How Estée Lauder and Google taught AI to sell fragrances

Unlike the current barrage of agentic stylists and shoppers tasked with picking outfits and hunting down deals, the “AI Scent Advisor" must spin everyday language into the intangible and totally nondigital experience of smelling.

The result is a chatbot with a fluttery, poetic tone.

“Where would we like to journey today?" It asked during one session. “The freshness of an orchard? The warmth of a blooming flower garden? The windswept allure of the coastline?"

The goal is to convince online shoppers to spend north of $100 on a scent suggested by a chatbot conversation filled with ethereal adjectives and descriptions of the English countryside.

“If a year ago I would have said to people, ‘Can we use AI to help you with scents?’ they’d be like, ‘Well, you can’t smell your iPad’," said Brian Franz, chief technology, data and analytics officer at Estée Lauder, which includes brands like MAC and Bobbi Brown.

To make it happen, the firm turned to Google Cloud, which sent a handful of AI engineers armed with Gemini models to spend a day sniffing floral musk, and then managed to crack the code.

Since a soft launch begun on Jo Malone in October, online shoppers who used the tool made purchases at almost double the rate of those who didn’t, helping Estée Lauder achieve that rare unicorn in generative AI: a tool that actually drives top-line growth.

Estée Lauder’s share price plunged in the past few years, hurt in part by a pullback in spending by Chinese consumers, though it’s been up more than 25% since CEO Stéphane de La Faverie took the helm in January. The company was also late to embrace online retail, a mistake de La Faverie is hoping to correct with a new digital push. In April, he appointed Franz to the company’s first-ever post of chief technology, data and analytics officer. Franz came from State Street where he most recently served as executive vice president, global chief information officer and head of enterprise resiliency.

In a recent filing with the Securities and Exchange Commission, the company said restructuring efforts, which include consolidating service providers and pursuing a net reduction in workforce, have cost $1.14 billion.

De La Faverie has said he’s betting heavily on the fragrance category, which has become increasingly popular among younger consumers, and the digital opportunities there. In October, Estée Lauder opened a Fragrance Atelier in Paris, with the goal of using advanced technology to help develop new scents.

So how did Estée Lauder train its chatbot to talk scent?

It started with an existing trove of scent data. In perfumery today, a given scent can be based on any combination of seven primary olfactory categories (among them woody, amber and “chypre"), and Estée Lauder also uses five newer categories (green, aromatic, musky, leathery and gourmand) that correspond to real ingredients in the recipe. Then there’s the less technical, but equally relevant sensory experience each perfume evokes: a pear orchard in the fall or a bucket of blackberries on a hot summer day.

Google engineers fed the model with both, but then had to tackle the issue of capturing the right tone.

Google’s Gemini models power a range of customer-facing experiences from Home Depot to Wendy’s. But choosing a signature scent is more personal than ordering a Baconator. It was important that the Jo Malone bot have a voice that evoked real, in-store associates, the beauty company said.

To get there, Google Cloud recorded and analyzed the way live scent stylists speak to customers, including a methodology known as TED, or “tell, explain, describe."

“When you can’t smell something, you have to be really evocative with language," said Jose Gomes, vice president, retail and consumer goods at Google Cloud.

That’s how the AI Advisor became so ethereal and philosophical, he added. The real associates love phrases like “journey" and “captivate."

Estée Lauder said it’s still considering bringing a similar AI experience to its other fragrance brands, which include Kilian Paris, Frédéric Malle, Tom Ford Beauty and Le Labo.

“If you put the consumer at the heart of this and you bring the technology, we’re really able to make a big difference in how the ELC companies perform," Franz said.

Write to Isabelle Bousquette at isabelle.bousquette@wsj.com