Evaluating railway performance
How can you reduce the cost and time of refining and developing products that will resonate with your customers?
A key indicator of a successful and profitable product is positive consumer feedback. However, acquiring this is expensive and can only be gathered after product development.
Partnering with a global consumer goods organisation, we created a data-led approach that explores the relationship between ingredient combinations, demographic data, and consumer preference scores. Revealing the relationship between these variables means the client can make accurate data-led forecasts for successful product development decisions, that could lead to cheaper costs.
DECODING THE DRIVERS OF CONSUMER PREFERENCE
The Challenge
Understanding what drives consumer preference is a persistent challenge in the consumer goods industry. Traditional product development relies heavily on physical trials and consumer testing. These are expensive, time-consuming, and can yield limited insight into product performance.
Our client wanted to shift from reactive testing to proactive design. They needed a way to predict how much consumers would like a product before it is developed and trialed, and to understand which recipe components were driving any preferences. The solution had to be statistically robust, explainable, and scalable across product lines.
Initial modelling demonstrated that average consumer scores across the population data lacked sufficient variability to support meaningful prediction. This raised a critical question: Could important insights be hidden when we only look at the traditional approach that considers average consumer response?
MODELLING PREFERENCE WITH EXPLAINABLE, INGREDIENT-LEVEL INSIGHT
The Solution
To uncover maximum decision intelligence potential, we designed a multi-phase programme of work, applying principles from experimental design to identify a minimal yet information-rich set of product variations to test. This reduced the number of physical trials required while maximising the insight gained.
We introduced a new layer of analysis: explainable modelling within consumer subgroups. Using extensive data from previous trials and focus groups provided by the client, we segmented the population and explored how ingredient factors influenced consumer scores within each group.
The approach combines statistical modelling, distributional analysis, and hypothesis testing to uncover relationships that are not visible at the population level. We analysed the full distribution of liking scores, enabling a richer understanding of consumer behaviour.
Explainability was central to the project. We used interpretable models to quantify the influence of each recipe component on predicted liking scores. This allowed us to show, for example, how increasing ingredient content might improve liking in one subgroup but reduce it in another. These insights were presented in a way that could be understood and trusted by both technical and commercial teams within the client’s business.
FASTER, SMARTER DECISIONS IN PRODUCT INNOVATION
The Impact
The client now has a trustworthy, strategic decision intelligence model to support their product development. They can:
- Reduce the cost and time of product development while increasing the likelihood of market success.
- Identify which ingredient changes are most likely to improve consumer preference.
- Understand how different consumer groups respond to different product attributes.
This work has enabled faster iteration, more targeted innovation, and a stronger foundation for data-led decision-making. The project demonstrates how expertly applied modelling can unlock value in complex, high-value environments.
By combining statistical rigour with domain understanding, we delivered a solution that is not only accurate, but also transparent and actionable, supporting decision makers to uphold accountability for safe, compliant and scalable deployment in product trials.
