely exclusively on historical sales averages risk missing those changes until the stock has already been ordered.

Those that connect forecasting with contextual data, inventory, recipes and operational decision-making have a better opportunity to respond.

Because the real measure of an effective forecasting system isn't how accurately it predicts sales. It's whether the business has the right products, in the right place, at the right time, without creating unnecessary waste.

And during the busiest trading period of the year, that distinction matters.

Orderly helps enterprise food retailers, restaurant operators and their supply chains make smarter ordering and inventory decisions. Discover our platform.