Understanding AI in Demand Planning: Transparency & Explainability

Introduction: The Complexity of Demand Planning

Demand planning involves juggling hundreds or thousands of products (SKUs), numerous locations, and a constantly shifting customer base. This complexity makes the quest for effective and user-friendly demand planning solutions more challenging but also more crucial.

The inevitable question: How does the solution work exactly?

Regardless of how user-friendly a solution is, the question of the logic behind is inevitable. Why does it generate a specific forecast? The reasons for this curiosity are varied:

1. Curiosity and Learning: Users may wish to understand the system for their knowledge enhancement
2. Threat Perception: Some may view automated forecasts as a threat to less advanced methods, which they are
3. Past Inaccuracies: A history of inaccurate forecasts breeds a desire to understand and possibly mistrust the system

These are legitimate concerns. Understanding the 'why' and 'how' behind forecasts is not just a matter of curiosity but also of operational necessity.

The Problem of the 'Black Box'

In many scenarios, as long as the forecasts are accurate, the underlying mechanism may not concern everyone. However, the moment accuracy dips, even those who were previously indifferent then start to question the system's inner workings. A 'black box' approach, where the process is opaque, might deliver accurate forecasts but fails to maintain long-term trust within an organization. Transparency is key to building and sustaining confidence in the system.

Explainable AI: A Diverse Spectrum of Understanding

Explainable AI (XAI) is about making the AI's decision-making process transparent, but the level of explanation depends greatly on the audience. The spectrum ranges widely:

• For the Layperson: Simplified explanations that outline basic principles and outcomes without delving into technical intricacies
• For the Business User: More detailed insights that link forecasting outcomes with business strategies and assumptions
• For the Expert: In-depth technical explanations suitable for those with advanced knowledge in the field, like a PhD in mathematics

Each level of explanation caters to different needs and understanding levels, ensuring that the AI's workings are cleared up appropriately for each user.

Why Explainability Matters in Demand Planning

In demand planning, different product categories may require distinct forecasting models. It's crucial not only to choose the right model for each category but also to be able to explain why a particular model was selected. This transparency:

• Enhances trust in the system's recommendations.
• Allows users to understand the rationale behind forecasts for different categories.
• Facilitates better alignment of forecasting models with business strategies and market dynamics.

Conclusion: Balancing Complexity with Clarity

In conclusion, while demand planning solutions must tackle the inherent complexity of managing vast arrays of products and fluctuating market conditions, they also need to be transparent. Explainable AI bridges this gap, offering clarity and insight into the AI's decision-making process, tailored to the understanding level of different users. This balance of complexity and clarity is vital for building a dependable, efficient, and trusted demand planning system.

In Horizon, you can always dig deeper and find out why certain models were used for a specific category. In this manner, you’re able to learn more about patterns & trends for certain types of products or for different regions.

If you want to see how Horizon does this, you can watch this brief video.