The Key to Resilient Supply Chains

In an era of global volatility, the “gut feeling” of experienced planners is no longer enough to navigate the complexities of modern supply chains. From global pandemics to sudden market shifts, the limitations of traditional forecasting are becoming painfully clear. But as businesses flock towards Artificial Intelligence, a critical question remains: When does AI truly provide a competitive edge, and what does it take to get there?

Red cargo ship Charles Island leaving an industrial port at sunset, cranes on shore and a tugboat nearby
Young man in headphones speaking into a studio microphone beside an audio mixer in a home podcast setup.
We sat down with Nicolas Pesci to discuss the real-world value of AI in supply chain forecasting.

The Limits of Tradition

To answer this, we sat down with Nicolas Pesci, author of a comprehensive thesis at the Berlin School of Economics and Law (HWR Berlin), who investigated the real-world value of Machine Learning (ML) versus traditional forecasting methods. His findings provide a scientific anchor for what we build with prognotix.

For decades, methods like Exponential Smoothing (Holt-Winters) or ARIMA have been the workhorses of demand planning. They work well, until they don’t. Pesci’s research highlights that these statistical models often struggle with non-linear patterns and high volatility.

“Traditional models assume the future will look largely like the past,” Pesci explains. “But in a digitalized world where social media trends or weather changes can flip demand overnight, these linear assumptions lead to costly stockouts or bloated inventories”.

Machine Learning doesn’t just offer “better guesses.” According to the thesis, the true value lies in three pillars:

  1. Automation of Complexity: ML can process thousands of SKUs and integrate external “features”, such as weather data, market trends, or promotional activities, without manual intervention.
  2. Scalability: While a human team hits a ceiling when managing large portfolios, ML-based systems like prognotix can scale effortlessly across diverse markets and product lines.
  3. Self-Correction: Unlike static models, ML learns from its own errors, continuously adapting to new consumer behaviors without needing a consultant to recalibrate the parameters every month.

The "Black Box" Challenge

One of the most striking insights from Pesci’s expert interviews is that the biggest hurdle isn’t always technical, it’s organizational. “Many planners are skeptical of ‘Black Box’ results,” says Pesci. If an algorithm predicts a 20% spike in demand, but cannot explain why, decision-makers hesitate to trust it.

This is where the concept of Data Literacy becomes vital. For a company to succeed with AI, it’s not enough to just buy the software; the team must be empowered to interpret and communicate these results. At prognotix, we bridge this gap by focusing on transparency and user-centric interfaces that turn complex data into actionable insights.

Four small matte black square gift boxes on a white surface, one box open with its lid tilted.
Photo by Alleksana @ Pexels

Is ML Always the Answer?

The thesis is refreshingly honest: ML is not a silver bullet. For companies with very stable, simple demand patterns and limited data, traditional methods might still be more cost-effective. Implementing ML requires “fuel”, high-quality, clean data. Pesci emphasizes that the Return on Investment (ROI) is highest for companies facing high complexity and those willing to invest in their digital maturity.

The integration of Large Language Models (LLMs) to explain ML results and the move toward more “interactive” forecasting systems are the next frontiers. Pesci’s research confirms what we at prognotix live every day: the most resilient supply chains of tomorrow are those that combine human expertise with the analytical power of science-backed AI.

This post is based on the findings of the bachelor thesis “KI-basierte Nachfrageprognose in der Supply Chain” by Nicolas Pesci (2026).