Material innovation hinges on understanding the complex interplay between formulation parameters, processing conditions, and performance outcomes. Yet, the increasing fragmentation and volume of materials data makes it challenging to extract actionable insights.

In this talk, we demonstrate how machine learning (ML) can be applied to uncover hidden correlations and predict material behavior in complex datasets. At the core is an AI-based tool that enables accurate, data-driven decisions—supporting formulation optimization, material innovation, and faster time-to-market. This capability is embedded within a broader Material Intelligence Platform (MIP), which structures and connects the underlying data.

We showcase an ML use case from the paint and coatings industry, where a regression model is trained on 80% of historical formulation data—such as viscosity, ash content, and residual humidity—to predict key performance metrics. The remaining 20% is used for validation, demonstrating high predictive performance and enabling researchers to pre-evaluate and optimize new formulations efficiently.

Looking ahead, we explore multimodal learning by integrating external sources—such as publications, patents, and databases like CAS—into the ML pipeline. This expands predictive capabilities beyond internal datasets, allowing researchers to forecast viable formulations even in early-stage R&D.

Key outcomes include:

• Predictive formulation optimization using supervised ML
• Fewer physical experiments and faster development cycles
• Improved transparency across research and development (R&D)
• Outlook toward combining structured lab data and unstructured scientific knowledge

This work positions AI as a driver of predictive innovation in data-driven materials science.