Demand forecasting in logistics: methods, benefits and practical applications

A supply chain management perspective
by aBCD Consulting

What demand forecasting is and why it is essential

Demand forecasting is the process by which a company estimates, based on historical data, market information and analytical models, the future demand for its products or services over a given time frame. It is not about guessing the future: it is about systematically reducing the uncertainty that governs every logistics decision, from inventory management and production capacity planning to warehouse organisation and transport scheduling.

In the modern supply chain, demand forecasting is the engine that drives all other planning processes. Accurate demand planning makes it possible to know in advance how many units of each item will be needed, where and when. This information cascades along the entire supply chain, determining reorder levels with suppliers, safety stock sizing, warehouse shift planning and customer delivery scheduling.

Conversely, inaccurate or missing forecasts generate two mirror-image inefficiencies, both equally costly. On one side, stockouts and delivery delays caused by higher-than-expected demand; on the other, excess stock and unnecessary logistics costs caused by lower-than-forecast demand.

At aBCD Consulting, we see demand forecasting as the most underestimated strategic lever in mid-sized companies: many organisations invest heavily in warehousing, transport and technology, yet continue to plan using Excel spreadsheets and personal intuition, missing out on potentially very significant benefits.

Building an effective demand forecasting process is almost always the most productive starting point for improving overall supply chain efficiency.

Benefits of demand forecasting in modern supply chains

The benefits of a structured demand forecasting process appear simultaneously across several dimensions of the supply chain, with a multiplier effect that makes it one of the most profitable investments in logistics consulting. The first and most direct benefit concerns inventory optimisation: accurate forecasts make it possible to size stock levels correctly, reducing both excess inventory, with resulting savings in working capital and storage costs, and stockouts, with a resulting improvement in customer service levels.

In our experience, companies that move from reactive to forecast-driven inventory management typically achieve significant reductions in average warehouse stock levels without reducing product availability, and often even improving it. A second key benefit concerns transport and distribution planning: knowing in advance when and where demand peaks will occur makes it possible to organise transport flows early enough to optimise loads, choose the most cost-effective modes and avoid costly, high-impact emergency solutions.

Demand forecasting also improves the quality of supplier relationships. Sharing purchase forecasts with key suppliers makes it possible to negotiate better terms, ensure material availability at critical times and reduce delays caused by unplanned urgent orders. Finally, robust demand planning is the foundation of the S&OP (Sales & Operations Planning) process, which brings sales, production and logistics functions together around a shared plan, reducing organisational tensions and improving the overall reliability of the supply chain.

Main factors affecting forecast accuracy in logistics

The quality of demand forecasting depends on the ability to correctly identify and model the factors that drive demand. Some of these factors are structural and predictable; others are situational and require continuous monitoring. In this section, we look at the three main categories of variables that every demand forecasting system must be able to read and interpret.

Seasonality and market trends

Seasonality is the best-known and most intuitive of the factors that influence demand: almost all products and markets show cyclical variations linked to the calendar (months, quarters, holidays, promotional campaigns) that recur with some regularity from year to year. Recognising and quantifying these seasonal patterns in historical data is the foundation of any forecasting model. A system that ignores seasonality will inevitably produce systematically wrong forecasts at demand peaks and troughs, causing the two classic problems of inventory management: overstock in the downturn and stockouts in the upturn.

Alongside seasonality, market trends represent the medium- and long-term component of how demand evolves: the structural growth of a product category, the decline of a distribution channel, the emergence of new customer segments. Being able to tell a structural trend from a temporary fluctuation is one of a demand planner’s most critical skills. An overestimated trend leads to unjustified investment in stock and capacity; an underestimated trend exposes the company to losing market share because it cannot meet demand.

At aBCD Consulting, we use time series decomposition models that explicitly separate the seasonal component from the trend component, allowing us to produce more accurate forecasts and manage inventory more precisely throughout the year.

Geography, competition and product type

Demand is not uniform, either geographically or in relation to the nature of the product. Two items from the same catalogue, sold through the same channels, can have radically different demand profiles depending on their type, price positioning and the competitive pressure they face. The geographical dimension is often underestimated in aggregate forecasting models: forecasts built at national or macro-regional level can mask very different local dynamics, with the risk of saturated warehouses in some areas and stockouts in others.

Effective demand forecasting must be granular enough to capture the geographical differences that matter for the company’s distribution network. Competition introduces an element of uncertainty that is hard to model using internal historical data alone: the launch of a competing product, an aggressive promotional campaign by a competitor or a price change can significantly alter expected demand patterns.

That is why the most sophisticated forecasting models combine external market information (sell-out data, market research, customer signals) with internal historical data. Finally, product type determines the forecast’s risk profile: fast-moving products with stable demand are relatively easy to forecast, while seasonal, promotional or short-life-cycle products require specific approaches and more frequent monitoring of forecasting parameters.

Changes in consumer behaviour

Of all the factors that affect demand forecast accuracy, changes in consumer behaviour are perhaps the hardest to anticipate and model: by definition, they represent breaks from the historical patterns on which forecasting models are based.

The pandemic of recent years offered the most extreme example, with sudden, profound shifts in buying habits that made forecasting models built on years of data obsolete within a few months. But even in more ordinary market conditions, consumer behaviour is constantly evolving. The growth of e-commerce has changed order profiles, reducing average order size and increasing frequency; growing concern for sustainability is changing product preferences; and volatile energy prices are affecting buying habits in non-linear ways.

To manage these changes, demand forecasting cannot rely solely on backward-looking statistical models. It must integrate leading indicators, such as real-time sell-out data, sales force feedback and trend analysis on digital channels, and it must be updated often enough to capture changes before they turn into the wrong stock levels or delivery delays.

At aBCD Consulting, we help companies build dynamic demand planning processes that can incorporate this qualitative information alongside quantitative data and update in a structured way as market conditions change.

Demand forecasting methods in logistics

There is no demand forecasting method that is universally better than the others: the choice depends on the availability and quality of data, the nature of products and markets, the forecasting horizon and the resources available. In this section, we present the three main approaches (qualitative, quantitative and advanced), explaining their logic, strengths and practical limitations.

Qualitative approaches: expert experience and intuition

Qualitative approaches to demand forecasting are based on the judgement, experience and market knowledge of people who, because of their role, have access to information that historical data alone cannot capture: sales staff, product managers, buyers and marketing managers. These approaches are particularly useful when historical data is scarce, unreliable or simply does not exist, for example when launching a new product, entering a new geographical market or assessing the impact of a new type of promotion. The Delphi method, for instance, involves the structured, iterative collection of opinions from a panel of experts, with the aim of converging on a shared forecast that integrates different perspectives. Sales force forecasts, built bottom-up by aggregating the estimates of individual account managers, are another widely used qualitative approach, particularly effective in B2B markets where customer relationships are stable and information on purchasing intentions is relatively accessible. The main limitation of qualitative approaches is their subjectivity and the difficulty of scaling the process systematically across large catalogues. They also tend to be influenced by recurring cognitive biases: sales optimism when forecasting, anchoring to the previous year’s results and underestimating variability. For this reason, qualitative approaches deliver the best results when they are rigorously structured and combined with quantitative data, within an S&OP process that explicitly integrates them.

Quantitative methods: historical data analysis and statistical models

Quantitative methods are the backbone of any structured demand forecasting system. They are based on analysing historical sales data and applying statistical models to extract recurring patterns (trend, seasonality, cyclicality) and project them into the future. The most widely used models range from moving averages and exponential smoothing, suited to relatively stable time series, to the Holt-Winters model and ARIMA models with their seasonal variants (SARIMA), which are more effective when trends and seasonality are pronounced. The choice of the right model depends on the structure of the time series: there is no universally best model, and it is good practice to test several and select the one that minimises forecast error over a validation period. A critical point we always stress strongly in our consulting work is that the quality of a quantitative forecast depends directly on the quality of the data behind it. Historical data that is incomplete, unclean or affected by uncorrected anomalies, such as exceptional promotions or past stockouts that artificially depressed sales, produces systematically biased forecasts. Before implementing any statistical model, it is therefore essential to invest in data quality. A well-designed process for collecting, cleaning and validating sales data is what makes any forecasting model effective, just as a well-structured warehouse process is what makes any technology effective. Quantitative methods are scalable, objective and traceable: they are the foundation on which to build professional demand planning.

Advanced solutions: artificial intelligence, machine learning and big data

In recent years, demand forecasting systems based on artificial intelligence and machine learning have made extraordinary progress, pushing the frontier of forecasting well beyond what traditional statistical models can offer. These systems can simultaneously analyse a very large number of variables (historical sales data, weather information, search engine trends, macroeconomic data, social media signals) and identify complex, non-linear patterns that traditional models would fail to capture.

Machine learning, in particular, excels at handling variability: it can continuously adapt its parameters as conditions change, improving its accuracy over time and responding faster to changes in consumer behaviour or market conditions.

Big data further amplifies these capabilities, making it possible to integrate external information sources that enrich the forecasting model well beyond the boundaries of a company’s internal data. That is why, for over a year now, we have been working with a specialist provider that can significantly improve forecast accuracy by combining more than a dozen algorithms and comparing time series against more than 1,200 indirect variables. However, and this is a point we always stress in our consulting projects, even the most advanced AI and machine learning systems only work well if the underlying process is well designed and the data feeding them is of high quality.

A machine learning model trained on historical data that is dirty, incomplete or unrepresentative will produce unreliable forecasts with an algorithmic confidence that can be even more misleading than human error. Technology, even the most sophisticated, is a multiplier: it amplifies the quality of the process that supports it, for better or worse.

Practical applications: optimising inventory, transport and warehousing through demand forecasting

Demand forecasting is not an analytical exercise for its own sake: its value lies entirely in the operational decisions it enables and the concrete results it delivers across the supply chain. The most direct and impactful application is inventory optimisation. Accurate demand forecasting makes it possible to calculate reorder levels and safety stock far more precisely than traditional empirical methods, reducing tied-up capital without increasing the risk of stockouts.

At aBCD Consulting, we use demand forecasts as the primary input for sizing stock by item and by storage location, calibrating reorder parameters according to actual demand variability and supplier lead times. On the transport side, demand forecasting makes it possible to plan distribution flows far enough ahead to optimise loads, consolidate shipments and choose the most efficient transport modes. Avoiding urgent, partly loaded shipments is one of the most direct contributions demand forecasting makes to reducing logistics costs and CO₂ emissions.

In the warehouse, short-term demand forecasts guide the planning of operational activities: shift sizing, advance preparation of recurring orders and dynamic allocation of space according to expected peaks. A warehouse that knows in advance the volumes it will need to handle in the coming days works more smoothly, with fewer surprises, less overtime and more consistent service quality. In short, demand forecasting is the thread that connects strategic supply chain planning with day-to-day operational execution: the stronger and more reliable this thread, the more cohesively and efficiently the entire supply chain works.

Common mistakes to avoid in logistics demand forecasting

In our demand planning consulting work, we regularly come across the same recurring mistakes, regardless of sector or company size. Recognising and correcting them is often more valuable than introducing new tools or technologies.

The first mistake, and perhaps the most common, is confusing the forecast with the sales target: the forecast is not what the company would like to sell, but the best estimate of what it will sell. When the demand planning process is contaminated by budget targets or commercial pressure, forecasts become systematically optimistic, leading to excess stock and operational disappointment.

The second mistake is not measuring forecast accuracy: without a structured system for monitoring forecast error (MAPE, bias, tracking signal), it is impossible to understand where the model goes wrong and how to improve it. Demand forecasting is an iterative process: it only improves over time if it is fed with systematic feedback on its own errors.

The third mistake is using a single model for all items. Fast-moving products with stable demand, seasonal products, and products at launch or end-of-life stage require different forecasting approaches, and applying the same model to all of them produces unsatisfactory results for most of the catalogue.

A fourth very common mistake concerns data quality: historical data that has not been cleaned or corrected for past anomalies (exceptional promotions, stockouts, recording errors) produces distorted forecasts that cascade across the entire supply chain. Finally, one of the most costly mistakes is implementing advanced demand planning tools on processes that are not yet mature. Technology amplifies what it finds, and a disorganised planning process that is digitalised without first being redesigned simply becomes faster at producing wrong forecasts.