In today’s volatile market, traditional demand forecasting is no longer enough. This article explores why a shift toward decision-making that embraces uncertainty—using AI and Simulation—is critical for supply chain success.
In today’s volatile and complex global marketplace, traditional demand forecasting methods are no longer sufficient. As supply chains become more intricate and unpredictable, businesses face increasing challenges in predicting demand accurately. The rise of multinational operations, the rapid turnover of products with short life cycles, and the sheer diversity of markets have only added to the difficulty.
But focusing solely on improving forecast accuracy can lead to missed opportunities and increased costs. In an environment where uncertainty is the norm, the most successful companies are those that prioritize smart decision-making over precise predictions. Rather than relying on outdated methods that struggle to keep up with the pace of change, these companies are turning to advanced simulation and AI-driven techniques that embrace uncertainty, adapt to new realities, and drive better business outcomes.
This article explores why shifting from traditional demand forecasting to a more dynamic, decision-focused approach is crucial for staying competitive in today’s ever-changing market. By leveraging the power of AI-simulation, businesses can not only improve their forecasting but also make more informed decisions that lead to tangible benefits such as reduced inventory costs, improved service levels, and a stronger bottom line.
Over the years, significant criticism has emerged regarding the inability of forecast practitioners to “improve” the demand forecast especially in a context of market volatility or unexpected events. This has led to large swings in forecast accuracy, escalating errors, poor planning and decision-making for supply chain leaders.
One of the primary reasons for this forecast inaccuracy is the inherent uncertainty in predicting the future, which is exacerbated by unexpected events such as wars, pandemics, inflation, and natural disasters, due to insufficient historical data or understanding of their mechanisms.
Some events, involving trends or seasonality, can be predicted with high accuracy using key drivers like price or advertising. However, limited data availability or highly random demand, such as in the automobile industry with service part failures, makes demand forecasting nearly impossible using traditional methods.
Supply chain leaders often share laughs over the inevitability of inaccurate forecasts. One popular joke is: “There are two theories to getting an accurate forecast. Neither one works.” Another favorite one goes, “I have a demand forecasting joke, but it’s not accurate.”
While amusing, these jokes serve to highlight the inherent uncertainties in forecasting. That’s because, in the context of increasing supply chain complexity, countless internal and external factors can impact a product’s demand.
Factors that impact demand forecasts include:
Many companies assume that more data equals better predictions. But without the right insights, data can mislead, causing costly forecasting errors.
Example:
Sales of ice cream and sunblock lotion may be closely correlated, but one variable doesn’t cause the other. Both are affected by the season and the weather.
Generally, the business questions that supply chain managers need to ask should be trying to understand causation such as “Does increasing the amount we spend on marketing promotions lead to an increase in sales?”
This error can happen in both statistical and judgmental forecasting. Too often, supply chain managers latch onto specific correlations or overlook the obvious to confirm assumptions for a brand or product.
Sometimes, the situation is too complex for traditional forecasting methods. Demand patterns and relationships can be unstable and unreliable due to fluctuations, changing over time and making accurate predictions difficult. While some patterns are predictable, others require understanding influencing factors. Commercial advertising can gradually influence buying habits, whereas social media buzz can be sudden and unpredictable. Factors like duration, magnitude of change, and domain knowledge significantly impact forecast accuracy.
Our practice with customers showed that many forecasts used for decision-making lack a clear distinction between forecasting and a robust decision-making process. Ironically, while products are manufactured with strict guidelines that are within a .001 tolerance range, companies often rely on gut feelings or familiar yet ineffective techniques to forecast demand for these same products.
One subjective method is “target setting,” which focuses on sales goals rather than accurate forecasting. Due to their subjective nature, judgmental methods are not consistently accurate over time and are generally unsuitable for firms with many products and SKUs.
No longer assuming that demand magically happens and can only be estimated by understanding past events, companies have started using more sophisticated quantitative methods that are grouped into one of two categories: time series methods and casual methods.
Traditional forecasting such as time series methods assume that future sales will follow past patterns, considering trends, seasons, and cycles. But in today’s volatile market, that’s often not the case. Examples include naive models, moving averages, exponential smoothing, decomposition, and ARIMA (Box-Jenkins). Despite their success, these methods have limitations: they need a lot of historical data, adjust slowly to changes, and perform poorly for long-term forecasts. Smoothing weights are hard to find, and large data fluctuations can cause significant errors.
Casual methods model and forecast relationships between variables using techniques like simple linear regression, multiple regression, and ARIMAX, which includes explanatory variables. However, too many variables can worsen the model’s accuracy. Forecasting accuracy relies on a consistent relationship between variables and accurate estimates of independent variables. They also require larger data storage and are harder to systematize.
Traditional methods work when patterns are stable, but they often fail when the market changes. This is where simulation and AI shine—they adapt quickly to new realities.The true challenge lies in forecasting changes to patterns, relationships, or the system itself and determining timing and magnitude. Another key difference lies in how the information is captured, prepared, and processed, testing the method’s effectiveness.
Forecasting demand is more uncertain than calculating the orbit of a satellite. In business, predicting future demand is not an exact science; it’s more like trying to guess the number of raindrops in a storm.
Senior managers and forecast practitioners must accept some unexplained variance in predictions from both quantitative and judgemental methods.
For demand forecasting in complex and uncertain environments, simulation modeling combined with AI techniques offers advantages over traditional time series models and statistical methods.
AI-Simulation technology captures complex relationships, uncertainties, and interdependencies, providing a comprehensive view of system dynamics. This allows businesses to perform scenario analyses and “what-if” simulations, exploring a range of potential outcomes based on varying parameters.
Simulation models can be tailored to specific supply chain characteristics and challenges, integrating diverse data sources and operational constraints for precise forecasting outcomes. They excel at handling nonlinear relationships and complex patterns that challenge conventional techniques, allowing analysis under abnormal conditions, rare events, or new designs.
By combining AI with simulation models, companies can continuously learn from past data and adapt to changing conditions, leading to smarter decisions and better business outcomes.This adaptive process improves forecasting performance and decision making.
Large manufacturing companies are shifting to a decision-making process based on both data and domain knowledge. By implementing AI-Simulation technology, these companies improve decision making in uncertain environments, resulting in better demand forecasting, reduced safety stocks, lower storage and transport costs, and optimal inventory strategies.
Companies using AI-simulation have reduced inventory by up to 8%, improving service and saving hundreds of thousands of dollars. It’s not just about predicting the future—it’s about making better decisions today.”
The best decision is always going to be the one that takes into account this uncertainty and that leads to a positive outcome even when the future is going to be different than what was forecasted.
The future is uncertain, but that doesn’t mean your decisions have to be. By embracing advanced simulation techniques, supply chain leaders can turn uncertainty into a competitive advantage, positioning their companies for success in any market condition.
