L’Oréal, Mondelez and Nestlé are using AI to reduce the number of physical formulations they must test before launching a product. Models predict ingredient behaviour, generate recipe options and search material combinations, while scientists and product teams decide what reaches the laboratory and consumer.
The benefit is not replacing experimentation. It is focusing expensive physical work on candidates with stronger expected performance across taste, safety, cost, sustainability and supply.
Beauty models search for new uses in known chemistry
L’Oréal has used AI in its laboratories for four years to predict how molecules affect skin and hair. The company says formulation has become four times faster.
One example repurposed molecules previously used in skincare for a collagen-based shampoo intended to add lift. Digital simulation allowed teams to explore combinations before committing to laboratory batches.
Known ingredients can reduce some uncertainty, but use in a different product changes concentration, interaction and exposure. Safety and performance still require appropriate physical testing.
Faster iteration matters as beauty trends change and L’Oréal seeks stronger growth. Speed should not incentivise a flood of minor products whose environmental and marketing costs outweigh consumer value.
Food companies reduce samples and manage ingredient volatility
Mondelez uses AI to generate and evaluate recipes across brands such as Oreo, Cadbury and Toblerone. Human experts review unusual combinations before selected formulas proceed.
The company says the tool supported Gluten Free Golden Oreo and a revised Chips Ahoy recipe, with 60% of AI-assisted biscuit recipes performing better across nutrition, sustainability and cost.
Recipe optimisation can also identify alternatives when an ingredient becomes expensive or unavailable, reducing dependence on one source. Substitution must preserve allergen, labelling, taste and manufacturing requirements.
Nestlé’s plan to remove artificial colours worldwide requires screening natural alternatives, production testing and shelf-life evaluation. AI can narrow the space, but consumer and stability studies remain essential.
Synthetic taste predictions should be checked across cultures and age groups. Historical sales data can reinforce established preferences and miss genuinely new products.
Nutrition and sustainability are multi-objective decisions, not decorative scores. A cheaper ingredient may increase processing or reduce supply resilience, while a lower-carbon option may perform poorly during storage. Product teams should show how trade-offs were weighted and prevent the model from hiding a weaker dimension behind one combined ranking.
Product intelligence extends into packaging and supply chains
Nestlé and IBM developed a generative tool for high-barrier packaging materials, connecting molecular structure with moisture, oxygen, temperature, cost and recyclability properties.
That broader optimisation reveals why product development is a systems problem. A formula succeeds only when ingredients can be sourced, factories can produce it and packaging protects it through distribution.
Barry Callebaut and NotCo are exploring plant-based chocolate ingredients, while Haleon’s Microsoft collaboration reaches research, forecasting and supply operations. The pattern is industry-wide rather than one isolated laboratory project.
Companies need experimental provenance: model version, objective weights, rejected formulas and physical results. Failed tests are valuable data and should remain available rather than disappear with a project team.
Metrics should include development time, samples avoided, consumer acceptance, nutrition, footprint and launch performance. Cost reduction alone can drive recipes toward technically acceptable but less enjoyable outcomes.
AI is giving product scientists a faster way to ask “what if?” across enormous combination spaces. The best organisations will use that freedom to explore more intelligently, then apply human taste and scientific discipline. A model can propose a shampoo molecule, biscuit recipe or packaging polymer; the product becomes real only after experts prove that people can safely use it—and actually want to.
Manufacturing teams and consumers then supply the evidence showing whether the proposed answer was useful.
