‘Why the e.l.f. not?’ The beauty brand built an AI model to write social media comments

After nine months of training on social media data, the model, nicknamed “elf-luencer,” is now “talking like e.l.f.” around 90% of the time, according to the cosmetics company’s chief digital officer Ekta Chopra. It uses emojis and the brand’s “elf-isms” like “elf-mazing” and “elf-ing awesome.”

“AI is very powerful if it has human-led concepts to it,” Chopra said, speaking on a panel about customer loyalty in the AI era at Fortune’s Most Powerful Women conference on Monday.

The AI model has saved a lot of time for community managers who would otherwise have to think up and draft social media comments. Now those employees can have a broader reach and devote more time getting to know customers and building relationships with influencers. 

“But every comment is touched by a human,” Chopra noted. They’re using the tech to make employees more efficient, not replace staff, she said.

Chopra oversees the company’s sprawling digital footprint, from AI and augmented reality to social media and ecommerce. Customers can find e.l.f. products just about anywhere they can click—the beauty brand built up a presence on the gaming site Roblox this year, on top of its usual website and mobile app shopping experiences.

Chopra appeared this week alongside Amy Brooks, the NBA’s first-ever president of new business ventures, Irana Wasti, chief product officer at enterprise fintech firm BILL, and Astha Malik, the chief business officer for marketing software company Braze.

While impressive, Malik noted that generative AI like e.l.f.’s would be powerless without the creative direction of the humans who came up with “elf-isms” and all the social media content that was used to train the model.

“Creativity is still something that belongs in the human arena,” Malik said. AI isn’t wiping out the marketing team. Brooks and Wasti agreed it’s making teams more efficient by saving customers time on tedious tasks like calls to customer service and helping their companies scale global operations.

The NBA is currently using AI to translate game recaps into French, Portuguese, and Spanish. And a new AI tool, NBA Insights, will add context to stats for fans who are checking in about how their team did in last night’s game. 

Powered by Microsoft Azure, NBA Insights will identify key narratives and player performances from a particular game. But longform game recaps are easier to manage than live commentary, Brooks said. Generating translations for entire games is tricky given the quick pace and deep knowledge needed to follow it.

“Think of commentary in basketball,” Brooks said. “They’re so nuanced it’s just going to be screwed up.” 

The NBA is starting out by testing live translation on four games per week.

“There’s going to be mistakes, but we need to learn and iterate off of that,” she said.

Still, Brooks knows that what fans really want is to feel connected to players. And while Steph Curry might not be comfortable with an AI chatbot that makes Warriors fans feel like they’re having an intimate one-on-one conversation with him, Brooks said players might be more open to a model that is trained to give amateurs pointers on their shooting form.