It’s Hard To Rave About A Plain White T

S

ay you’re looking to update your work-from-home wardrobe. You go online and start browsing everyday basics. You’re confronted with an abundance of choices: striped tees, scoop-neck tees, pocket tees, graphic tees, V-neck tees and button-back tees. And that's just the tees.

In the past, a beleaguered copywriter had to muster the energy to spin up enthusiastic descriptions for each and every one, while taking care to accurately describe the materials and fit. It might take them weeks. Now some retailers are testing whether they can have generative artificial intelligence, using the same underlying technology as ChatGPT, do it instead.

Last year, Stitch Fix began using AI to produce product descriptions for its online catalogue of thousands of items. "We can write 10,000 product descriptions in 30 minutes," said Sachin Dhawan, Stitch Fix's chief technology officer. It typically takes less than a minute for an employee to review the copy, which is approved nine times out of ten. "They are super good," he added.

Retailers are looking at the technology as a way to cut down on the time and expense involved in running an e-commerce operation, especially as profits have been pressured by higher costs and higher levels of discounting. Copywriters make an average of $29 an hour, or about $60,000 annually, according to ZipRecruiter. AI can do the job less expensively in a fraction of the time. In a Coresight Research survey, more than a third of retailers said AI helps to create product descriptions that represent their brand's voice.

At Adore Me, a lingerie brand acquired by Victoria's Secret, copywriters used to complain that writing product descriptions was one of their least favorite tasks, taking them 20 or 30 hours a month. "When you have to try to be original, fun and creative 100 times in a row, it's not the most fun thing," said Ranjan Roy, Adore Me's vice president of strategy.

Now they can use AI to jumpstart the process. "It's better than staring at a blank screen," said Brian Hennessey, CEO of Talkoot, which uses AI to help online retailers fast-track copy. While a company may end up rewriting 80% of the description for its most visible and important products, he said, AI can at least help get them started. They rely more heavily on AI for their volume drivers, where they may only change 20% of the copy.

"A sock is a sock. A white T-shirt is a white T-shirt," said Hennessey, who was previously the global writing director for Adidas. "I know what they are. AI knows what they are."

Hennessey's customers, who include Adidas, Under Armour, Reebok and Burton, have saved 40% on content production, plus reduced errors by three times and increased conversion rates by 10%.

B

urton, the snowboard company, is using AI to personalize its shopping experience. It knows that its website tends to attract more hardcore snowboarders, while its Amazon page gets parents and newbies. So the company is using AI to write separate product descriptions, publishing more insider language on its website and more descriptive, educational language on Amazon.

AI is also good at optimizing language for search. Adore Me has seen a 40% increase in aggregate non-branded search, meaning that it's showing up higher in the search results when someone who is perhaps unfamiliar with the brand goes online and does a general search for bras, underwear or swimsuits.

Honing an algorithm's salesmanship takes time. When Adore Me first began dabbling with the technology, the descriptions were generic, off-brand and used terms that humans would probably never use, like "whilst." It then began training the algorithm on its own data, first feeding in all 23,000 product descriptions it had written since the company was founded a decade earlier.

"You would think bigger is better," said Adore Me's Roy, but it included all sorts of inconsistencies from different writers with different styles from different time periods.

Ultimately, the brand fed it copy from the past two years, which had a much more consistent brand voice, and tweaked from there. It instructed the algorithm not to include certain words, like "green" and "eco-conscious," for compliance purposes. It made sure that each description was optimized for search, like always including references to the matching bra or panty.

"It's nutso," said May Habib, CEO of Writer, an AI company that worked with Adore Me on product descriptions. "They’re almost able to eliminate all human effort in the writing process completely."

Adore Me said it hasn't laid off any copywriters, but that the job description has changed to focus on more high-level marketing strategy and creative campaign work.

Stitch Fix also said its copywriting team remains intact. It keeps humans "in the loop" to lend their judgment, even as it has spent years training its algorithms to take over more functions. Its stylists, for instance, review computer-generated clothing recommendations before sending them to shoppers who have signed up for a subscription box. But its a-la-carte offering, called Freestyle, independently generates recommendations in near real-time. It all helps to lighten the workload. While it's difficult to say how much algorithms are to blame, consider that the number of stylists that Stitch Fix employs has decreased by a third since 2019, while the number of data scientists and engineers has risen by a third, according to the company's annual reports.

More companies are moving in this direction. Big retailers and food and beverage companies are showing the most interest in AI-generated copy, according to Hennessey. With the recent buzz around ChatGPT, he said he's encountering high expectations, with some companies ready to let the technology take over completely.

"It's like coming up to a farmer who's been plowing behind a mule for 40 years. You tell him you made a tractor, you just have to get on it and you can sow your fields in record time. And having the farmer go, ‘But I still have to drive it?’" said Hennessey. "We’re trying to combat the idea that AI can do absolutely everything."