By Nahla Bekkara, Partnerships Lead and Enterprise Growth, Haut.AI
Keywords: Artificial intelligence (AI), Beauty R&D, Cosmetic formulation, Machine learning, Formulation development
Artificial intelligence (AI) is already changing how consumers discover beauty products, receive recommendations and shop. Industry research closely tracks these consumer-facing applications, from discovery and search to personalization and commerce. Inside research and development (R&D), the picture is different. Peer-reviewed studies demonstrate that machine learning can help predict ingredient safety and formulation behavior, but the field remains at an early stage of adoption. The distance between published scientific progress and broader industry use is an important development for beauty professionals to understand.
AI adoption is concentrated in consumer-facing functions
AI adoption across industries continues to grow. According to the Stanford Institute for Human-Centered Artificial Intelligence’s AI Index Report 2026, 88% of organizations reported using AI in at least one business function, up from 78% the previous year.
In consumer goods and retail, marketing and sales recorded the highest AI usage at 51%. Product development did not appear among the leading AI functions in any sector covered by the survey, underscoring how industry-wide adoption data remain concentrated in functions closer to the customer.
Beauty industry research reflects a similar emphasis on consumer-facing applications. McKinsey’s State of Beauty 2026 discusses AI in the context of commerce, discovery, personalization and search. It projects that AI platforms could drive up to 35% of e-commerce transactions within three-to-five years.
Although these findings do not mean AI is absent from product development, they highlight the gap between the functions emphasized in industry surveys and the applications explored in formulation research.
Published research demonstrates practical applications
Research into machine learning for cosmetic formulation has examined specific technical challenges, including predicting ingredient safety and assessing formulation performance.
A 2025 study reported 86% accuracy in predicting skin sensitization potential using 24 physicochemical descriptors. The model’s results included 80% sensitivity and 90% specificity. Such predictions may help researchers assess ingredient safety earlier, though further evaluation is needed before applying findings to specific products or regulatory decisions.
The same paper evaluated five machine learning models for predicting cleansing formulation performance. The strongest reported result had an R-squared value of 0.770, meaning the model accounted for a substantial proportion of the variation observed in the study.
A separate 2025 study in Cosmetics examined the phase stability of 812 shampoo formulations, 294 of which were stable. The researchers found that a large language model could distinguish stable and unstable formulations with useful accuracy using as few as 20 training examples. Its performance was comparable to conventional machine learning approaches that required two-to-five times more data to reach a similar level.
These are measurable research findings tied to defined technical challenges. They offer evidence of what AI can do under specific study conditions, not proof that the same results will hold across ingredients, product categories or commercial development environments.
Research progress and industry adoption remain distinct
A 2026 review in Chemical Papers examined machine learning applications in areas including raw material screening, ingredient matching, quality control and personalization. The authors concluded that machine learning in cosmetic formulation remains under development rather than at a stage of mature deployment. They identified data reliability, model interpretability and regulatory acceptance among the barriers to wider use.
For formulators, these considerations are central to evaluating an AI application. A model’s performance must be understood in the context of its training data, the specific question it addresses and the conditions under which it has been tested. Predictions can inform research priorities and help assess potential outcomes, but they do not remove the need for experimental validation, scientific expertise or appropriate safety and regulatory review.
The evidence points to a field with demonstrated technical promise and work still to do before wider adoption. The opportunity is not to treat AI as a substitute for established science, but to identify where its measurable capabilities can contribute to product development and how those capabilities can be validated in practice.
Explore AI applications at BIFA Symposium
For cosmetic R&D teams, the opportunity lies in identifying where AI can provide useful evidence alongside established scientific methods, while addressing the questions of data quality, reliability and validation that stand between research and practical application.
These questions will be explored at the Beauty Ingredients & Formulation Asia (BIFA) Symposium, part of Cosmoprof Asia 2026, during “AI in Beauty R&D: From the Shop Window to the Laboratory.”
Moderated by Nahla Bekkara, Partnerships Lead and Enterprise Growth at Haut.AI, the session will feature Dr. Eisuke Takai of Shiseido Co., Ltd., Japan, and Dahlia Hamzah of Givaudan Active Beauty. Taking place Wednesday, November 11 from 12:05 to 12:40pm at AsiaWorld-Expo in Hong Kong, the panel will examine current AI applications and emerging approaches to beauty research and product development, bringing scientific findings and industry perspectives into the same discussion.
