From Physical Data to Sensory Insight: Supporting Enhanced Product Testing

Could analytical measurements predict how a product tastes before a human ever tries it? That was the challenge facing a global food manufacturer determined to accelerate innovation without compromising quality.

Traditional sensory panels with human tasters are costly, slow, and subjective. Smith Institute created an explainable modelling framework that links comprehensive mechanical measurements to human sensory feedback.

This approach reduces reliance on lengthy panel testing, speeds up product development, and builds trust in data-driven decisions. The result was a bespoke modelling framework that can predict accurate product rankings from analytical measurements. Harnessing the potential of data has created a faster, evidence-based process that empowers more efficient and effective product development.

TURNING MECHANICAL SIGNALS INTO MEANINGFUL INSIGHT

The Challenge

In consumable product development, understanding how a product feels and breaks down in the mouth is critical to predicting market success. Traditionally, this relies on trained sensory panels to score attributes such as firmness, elasticity, and breakdown. While effective, this process is expensive, time-consuming, and inherently subjective.

The client created an innovative advanced measurement device that captures detailed mechanical data. The ambition was to use this data to predict the trained sensory panel scores and reduce dependence on panels. However, the relationship between measurements and sensory attributes was complex and poorly understood. Adding to the challenge, sensory scores themselves are noisy, containing random errors or irrelevant variations, making it difficult to identify true product rankings. Without a robust modelling approach, the client risked slower innovation and missed opportunities in a competitive market.

BUILDING TRUSTWORTHY PREDICTIONS

The Solution

Smith Institute delivered a three-phase programme designed to transform raw physical data into actionable sensory insight.

Phase 1: Feature Engineering and Assessment

We began by aligning the mechanical data with human sensory feedback. This stage focused on cleaning and structuring the raw signals to ensure consistency and reliability. Using physics- and statistics-based analysis, we built a set of explainable features grounded in the complex mechanics of human texture perception that highlighted relationships with the sensory attributes, setting the stage for future predictive modelling.

Phase 2: Predictive Modelling with Explainability

We explored approaches to build predictive models for selected sensory attributes using the most informative features. Explainability was prioritised throughout, using interpretable models such as regularised linear models, decision trees, and ensemble methods with clear visualisations of feature contributions. This transparency was essential for building trust and enabling adoption within business-as-usual operations.

Phase 3: Noise Reduction and Ranking Accuracy

The final phase addressed a critical barrier: the inherent noise in sensory data. We implemented recently published mathematical methods, applied here for the first time, to robustly sort and rank sensory attributes, revealing the underlying structure of sensory perception. Building on these insights, we developed a bespoke model capable of inferring true product rankings directly from mechanical measurements. This closed the loop between mechanical and sensory domains, providing a scalable and extensible framework for future formulations and datasets.

FASTER, SMARTER PRODUCT DEVELOPMENT

The Impact

The client now has a validated, explainable modelling framework that transforms analytical measurements into reliable sensory predictions. This will enable them to:

Forecast sensory scores for new formulations without immediate reliance on panels.
Identify which mechanical features drive specific sensory responses.
Reduce development time and cost by focusing human testing resources where they add the most value.
Build organisational confidence in AI tools through transparent, interpretable outputs.

This shift moves product development from reactive testing to proactive design. Decision-makers gain a scalable, evidence-based process that supports faster innovation and smarter investment. By combining domain expertise with rigorous statistical modelling, Smith Institute delivered a solution that is technically robust and operationally relevant, breaking new ground in data-driven R&D in consumer goods.