Crystal Box – Chair Supply Chain of the Future

The project · from brief to prototype

What happened, step by step.

01

Challenge

Forecasters at major French companies are pushed to adopt machine-learning tools but cannot explain the predictions to stakeholders. So they often override even correct AI forecasts, and the investment goes unused.

02

What they built

An add-on for forecasting tools with two features. Explainability shows how the model reaches a recommendation and which factors drive it. Add Data lets forecasters feed in context such as exchange rates, weather or strikes that lies outside the training data.

03

Impact

Forecasters could present AI insights with confidence. For companies working with the chair, it addresses low AI adoption so their machine-learning investments start paying back.

Full project text

The team’s original texts from the project page.

A forecasting tool add on to explain AI driven predictions while enriching them with human knowledge

Challenge

Don't you hate making decisions without understanding the reasoning behind them? Don't you hate being asked to trust something you can't explain to others?
AI-driven sales forecasting promises accuracy but delivers opacity. Forecasters at major French companies face mounting pressure to adopt machine learning tools, yet they can't justify AI predictions to stakeholders. This black box problem creates a trust gap—even when AI predictions are correct, forecasters avoid them because they can't explain the logic behind the numbers. Without transparency into AI decision-making, valuable predictive insights remain unused, undermining the substantial investments companies make in advanced forecasting technologies.

Solution

Crystal Box addresses the core challenge of AI explainability in supply chain decision-making. Unlike traditional forecasting tools, it allows forecasters to see why the AI’s predictions change and adjust models with their own contextual knowledge by focusing on two key feature areas:

Explainability: helping users better understand and interact with Machine Learning models during forecasting decisions, providing clearer explanations of how AI recommendations are generated and what factors influence them.

Add Data: allowing users to integrate contextual factors - such as currency exchange rates, weather, or strikes - to adjust predictions based on knowledge that lies outside the model's initial training data.

Impact

Crystal Box transforms AI forecasting from a trust barrier into a competitive advantage. For users, it eliminates the fear of unexplainable predictions. Forecasters can now confidently present AI insights to stakeholders and make strategic decisions based on transparent reasoning.
For sponsors, this directly tackles the costly problem of low AI adoption rates that prevent companies from realizing returns on their machine learning investments. Instead of forecasters overriding accurate predictions due to lack of understanding, Crystal Box enables them to leverage AI capabilities fully. This accelerates feedback loops, improves resource allocation, and delivers the forecasting value companies expected from their substantial AI investments.

Gallery