Demand Forecasting
On this page
Do you know how to measure the accuracy of your forecasts?
Let's plan an exchangeDiscover our sales forecasting softwareDemand forecasting involves estimating what customers would purchase over a given period if the product were available at the proposed price, and anticipating how that volume responds to price, promotions, and contextual factors (weather, calendar, competition). It differs from sales forecasting, which projects the actual volumes sold, which are limited by inventory. It serves as the basis for all pricing, promotional, and purchasing decisions.
The Essentials
- Demand forecasting estimates what customers would buy in the absence of disruptions, whereas sales forecasting projects sales under specific constraints.
- Learning from a history in which disruptions have masked demand leads to a persistent underestimation of needs.
- It estimates the expected impact of a price or promotion on sales volume before making a purchase.
In 6 Questions
An estimate of actual demand, not just past sales.
Planning, purchasing, pricing, logistics.
Months in advance for purchases, every week for operations.
SKU × store × week, category × month, product family × quarter.
Optimize inventory, promotions, and profit margins.
Break-adjusted history, price elasticity, and machine learning, measured by MAPE.
Why Forecast Demand?
Because inventory levels, promotions, and prices are all determined based on an assumption about demand.
- Optimize inventory: Avoid stockouts (lost revenue) and excess inventory (tied-up capital, damage).
- Calibrating promotions: Understanding natural demand makes it possible to measure the incremental effect of a promotion.
- Managing margins: adjusting prices based on projected demand, using an approach similar to predictive pricing and yield management.
Demand Forecasting vs. Sales Forecasting: What's the Difference?
Demand is what customers want to buy; sales are the portion of that demand that the store was able to fulfill.
| Criterion | Demand Forecasting | Sales Forecast |
|---|---|---|
| What we estimate | Question: Whatwould customers buy if there were no inventory constraints? | Sales: Theactual volume of sales |
| Effect of a Breakup | Application: Thelost application is reconstructed | Sales: Noinventory, no sales |
| Integrated levers | Request: Prices, promotions, competitors' prices, background information | Sales:Procurement Plan, Product Assortment |
| Primary Use | Request: Setprices, tailor promotions | Sales: Planorders, revenue, and logistics |
In practice, we first forecast demand, then derive sales figures by applying inventory and product mix constraints. For a product that is always in stock, the two figures are the same; for a product that is frequently out of stock, they differ significantly. Sales-side methods are detailed in the sales forecasting worksheet, and the mechanism by which underestimated demand leads to stockouts is explained in our article on stockouts caused by poor forecasting.
Factors that affect demand: price, promotions, and the broader context
A forecast of effective demand for pricing purposes isolates the effect of each lever.
- Price:Price elasticity measures how much demand decreases when the price increases. It allows us to simulate a price change before implementing it.
- Promotions: They create a spike, often followed by a dip, and draw some of their sales volume away from similar products (intra-category cannibalization).
- Competitive pricing: A widening price gap shifts demand from one retailer to another, especially for the products most frequently compared.
- The context: weather, holidays, school breaks, or unforeseen events—all covered in our article on exogenous events in forecasting.
- Availability: When a product is out of stock, some of the demand for it shifts to a substitute, while the rest is lost.
Real-life example: a T-shirt from the summer collection
A clothing retailer plans to sell 28,000 units of a T-shirt, with sales peaking in mid-June, and orders 32,000.
28,000 units planned, 32,000 ordered
Textile Sign · Demand Forecast for a T-Shirt from the Summer Collection
units projected for the season for this product, based on historical data, weather forecasts, and fashion trends (NLP analysis of social media).
units ordered from suppliers based on this forecast, with a 15% margin
Anticipated sales peak, used to plan logistics and promotional windows
Starting in January, the models combine five years of historical data, fashion trends analyzed on social media, the regional weather forecast, and the sales calendar. The forecast guides purchasing decisions (with a 15% safety margin), logistics, and promotional windows.
How can you build a robust demand forecast?
By combining several approaches and continuously measuring accuracy.
| Brick | Contents |
|---|---|
| Revised History | Terminations, promotions, and anomalies adjusted prior to the apprenticeship. |
| Classical Statistics | Moving averages, ARIMA, Holt-Winters. |
| Machine learning | Gradient boosting, recurrent neural networks. |
| Business Expertise | The category manager's judgment on specific cases. |
| Measurement | MAPE: 5 to 25 percent, depending on the product's stability. |
Modern tools weight these sources based on context. This is exactly what our sales forecasting software does; when applied to end-of-season sales, it calibrates the markdowns in our “markdown” and “ markdown ” offerings. These forecasts also feed into pricing strategy simulations. The quality of the historical data is just as important as the model itself: find out what sales history is needed to make accurate forecasts.
The 4 Common Mistakes in Demand Forecasting
Unadjusted sales figures, too few external variables, only one forecasting level, or no performance metrics.
- Understanding gross sales: Without adjusting for out-of-stock situations, the model confuses a lack of inventory with a lack of demand and perpetuates the out-of-stock situation.
- Underestimating external variables: without factors such asweather, holidays, or competing promotions, the model fails to account for key variations.
- Forecast at only one level: the forecast must be consistent across all reference codes, stores, and categories.
- Failure to measure performance: without monitoring the MAPE or conducting back-testing, you won't notice when the model starts to degrade.
Frequently Asked Questions
Short answers to the most frequently asked questions about demand forecasting.
What is demand forecasting?
Demand forecasting estimates how much customers would buy over a given period if the product were available at the listed price, and how that volume would respond to a price change, a promotion, or the current context. It starts with sales history adjusted for out-of-stock periods, then incorporates price, promotions, competitor prices, weather, and the calendar, often using machine learning. It is used to set prices, tailor promotions, and determine purchase quantities.
What is the difference between demand forecasting and sales forecasting?
Demand forecasting estimates what customers would like to buy; sales forecasting estimates what will actually be sold, once constraints related to inventory, product assortment, and shelf space are taken into account. For a product that is always in stock, the two figures are the same. For a product that is frequently out of stock, they differ: forecasting based solely on sales will cause the product to be out of stock again in the next cycle. Therefore, we first forecast demand, and then derive sales figures from it.
What is the difference between demand-based pricing and competition-based pricing?
Competition-based pricing sets the price relative to competitors’ prices, using an index or a target spread. Demand-based pricing uses customers’ price sensitivity—measured by price elasticity—to determine the price that maximizes either profit margin or sales volume. The former protects the price image for the most frequently compared products; the latter maximizes profit margin where customers are less price-sensitive. Most retailers combine the two approaches depending on the role of each product.
How does price affect demand?
Through price elasticity: when the price rises, demand falls to a greater or lesser extent depending on the product. A necessity or a product with no substitute shows little response; a product that is frequently compared or easily replaceable shows a strong response. A demand forecast that incorporates this elasticity allows you to simulate several prices before choosing one, rather than observing the effect after the fact. It must also take into account competitors’ prices, which affect customer sensitivity.
What demand forecasting tool is best for a multi-store network?
You need a tool that forecasts at the SKU × store level, automatically accounts for out-of-stocks and past promotions, incorporates price, weather, and calendar data, and links the forecast to pricing and restocking decisions. Spreadsheets quickly reach their limits when dealing with more than a few hundred SKUs. Booper offers AI-powered sales forecasting linked to its pricing solution: the forecasted demand directly feeds into pricing recommendations, store by store.
What level of accuracy should we aim for?
Less than 10% MAPE (Mean Absolute Percentage Error) for stable products. For new products or volatile seasonal items, 20% to 30% is still acceptable due to insufficient historical data. Accuracy also depends on the time horizon: forecasting for next week is more reliable than forecasting for the next season, and a store-level forecast is less accurate than an aggregated forecast. Setting a single target for the entire catalog leads to an unfair assessment of the model.
What level of detail should be used?
It is chosen based on the intended use of the forecast. For operations, restocking, and store replenishment, forecasts are made at the reference level × store × week. Sales planning uses the category × month level, and purchasing uses the product family × quarter level. These levels must be consistent with one another: the sum of the detailed forecasts must equal the aggregate forecast; otherwise, each department will work with its own figures, and trade-offs will become impossible.
How do I add promotions?
Using a so-called “uplift” model, which separates natural demand—known as the baseline—from the specific effect of the promotion. This separation makes it possible to accurately predict the impact of a future promotion, based on its depth and prominence. Without it, promotional spikes inflate the baseline demand, and the model overestimates sales during normal weeks. It is also important to account for the dip that often follows the promotion, once customers have stocked up.
How should stockouts be handled in a forecast?
They must be reprocessed; otherwise, the model will learn that there was no demand, when in fact there was no supply. Sales observed during a stockout underestimate actual demand, and the model carries this underestimation over to the next cycle, causing another stockout. This is the most costly and most common flaw.
How can you forecast demand for a new product?
Compared to past launches with a similar profile, taking into account the category, price point, and type of promotion. The initial forecast is intentionally broad and narrows after two to three weeks of actual sales. It is unrealistic to expect high precision from day one.
Key Takeaways
- Demand forecasting estimates what customers would buy, not just what they have bought.
- It isolates the effects of price, promotions, and context: this is the starting point for pricing.
- Its performance is measured (MAPE) based on a backtest adjusted for out-of-stock periods.
Would you like to forecast demand before setting your prices?
Booper combines demand forecasts with pricing scenarios to help you make informed decisions.
Let's talk about your demand forecast →Discover our sales forecasting softwareThese articles can interest you
AI-based sales forecasting estimates demand for each product, store by store, by separating baseline demand from promotional demand. AI takes in more data points than a spreadsheet, but its value lies solely in the decisions it helps inform.
In retail, this decision is primarily about pricing or a promotion. A useful pricing forecast—one that estimates how much will be sold at each considered price—is evaluated based on its bias rather than its MAPE, and is validated against control stores.
How much historical data is needed for forecasting? It depends on the category, but reliability matters more than depth: historical data in which breaks, promotions, and price changes are dated is better than a long, unprocessed data set.
Two factors are often overlooked: the history of prices paid—without which elasticity cannot be measured—and the years of inflation, as reflected in volume figures. The adjustment begins with the items that account for the bulk of sales.