From new tariffs and trade uncertainties to geopolitical tensions and extreme weather events, external forces have upended traditional approaches to demand forecasting. Among the most affected sectors are the FMCG and retail sectors, which have historically been more vulnerable due to changing consumer behaviors, rapid product evolution and complex global supply chains. And although these sectors were among the first to modernize and integrate new technologies, a considerable gap has emerged between the acquisition of AI capabilities and their effective implementation.
A Gartner June Report found that only 23% of supply chain leaders had a formal supply chain AI strategy in place within their organization. The report found that most CSCOs were taking an unstructured approach, focusing on short-term projects rather than long-term transformation strategies. This is partly due to increased pressure on C-suite executives to deliver rapid ROI. But while use cases can be useful in times of stability, those days are long gone.
Model Drift: the silent killer of precision
Supply chain managers are always on the lookout for model drift, when algorithms based on historical data lose their predictive power due to changing external conditions. These conditions are becoming more and more irregular and the impacts more and more serious. Excess inventory can jeopardize working capital, stock-outs can erode consumer confidence, and inaccurate forecasts can lead to poor pricing strategies that can reduce margins. According to McKinseyModel drift is one of the main reasons why AI implementations fail to scale effectively in businesses.
Unexpected tariffs were big accelerators of drift last year, and companies such as Walmart and Costco faced tariff boost. When costs fluctuate overnight, supply chains simply don’t have enough time to recalibrate, resulting in empty shelves or excess inventory. The situation is even more dire for businesses dependent on imported goods, where tariffs may increase retail prices from 3% to 5%. Static forecasting models are unable to keep pace with such a dynamic and unpredictable world, and drift has become a constant risk requiring constant vigilance.
The good news is that adaptive AI/ML capabilities have been designed for the moment. But for retailers to thrive, they must reinvent demand forecasting across three critical dimensions: signals, models and collaboration.
Three key levers of self-correcting demand forecasting
Better signals: seeing the invisible. Most traditional forecasting tools rely heavily on structured internal data (i.e. limited to what is available within their company): sales history, point-of-sale trends, inventory levels, among others. But there is also a whole world of external variables that can influence the evolution of demand and the way consumers purchase products. It is these unstructured signals that are often the first indicators of problems, and this is an area that supply chain leaders are actively addressing with applied AI.
For example, let’s say a retail distributor in the United Arab Emirates noticed a surprising drop in sales over the course of a month. Standard event calendars indicated nothing, until analysts conducted further research and discovered that regional flooding was likely causing product shortages. Traditional models would have missed this, and modeling one-off events is notoriously difficult. Future forecasts could use AI agents to detect anomalies and automatically investigate disruptions, enabling faster, more accurate and region-specific forecasts.
Better models: continuous and contextual AI. Even advanced machine learning models degrade over time if they are not modified or retrained. Adaptive AI models, like agentic AI, continuously learn from real-time data, making them resistant to drift and better aligned with market dynamics.
To use a hypothetical scenario, consider a beverage brand that faced a sudden tariff hike that raised the price of its flagship drink from $8 to $12, prompting price-conscious customers to turn to a competitor’s $10 alternative. While traditional models would not account for this change in behavior, an adaptive AI platform would have flagged the decline in sales within days, allowing the brand to adjust its promotions and save revenue. It is this type of robust modeling that is needed today.
Better collaboration: consensus forecasts based on GenAI and RGM linkage. Other major failures in current demand forecasting are due to organizational fragmentation and outdated consensus forecasts. Sales, marketing, operations and finance teams often operate in silos, creating conflicting forecasts that dilute execution. And even when all of these different signals are successfully combined into one model, you still need a group of planners from each department to sit around a table and agree on the path forward. This is not just an operational challenge; it can also be a clash of personalities.
However, using GenAI-based simulations, a company can reduce its planning cycle from 30 days to just seven days, by aligning its pricing, promotion and supply scenarios on a shared forecasting platform. Not only does this save time and increase forecast accuracy; this allows for faster responsiveness and can significantly improve internal collaboration. It is also worth noting that demand forecasting influences (and is influenced by) revenue growth management (RGM), particularly in areas such as pricing, discounting and promotion strategies. When RGM and demand forecasting co-evolve, margin results improve dramatically and businesses that connect the two unlock a powerful feedback loop.
Wide applicability
Demand forecasting optimization doesn’t just apply to retail and consumer goods companies. AI-driven innovations have high cross-industry relevance, including:
Manufacturing. Predictive maintenance models improve when upstream retail demand signals are taken into account, allowing production schedules to better match market reality.
Automotive. Real-time tracking of changes in consumer interest, economic trends and weather events can help improve dealer inventory.
Industrial. Companies that produce tangible goods for manufacturing, construction, infrastructure and defense systems already benefit from AI-Driven Demand Sensingespecially when they combine IoT, weather and macroeconomic data into their planning platforms.
The FMCG and retail sectors can no longer afford to rely on fragile, backward-looking forecasting systems. With adaptive AI, supply chain managers can anticipate problems by detecting anomalies before they develop further, updating models as market conditions change, and aligning departments around shared, data-driven decisions. In a world of constant disruption, flexibility is quickly becoming the new norm, and only the most resilient will survive.
Sunder Balakrishnan is Director of Supply Chain Analytics at LatentView analysis.