Consumer Goods R&D with Automated Product Stability Forecasting
Consumers today are more educated than ever, which has driven a change in preferences towards, among others, increased transparency in ingredient sourcing, waste and product testing. This has led companies to seek new ways to stay competitive — only to find that the necessary experimentation and adjustments can be highly time- and resource-consuming. As a result, consumer goods manufacturers have seen a shift towards digitalization and automation in testing to reduce costs and time to market.
Our team of data and analytics experts at Starschema recently helped a Fortune 50 cosmetics company digitalize their testing process to speed up the research and development cycle without compromising quality along the way. The project offers a good look at how an industry leader approaches a cutting-edge forecasting solution that’s also relatively easily applicable for any business that does comparable R&D work, so let me take you deeper!
“Will This Face Scrub Blend?”
The use case that inspired the company to attempt to streamline its R&D lifecycle was not unlike the viral series of infomercials titled “Will it Blend?”. In it, all sorts of unusual items got dropped into a blender, with at least semi-expected results. The key here is the expectation: nobody was really surprised to see sophisticated devices like smartphones emerge, at the very least, unusable from an industrial-grade blender, but the exact nature and degree of the effect became the stuff of much online speculation.
Companies conducting large-scale R&D operations can have a similarly solid idea of how a product will behave over time when exposed to various effects such as temperature, humidity and light. But while researchers may make highly educated guesses based on substantial domain knowledge and experience, to ensure consistently reliable and actionable forecasts at the scale required by an enterprise would require customized state-of-the-art data solutions.
Forecasting product stability is a critical new frontier in product development, as it ensures that the products will maintain their efficacy and safety throughout their shelf life. Moreover, stability forecasts enable companies to optimize their product formulations, reduce the risk of product recalls and minimize waste — already good business in itself, but even more so considering the increase in customer awareness and demand that I mentioned in the opening.
Our team ultimately delivered a complex solution involving data engineering, visualization and science that automates the analysis of the ingredient concentration of in-development products to help forecast their stability. This reduces the time and cost of the product development process while achieving similar, or even improved, results.
The Three Main Components
Our solution for digitalizing the testing process comprised three main components: ETL processes, a dashboard and an algorithm, all informed by close collaboration between subject matter experts (SMEs), scientists and managers. The benefits of each component would also be individually significant for the client.
The ETL processes enable the company to extract and transform data from various sources. We worked closely with SMEs to build ETL processes that are tailored to the company’s specific needs to ensure that they are able to make more data-driven decisions.
The dashboard promotes this data-driven decision-making by making it easy to consume and engage with insights without deep data science or programming knowledge. This way, managers can monitor testing performance via real-time product performance KPIs and compare the product to similar products based on ingredients, functions of ingredients and product type to quickly identify opportunities for improvement. For example, if a new product is performing poorly in a particular testing scenario, the dashboard helps compare the results to similar products based on common ingredients (preservatives or active ingredients), packaging and product type.
The algorithm serves as the final and most essential component of the solution by enabling the company to forecast product stability based on ingredient concentration. By analyzing data from previous products, the algorithm can predict how a new one will perform, which goes a long way toward minimizing the risk of launching products that don’t meet quality standards and regulatory expectations. The algorithm we built uses a combination of supervised and unsupervised machine learning techniques, including regression analysis and dimension reduction algorithms and is constantly updated to incorporate new data and feedback from SMEs to allow for continuous improvement in forecast accuracy.
Conclusion
As industries like pharmaceuticals and cosmetics continue to evolve, digitalization and automation will become even more critical in ensuring that products are safe, effective and compliant with regulations — and often equally strict customer expectations. The example I discussed above is but one way in which a collaboration between SMEs, scientists, managers and an implementation partner can produce a customized and comprehensive solution that streamlines testing and evaluation processes and allows companies to develop safe, effective and innovative products faster and more efficiently.
About the Author
Gábor Berei is a Data Scientist at Starschema with several years of experience in the banking, energy, cosmetics and healthcare industries. His main focus is helping large corporations build machine learning-driven tools to enable advanced analytics solutions.
Connect with Gábor on LinkedIn.
REACH OUT TO STARSCHEMA HERE:
READ MORE STORIES FROM STARSCHEMA:
