Every strong supply chain begins with one crucial skill and that is understanding what customers will want and when. That’s what demand forecasting is all about. Demand Forecasting is a critical foundation to most other operational decisions in supply chain planning. It will drive optimum inventory level, efficient production planning, better resource allocation and improved logistics.
How India’s Market Has Changed
Over the past two decades, the Indian market has completely transformed. First came the rise of e-commerce, which changed how people shopped. Then came quick commerce, which made consumers expect faster deliveries and instant availability.
Companies have had to deal with a significant shift in the proportion of business that is driven through these channels than the earlier brick and mortar stores. Add to this the growth of the middle class, rising disposable incomes that are driving premiumisation and the increasing competition driven by strengthening of regional players and smaller focussed startups. Together, these factors create a complex mix that makes demand forecasting and the entire supply chain planning process a highly challenging task.
Time-Series- the OG forecasting method
Traditionally the standard approach to demand forecasting was time-series analysis. It used past sales data to find patterns in trends and seasons through methods like moving averages and exponential smoothing and ARIMA. While this approach worked fairly well, it started to fall short when markets became more unpredictable.
Take festivals like Diwali for example. Over the past six years, its date has shifted from October 21 to November 14, making it difficult for time-series models to capture the true seasonal impact. With this, regional differences in buying behaviour and the growing share of sales through online channels, forecasting demand accurately becomes very challenging. In most cases, planners have to manually adjust the numbers to make sense of it all.
The Rise of Machine Learning in Supply Chain
With the advent of powerful computing, the explosion of big data, and easy access to cloud platforms, machine learning has transformed how companies forecast demand. Unlike traditional time-series models that focus on a single variable, machine learning can consider many different factors that influence the demand.
Factors such as holidays, events, pricing, promotions, and even weather can all be included as inputs to generate a more accurate forecast. By understanding how these multiple factors interact, often in complex and non-linear ways, machine learning can reveal patterns that older models simply cannot capture. The result is a forecast that’s far more reflective of real-world behaviour.
The bandwagon trap
It’s no surprise that companies have moved to machine learning–based forecasting models. Many have jumped onto the bandwagon, but the reality has not always lived up to the hype. In several cases, the results have been not much better than what traditional time-series models could deliver. Why is this the case?
The problem often lies not in the model, but in the data. Many businesses are “ambition-rich but data-poor.” They understand the value of advanced models but lack the depth of information those models need to work effectively.
Machine learning is only as good as the data it’s trained on.
Without reliable, historical data on key drivers like promotions, pricing, holidays, or marketing spend and if you cannot capture this information accurately for future periods, then even the smartest algorithm cannot generate a precise forecast. Without these necessary driver data, the quality will be just about equal to than the time-series based models.
So where do I start?
The best place to start is with the data you already have within your systems and data warehouse. Most companies have dependable order or shipment data in their ERP systems. If that is the only driver data you have, then time series models might still rule the roost for you.
The next step is to enrich your data by adding known external factors, such as national and regional holidays or major events. Even this simple addition can improve forecast accuracy and significantly reduce the manual effort required to clean the data when using time-series models.
The most advanced companies today do not just stop at sales history. Over time, as your organization becomes more data-mature, you can start including additional drivers like pricing, promotions, and marketing events, usually captured through Trade Promotion Management tools or enterprise data warehouses. This is where the true power of machine learning begins to show. Weather data or broader economic indicators can also play a role, though typically only for very short-term or long-term forecasting.
Demand forecasting is essential for effective supply chain planning. Accuracy is a journey that depends on the quality and depth of your data. For some, traditional time-series models remain practical and reliable. For others, integrating multiple drivers unlocks the true potential of machine learning. The key is not to chase buzzwords, but to build a forecasting approach that matches your data maturity, business needs, and long-term strategy. Because the goal is not just better prediction, it’s smarter, faster decision-making in an increasingly complex market.