Shopping on online marketplaces such as eBay and Etsy, or even secondhand clothing seller ThredUp, can feel like treasure hunting at a giant flea market, except even more daunting as there are millions of unique listings to sort through.
So it is no wonder that online marketplaces are eager to add artificial-intelligence functionality to their search bars. The technology has the potential to behave like a ChatGPT version of a personal shopper, acting on sentence commands to come up with relevant results and digging up products the consumer might not even have known existed on the platform.
Getting there could take a bit of time, though.
Gift mode
Earlier this year, Etsy launched Gift Mode, an AI-powered feature that lets shoppers click a few details about the person they are shopping for. The marketplace could certainly do with some help sifting: Etsy has more than 100 million listings and the average search result yields upward of 10,000 items.
In April, eBay—which had about 2 billion live listings as of year-end 2023—introduced a generative AI-powered feature called “shop the look” that gives users a selection of curated outfits based on their shopping history.
ThredUp also launched a new AI search product earlier this year. “In secondhand, where you have 4 million items and you’re adding 100,000 a day, one of the biggest things we hear from customers is ‘oh, I can’t find something I want,’” ThredUp Chief Executive James Reinhart said at an industry conference earlier this year.
Tagging attributes
Reinhart said ThredUp used to tag four to six attributes—such as color and pattern—that a customer might search by, using human intuition. With AI, it can tag hundreds of attributes and then cluster those attributes “in ways that it can intuit what the user is actually looking for.” Enhanced search is already leading to more searches on ThredUp, according to the company.
A fully conversational online shopping experience might take some time to materialize, and it isn’t yet clear whether the switch will be worth it. For one, online shoppers are impatient, and large language models still take a relatively long time to spit out responses. Google previously found that, as page-load time goes from one second to five seconds, the probability of a user bouncing away increases 90%.
Etsy CEO Josh Silverman previously said that while Etsy measures latency in search by the millisecond, latency in GenAI is measured in seconds, adding that such a length of time is “more than most consumers will bear.”
Cost issues
Cost is another issue. Morgan Stanley analysts estimated that large-language query costs are about five times more expensive than traditional models.
For Etsy to maintain the same earnings before interest, taxes, depreciation and amortization (Ebitda) at the higher search cost, Morgan Stanley estimated that the buyers’ basket size would have to increase by 13% on average every year through 2025. That would be a tall order for a company that has seen that metric decline over the past year.
In the near term, a more promising use case for AI might be reducing friction for sellers, which itself could encourage more listings and sales.
For example, eBay last year launched a feature called magical listing that allows sellers to instantly populate detailed product information through generative AI based on the product’s title, category and other items.
Jamie Iannone, CEO of eBay, said at an industry conference in March that more than 90% of the sellers accept the automatically generated content, some with edits. The company is working on a version that can spit out a product description based on the image of the product.
Like AI search results, further breakthroughs in AI can’t come fast enough for online marketplaces.
Write to Jinjoo Lee at jinjoo.lee@wsj.com