Case Study
Developing new flavours
The challenge
Companies developing new flavours, fragrances, food and beverage products rely on extensive sensory evaluation testing to find the product formulations which command the highest consumer acceptance. The testing provides critical input in the product development process and guides the decisions about which product variation to launch. However, it is cost prohibitive to collect a lot of data and as a result companies are forced to make important decisions with insufficient information.
Givaudan, the world’s largest company in the flavour and fragrance industry, had a very efficient way to let panelists evaluate flavours. They were conducting the following experiment:
- They took seven ingredients to create a target flavour composition.
- They synthesized a total of 40 flavours varying ingredient volumes.
- They created a focus group of 70 people to evaluate the flavours.
- The subjects would rate flavours with a score from 1 to 9.
- The total data set contained 70 x 40 evaluations of eight columns - seven ingredients and the main Key Performance Indicator - the liking score.
Even with the best coordinated process, this method of collecting data was extremely time consuming and expensive.
Givaudan was looking for ways to design and launch new flavours faster and outsmart the competition.
The typical problems with sensory evaluation testing
It takes 3+ weeks to find people to bring to the lab.
Thousands of dollars spent to accommodate the 70+ test subjects.
A normal human can only evaluate about 15 flavours before they lose the ability to distinguish between them.
The subjects are from the target population of consumers, but they are not experts, so there is a lot of “noise” in the data.
Different people have wildly varying “likes”.
Even with all this testing, there is still uncertainty if the best combination will be chosen.
The DataStories solution
Givaudan sent us their experimental data in Excel and confirmed the main things they wanted to discover from their data. The following questions were asked:
Which ingredients drive liking?
Which optimal combination of ingredients will make a flavour that many people like a lot?
We analyzed the experimental data and built robust predictive models for each subject. We segmented the panel into groups of people driven by the same ingredients (in the same direction). We delivered predictive models which allowed Givaudan to evaluate millions of product combinations in the same search space without calling any more people to the lab.
The impact we created
With DataStories solution, Givaudan was able to:
Shorten the time to conduct focus groups from 6 months to 2 weeks (92% reduction).
Obtain immediate insights into what drives liking instead of a lengthy process which took several months.
Just plug the flavor data in the predictive models and predict liking scores with confidence instead of having to search for 40-80 people to evaluate a new flavour.
Evaluate millions of flavors per session instead of just 13-16 flavors per session.
So if you are a flavourist, don’t worry that the tools you currently have cannot do this type of advanced modelling. With our expertise and technology, we can help you create flavours for 20x less money and for 6x less time.
Related Publications:
- Katya Vladislavleva, Kalyan Veeramachaneni, Matt Burland, Jason Parcon, Una-May O'Reilly. Knowledge mining with genetic programming methods for variable selection in flavor design. In Juergen Branke et al. (Editors), GECCO'2010: Proceedings of the 12th annual conference on Genetic and evolutionary computation, pages 941-948, Portland, Oregon, USA, 2010.
- Kalyan Veeramachaneni, Katya Vladislavleva, Matt Burland, Jason Parcon, Una-May O'Reilly. Evolutionary Optimization of Flavors. In Juergen Branke et al. (Editors), GECCO'2010: Proceedings of the 12th annual conference on Genetic and evolutionary computation, pages 1291-1298, Portland, Oregon, USA, 2010.
- Kalyan Veeramachaneni, Ekaterina Vladislavleva, Una-May O’Reilly. Knowledge mining sensory evaluation data: genetic programming, statistical techniques, and swarm optimization. In Journal on Genetic Programming and Evolvable Machines, March 2012, Volume 13, Issue 1, p.103-133
- Katya Vladislavleva, Kalyan Veeramachaneni, Una-May O'Reilly. Learning a Lot from Only a Little: Genetic Programming for PanelSegmentation on Sparse Sensory Evaluation Data. In A.I.Esparcia-Alcazar et al. (Editors), Proceedings of the 13th European Conference on Genetic Programming, EuroGP2010, Lecture Notes on Computer Science, Volume 6021, pages 244-255, Istanbul, 2010, Springer.