By Paul R Salmon FCILT, FSCM
In the era of big data, machine learning (ML), and artificial intelligence (AI), traditional time series forecasting is increasingly being described as “old hat.” Once the cornerstone of supply chain and logistics planning, its limitations are becoming stark as organisations face unprecedented levels of complexity and volatility.
But is time series forecasting really obsolete? Or does it still have a role to play in modern, AI-driven logistics environments – including defence supply chains where uncertainty and resilience are critical?
📈 The Legacy of Time Series Forecasting
For decades, time series models such as ARIMA, exponential smoothing, and Holt-Winters formed the backbone of demand planning. These methods worked by analysing historical patterns to extrapolate future demand.
In predictable environments, these models were invaluable. For instance, in defence logistics they have long been used to forecast:
Routine demand for consumables such as fuel and food. Maintenance cycles for military vehicles and aircraft. Spare parts requirements for supply depots.
However, the assumptions underpinning these methods – that the future broadly resembles the past – are increasingly under strain.
🌐 The New Reality: Complexity and Volatility
Modern supply chains operate in an environment marked by:
Geopolitical shocks: conflicts in Ukraine, the Red Sea and Taiwan have disrupted global trade. Rapid technological change: autonomous systems, additive manufacturing (3D printing), and IoT sensors are altering demand patterns in defence. Environmental factors: climate-related disruptions are impacting global logistics networks.
In such environments, relying solely on historic demand patterns can leave organisations exposed.
For example, during Operation Pitting (the UK’s evacuation of Afghanistan in 2021), demand for aircraft spares and fuel surged unexpectedly. Traditional time series forecasts, calibrated on peacetime consumption, could not have predicted such an event.
This is where AI and ML offer new possibilities.
🤖 What AI/ML Brings to Forecasting
Unlike traditional time series models, ML and AI approaches can:
Ingest high volumes of diverse data: including weather, social media sentiment, satellite imagery, and even maintenance logs. Capture non-linear relationships: between variables such as operational tempo and spares usage. Adapt dynamically: learning from new data streams to update forecasts in near real-time. Predict rare events: by identifying patterns invisible to human analysts or traditional models.
In defence logistics, imagine an AI system that correlates:
Flight hours logged on Typhoon aircraft Supply chain disruptions in key OEMs Geopolitical risk indicators
… to predict spares shortages before they materialise.
In commercial logistics, Amazon and DHL already deploy similar systems to fine-tune inventory and delivery forecasts.
⚖️ Why Time Series Isn’t Completely Dead
Yet, it would be a mistake to write off time series forecasting entirely. It remains:
Fast and lightweight: useful for short-term predictions where patterns are stable. Transparent and explainable: critical in highly regulated environments like defence. Resilient in small data contexts: where ML models would overfit.
For example, at a forward operating base, demand for bottled water may follow stable daily patterns. A simple time series model can handle this without requiring massive datasets or computing power.
In fact, many modern tools – such as Facebook’s Prophet – blend traditional time series techniques with ML to create hybrid models.
🚀 The Case for Hybrid Forecasting
Rather than seeing time series and AI as rivals, forward-thinking organisations are combining them:
Defence Support Chains: Hybrid models are being explored to balance peacetime efficiency with surge capability during operations. OEMs and Defence Suppliers: Use ML to detect anomalies in consumption patterns and traditional forecasting for steady-state replenishment.
The UK’s Defence Support Transformation Programme has already highlighted the need to move from reactive to predictive support. AI-enhanced forecasting is a cornerstone of that vision.
🛡️ Implications for Defence and Logistics Leaders
Leaders in supply chain and logistics should ask:
Where do we need advanced AI? Highly volatile, high-value parts or operations benefit most. Where is traditional forecasting sufficient? Low-variability, high-volume items may not justify the complexity of AI. How do we build explainability into AI models? Defence stakeholders often demand transparent, auditable decision-making.
The challenge isn’t whether to replace time series forecasting, but how to integrate the best of both worlds.
✍️ Final Thoughts
In a world defined by complexity, the blunt tools of the past are no longer enough. Time series forecasting isn’t “dead,” but it is no longer sufficient on its own.
For defence and logistics organisations, the future lies in augmented forecasting systems that combine the transparency of traditional methods with the predictive power of AI and ML. Those that make this transition early will be better positioned to respond to shocks, reduce costs, and maintain readiness in the face of uncertainty.
