Approximately 3.7 billion people have a tube of conditioner on their bathroom shelf. The majority of those individuals have never once paused to consider how it came to be—what choices, experiments, and setbacks preceded the formula they are holding on a Tuesday morning. That unglamorous, mostly undetectable process is being subtly recreated. Not by a fresh batch of chemists. through artificial intelligence.
For the past few years, Unilever, the Anglo-Dutch consumer conglomerate behind Dove, Knorr, Hellmann’s, and numerous other well-known brands, has been developing what it refers to as a “always-on” AI insights system. The numbers are compelling enough to make you stop and think. Consumer insights are now evaluated 60% quicker. The process of demonstrating that a product does what it claims to do on the packaging, known as claims generation, is 75% faster. Formulation cycles that used to take five or six rounds have been reduced to just one or two. It’s possible that the majority of customers won’t notice this change. However, it is present, ingrained in the items they instinctively grab for.
The size of the underlying machinery is what distinguishes this from a standard corporate technology announcement. Just the Beauty and Wellbeing division of Unilever, which includes skin care, hair care, and the expanding vitamin and supplement market, is worth €12.8 billion. A company of that size’s changes in product development have an impact on more than just its own output. It modifies what is produced, what ends up on shelves, and subtly, what the rest of the industry is under pressure to do next.
The AI tools in use are acting in a truly unique way. Instead of just crunching sales data, they pull information from more than 1,000 external sources each month, such as social media sentiment, search behavior, retail movement, and competitor activity, and compare it to Unilever’s internal research archive, which contains 150,000 scientific documents accumulated over more than a century. In real time, scientists can query this archive in their native tongues, just like they would a colleague who has read it all. It’s the kind of access that used to take weeks of work and a research team.
Virtual cohorts, which are essentially digital test groups created from Unilever’s microbiome datasets, are another option. These AI-generated sample populations can mimic how actual people might react to a new formula because they represent a variety of skin and hair types, ages, and geographical locations. It is possible to analyze about 2,500 subjects at once. It sounds, and is, clinical. However, it also takes the place of some of the costly and time-consuming human testing that was previously required for every significant product launch.
The same reasoning is being applied in a different context on the food side. AI search tools and voice assistants are increasingly finding Unilever’s brands instead of in grocery store aisles. The team identified a structural issue as the reason behind Hellmann’s low ranking for “Game Day sandwich recipes” prior to the Super Bowl: there wasn’t enough listicle-style content for algorithms to comprehend. They improved the keywords, reorganized the content, and saw a ten-position increase in visibility rankings. It sounds unremarkable. Most likely, it is. However, it also indicates a more significant change.
Observing all of this suggests that the consumer goods sector is about to enter a stage in which the gap between a cultural moment and a final product will continue to close. Theoretically, a reformulation brief by Thursday could be informed by trends observed on social media on Monday. That’s not totally acceptable because there are legitimate concerns about whether speed always equates to quality and whether optimizing for algorithmic visibility alters the true purpose of food and cosmetic products.
However, it’s difficult to ignore the fact that the slower, more costly, and more unpredictable alternative wasn’t particularly beneficial to customers either. AI’s place in the innovation process is no longer the question. It obviously does. What is lost when a machine outpaces human thought is a more difficult question.
