How XAI builds trust in machine learning forecasts
Ruben Crevits - June 20, 2023
Trust is a big challenge for companies when it comes to fully embracing AI solutions. So, how can we overcome this hurdle? The recent Data2Move seminar I attended, hosted by the European Supply Chain Forum (ESCF) in Eindhoven, provided valuable insights on enhancing human-AI interaction.
The seminar focused on using AI in demand forecasting, an area where machine learning has already achieved impressive successes, like the renowned M5 competition. And let me tell you, these successes are much needed because demand forecasting has become way more complex in recent years.
During the seminar, Christina Imdahl, the research director at Data2Move, talked about what she called the "decision explosion" in supply chains. Basically, it means that crucial decisions need to be made faster than ever before, with shorter time intervals, and taking into account a whole bunch of constraints and considerations. It's a lot to handle. That's why some level of digitization and automation is necessary to keep it manageable. A perfect match for AI, don’t you think?
The challenges of trust in AI forecasting: a fascinating research project
Imdahl shared an intriguing research project she led, where she and her team explored the issue of trust in AI forecasting. They wanted to understand how human planners in a consumer goods company interacted with an AI forecasting model. What they discovered was fascinating. The supply chain planners tended to adjust the forecasts for fresh, expensive, or discounted products, relying on their experience and intuition. Surprisingly, their adjustments were based on a mistaken belief that the AI model didn't consider seasonality or promotional campaigns. Unfortunately, this unintentionally reduced the accuracy of the forecasts.
Blog post
Implementing XAI in practice
Now, let's discuss how we can address this trust challenge. In our session at the seminar, my colleague Thiebe Sleeuwaert and I showcased how we put the concept of eXplainable AI, or XAI, into action at OMP. We presented three practical cases where we used deep learning to interpret demand sensing signals. But here's the crucial part: for each forecast, we provided a clear explanation of the AI’s scope and limitations, leaving no doubt about what has been taken into account and what has not.
Blog post
Our presentation triggered a lot of interest among the attendees. They were genuinely curious about our decision to use a deep learning model and how our customers responded to this functionality. Attendees were particularly eager to learn more about the forecast patterns and the methods we used to prove the forecast quality. We had an engaging discussion that shed light on the practical aspects of implementing XAI.
Empowering real-world supply chains with AI forecasting
The challenge of trust in AI solutions is something supply chain innovation leaders face worldwide. They see the immense potential of AI but encounter resistance from skeptical planners who doubt the inputs and outcomes. This is where XAI steps in as a game-changer. By enhancing the human-AI interaction, XAI elevates trust levels, enabling AI forecasting to revolutionize real-world supply chains.
In a world where trust is essential, the success of machine learning in forecasting raises concerns about its "black box" nature. That’s why XAI has become a hot topic. If you're curious about the level of explanation ML-powered forecasting models should provide, explore this captivating blog post.
Ruben Crevits
Senior Data Scientist at OMP BE
Biography
Ruben specializes in statistical and machine learning forecasting. He passionately supports fellow data scientists, striving to make AI valuable for our customers.
