SALES FORECAST

Glossary

Definition

Sales forecasting (or demand forecasting) involves predicting future sales volumes for a product or category over a given period (week, month, season). It is based on an analysis of sales history, market trends, seasonality, planned promotions, and external events (weather, holidays, social trends). Forecasts inform purchasing, pricing, and logistics decisions.

Why it's important

Optimizing inventory: Reliable forecasts help prevent overstocking (storage costs, markdowns) and stockouts (lost sales, dissatisfied customers).

Managing prices: Anticipating high demand allows you to raise prices (dynamic pricing), while anticipating a decline allows you to plan promotions.

Improving profit margins: buying the right quantity at the right time reduces purchasing costs (through volume discounts) and minimizes end-of-season markdowns.

A concrete example

A toy retailer uses a forecasting model to estimate Christmas sales. The algorithm analyzes sales data from the past three years, factoring in TikTok trends (viral products), the weather forecast (a harsh winter means more indoor games), and planned promotions. It predicts a 25% increase in board games and stable sales for high-tech toys. The distributor adjusts its orders accordingly: a 30% increase in board games and stable inventory for high-tech toys. Result: stockouts avoided for board games, no overstocking of high-tech toys, and optimized margins.

Methods

Traditional statistical models: moving averages, exponential smoothing. Simple but not very accurate for complex product assortments.

Machine Learning: algorithms (Random Forest, XGBoost, neural networks) that learn complex patterns and incorporate dozens of variables (price, weather, competitor promotions, events, web trends). Significantly higher accuracy.

Hybrid approaches: combine statistical models (for the baseline) and machine learning (to refine forecasts for key products).

Common Mistakes

Ignoring external factors: A forecast that doesn’t take into account the weather, school holidays, or viral trends misses major fluctuations.

A single template for all products: a bestseller and a niche product don’t follow the same patterns. Segment your templates by category or sales profile.

Don’t update the models: a model trained on 2020 data won’t capture post-COVID changes. Retrain them regularly using recent data.

Learn more

Research & Data: Price analysis to review your sales history and improve the quality of your forecasting data.

Solutions: Sales forecasting and AI to deploy high-performance forecasting models and automate the forecasting process.

Tip: Operational pricing to align your forecasting, procurement, and pricing processes.

Resources: Check out our case studies on the implementation of AI-powered forecasting in retail.

Mini FAQ

It depends on the industry and the level of detail. Forecasting at the category level, with 70 to 90% accuracy, is easier than forecasting at the SKU level, where accuracy typically ranges from 50 to 70%. The goal is to continuously improve accuracy.

No, they complement it. A good model incorporates domain expertise, for example through manual adjustments for product launches or exceptional events.

No. Focus your efforts on best-sellers, which often account for 80% of revenue, as well as high-turnover products. Long-tail items can follow simpler rules.

‍