One of the biggest ways that science impacts everyday life isn’t in glamorous ‘high tech’ applications, but in how it improves the products that fill our homes. This month’s blog looks at how ML-driven analytics are increasingly being applied in R&D for FMCG.
FMCG (Fast Moving Consumer Goods) is an ugly piece of industry jargon that disguises one of the most important industrial sectors – one that makes much of the stuff of everyday life. Food, drinks, toiletries, cleaning suppliers, cosmetics – these are products that sell quickly at relatively low margins and need to constantly adapt to shifting demands from customers, retailers, and regulators. It’s a sector packed full of busy scientists wrestling a constant set of R&D challenges in which time-to-market, cost, and sustainability are critical.
So it’s unsurprising that at Intellegens we’ve seen a surge in interest from the sector in the application of machine learning to their formulation, re-formulation, and packaging problems. We’ve previously shared interesting examples like Yili improving cream formulations, Plantsea developing sustainable alternatives to the plastics in laundry capsules, and IFF studying sensory properties. Other projects have touched on areas like disinfectants, flavourings, and plant-based foods.
I quizzed our customer service team on what users from the sector found most valuable in their use of the Alchemite™ machine learning software. They said these scientists highlighted the practical focus of the software, enabling them to generate machine learning models quickly, getting straight to useful results without too much distraction handling data science or statistics. And what were those useful results? Often, the most useful tools are the analytics that create plots exposing key relationships in data, and enabling these relationships to be explored. I asked our team to chat to a couple of these customers and ask them which analytics they found most useful. With thanks to those users for their input, this is what they highlighted…
The Influence plot is a quick, intuitive way to understand which inputs to a system drive the key target outputs. In the FMCG context, it can help you see, for example, which ingredient or processing parameters you should focus on as you modify a formulation – or, as was the case in the Yili example, whether there are additives that have lower than expected impact which could be removed. It sounds simple, but these can be difficult effects to spot in complex systems where multiple parameters may interact in a non-linear fashion.
Another powerful workflow is enabled by the Synergies analytic, which highlights pairs of ingredients that have a combined impact on product performance greater than the sum of their parts, highlighting key development avenues. These relationships can be studied further in the Interactions plot, providing a three-dimensional understanding of the subtleties in multi-component formulations.
At Intellegens, we’re always enhancing analytics in response to customer feedback. We’re also thinking about how to make it faster and easier to generate ML models that target the most relevant factors efficiently. For example, a recently released sustainability extension (demonstrated in last month’s webinar) eases the inclusion of carbon footprint in studies. Look out for a similar tool on cost modelling, coming soon. This focus on the FMCG hot-button issues of cost and sustainability is in part driven by our work with the sector.
FMCG may be inelegant terminology, but perhaps more than any other sector, it covers work that is relevant to us all. It’s good to see ML being used to plot the way to tastier, cleaner, safer, cheaper, better-smelling households.
