Consumers don’t often search for a “two-tiered bubble hem deconstructed leather jacket,” but they might be inspired and drawn to one if they see it in a store or online. Meanwhile, searching for “cool, minimalist sneakers” can perplex an AI, which is not designed to search by vibe.
Capturing that “gut feeling” of intuitive style has been a challenge for current agentic AI models, which lack a deep understanding of taste. Therefore, AI results tend to default to retail bestsellers or generic attributes that often miss the nuances of consumer preference.
In the presentation “From Search to Serendipity: How AI Can Understand Taste,” at the recent Coterie by Informa event in New York, Anabel Moldonado, founder and CEO of Psykhe AI, explained why standard metadata (e.g., “feminine,” “modern”) is too basic to drive meaningful differentiation or reasoning between products.
The crucial difference, she said, is the distinction between “intent-led purchases,” where users provide specific inputs into search bars, and “serendipity,” which involves surfacing items a user wasn’t actively looking for.
“Personality is an important layer, and there’s a strong connection between personality and aesthetic preference,” said Moldonado, who has a background in neuropsychology. “It’s something intrinsic in us, apart from everything we talk about, like celebrity influencers and trends.”
The Big Five
To integrate personality and taste into search preferences, Psykhe AI uses a Big Five model of personality to create a “taste layer.”
By mapping five specific personality traits to aesthetic preferences, the model can predict why a user might prefer a maximalist, busy interior style versus a minimalist one. The approach, which factors in a person’s openness, conscientiousness, extraversion, agreeableness and neuroticism, moves beyond simple demographic data to include stable, intrinsic drivers of consumer behavior.
“It’s not as simple as saying he’s an extrovert, he likes bright colors,” said Moldonado about Psykhe’s patented process, which is signed on by Kirna Zabete and Anne Klein, to name a few. “It’s a very advanced way of feature engineering.”
When Psykhe AI works with retailers, the company ingests the product catalog and uses computer vision and natural language processing (NLP) to look at an item’s text, description and image and assign it a score that corresponds to a person.
Cross-Category Generalization
Fashion doesn’t happen in a vacuum, and taste crosses over to other categories as well, helping companies get a broader picture. This is especially helpful for retailers selling across multiple areas.
By mapping both products and people into the same vector space, Psykhe AI’s technology can identify preferences from minimal user interactions and generalize insights across categories—for example, using a customer’s clothing preferences to predict their interest in home décor.
If a user shows an affinity for a specific trendy aesthetic in apparel, the model extrapolates that preference to other categories like home goods or toys. This creates a continuous, cross-category understanding of the customer’s digital persona—or digital style DNA.
“We place a lot of weight on the current session because we have the real-time per-user tracking,” said Moldonado. “Every time someone goes on a new page, we’re ranking from those signals. So there are longitudinal ones, the things that are fairly stable, but there are also ones that slant to entertaining.”
In an Anne Klein case study, the company saw a 22% increase in revenue per visitor (RPV) after working with Psykhe AI.
According to Gary Haas, vice president of digital & ecommerce at parent company WHP Global: “Fashion isn’t a filter problem, it’s a recognition problem. Customers are trying to find the version of the brand that feels like them. But within a single catalog, multiple aesthetic expressions coexist without being explicitly defined in words. Psykhe AI allowed us to resolve that in real time—adapting product discovery to each shopper’s style DNA.”
