Digital twins in sensory research

But even the best sensory panels operate in the real world. Panelists can be absent, study designs can vary, attributes maybe missing, and gaps in the data can appear. When this happens, the impact can be bigger than expected: reduced analytical power, slower reporting, lower comparability across studies and, in some cases, costly retesting.

This challenge becomes even more important in high-throughput research environments, where several studies may need to be completed in a short period of time. In those cases, speed and continuity are critical.

At Haystack, we asked a simple but powerful question: what if we could create a reliable digital version of an expert panelist, not to replace them, but to support the research process when data is incomplete?

And that is exactly what we have been building: digital twins.

‍

What this means for your research

For clients, the benefits are very concrete.

Digital twins keep research teams moving when real-world research gets messy. When a panelist is absent, the study continues without delay. If a few scores are missing, the dataset draws on participant-specific data rather than generic averages. And when timelines are tight, teams work with a complete and robust view of the results.

Importantly, this is not a one-size-fits-all model. We build digital twins around the client’s own available data, category and panel context. We work iteratively and validate the model on the specific dataset it needs to support. This avoids a black-box approach or a generic assumption that the model will work in any context. Instead, we tailor and test the solution for the specific category, panel and dataset before applying it.

This supports faster reporting, fewer repeat tests and less pressure on expert panels. It can also help reduce panelist fatigue, because not every gap needs to result in extra fieldwork.

The approach is especially relevant for clients running frequent sensory studies, long-term tracking programs or high-volume innovation pipelines. In these contexts, even small efficiency gains can have a large impact over time.

Digital twins support product innovation, reformulation, quality control, competitive benchmarking and sensory tracking. They help teams compare products more consistently, protect decision quality and move faster from data to action.

In short, they help clients make better decisions when conditions are not perfect. And in real-world research, they rarely are.

‍

Beyond averages: What are digital twins?

Digital twins are synthetic models trained on historical panel data within a specific product category. They are designed to reproduce an individual panelist’s scoring pattern across products and attributes in that category. The model also takes into account the product context, the relationship between attributes, and the way the panelist behaves compared to the rest of the panel.

So when data is missing, a digital twin does not simply replace it with an average. Instead, it makes a more individual estimate based on how that specific panelist has scored in the past, in similar contexts.

That difference matters. Filling gaps with averages may seem harmless, but it can quietly flatten variance and weaken the relationships between attributes. Yet this underlying structure is exactly what analyses such as PCA (Principal Component Analysis), sensory mapping and benchmarking rely on. A digital twin helps preserve more of that structure by taking into account each panelist’s scoring style, the product context and the relationships within the data.

To validate this approach, Haystack has tested digital twins by deliberately removing data and checking how well the model could estimate the missing values. This included missing attribute scores, missing product data and missing panelists.

Across our validation work, digital twins consistently came closer to the actual scores than simple averages. Even when several panelists were removed from the data, the model was still able to accurately predict their actual scoring patterns. In practice, this means digital twins can help increase sensory capabilities, both in general and even more strongly in specific cases or projects. They do this not by replacing the panel, but by making better use of the expert data that already exists.

‍

Ready for research that keeps moving?

Missing data will always be part of real-world research. Panelists will sometimes be absent, studies will not always be perfectly complete, and timelines will continue to be tight.

The opportunity is to deal with those realities more intelligently.

At Haystack, we develop digital twin solutions that help clients protect research quality, reduce operational friction and move from incomplete data to confident decisions. Because the future of sensory research is not about choosing between speed and quality. It is about building smarter systems that deliver both.

Curious how digital twins could strengthen your research process? Let’s explore what is possible for your category, your data and your innovation challenges.

‍