Etsy ’s sellers aren’t the only ones crafting.
Up Close: In Conversation with Rithum's Ellie Williams
The company, which has been using machine learning ( ML ) for years, has developed a mission-oriented, responsible approach to using the technology—both to promote internal efficiency and to create effective consumer touchpoints.
Etsy’s journey with ML began in 2016, when the company started thinking about how the technology could help customers more easily search its broad, multifaceted catalogue.
Because the marketplace allows sellers to hawk everything from custom apparel and hand-painted footwear to wedding favors and homewares, it needed a way to segment its listings to make them discoverable and digestible for curious consumers, said Manju Rajashekkar, Etsy’s vice president of engineering.
“It’s really hard to pigeonhole the entire inventory into a specific catalogue,” he explained. “We started out with search, [then] we applied it to recommendations, which is the other side. We [refer to] search as something where you have an intent; recommendation is more of a discovery, where we try to understand what your preferences are and recommend things at different parts in the buying journey.”
Since the early days of ML adoption at Etsy, search and discovery have advanced much further, as has its penchant for using AI to serve its employees.
AI-ming at consumers
Etsy’s existing ML infrastructure—which includes a trove of data about the attributes of its inventory and about consumers’ preferences and behavioral tendencies—continues to help the company steer its strategy.
Earlier this year, the Brooklyn-based firm launched Gift Mode , a new feature that uses machine learning and AI to help consumers shop by persona based on the recipient’s interests.
A shopper fills out a few quick questions about who they’re shopping for, what that person’s interests include and the occasion for the gift, and Gift Mode presents them with a variety of lists to explore based on those attributes.
A shopper might indicate they need a gift for their father’s birthday, and note that he loves traveling and great food. In turn, Etsy will present the user with a few personas to choose from, like “The Adventurer,” “The Breakfast Lover” and “The Sandwich Specialist.” Each persona has a corresponding gift list.
Alternatively, a customer can select a persona directly from the Gift Mode page, skipping the input process. So, for instance, if a popular persona, like “The Cowgirl” or “The Cheese Lover” appeals to them, they can discover products that way.
The company already offers more than 200 personas, and it has plans to create more this year, in addition to listing more products on existing persona pages.
Etsy also uses generative AI to handle specific consumer queries. For example, a consumer could tell the conversational AI model, called Guided Search, “I am looking for a New York City-themed baseball hat.”
The model, available in Etsy’s app, will then share product listings relevant to the consumer’s query. It may also recommend other search terms for a consumer to explore. In this case, those ideas might include “retro NYC baseball hat with skyline design,” “vintage NYC baseball hat” or “embroidered NYC baseball cap.”
Rajashekkar said the tool can help the experience feel more natural and may put the customer at ease during their shopping session.
“A buyer has an intent, and they’re trying to find something, and the Guided Search assistant assists them every step of the way to help them find things that match their interest and match their particular purpose,” he said. He likened the experience to walking into a store and being helped by an associate.
Because Etsy’s business has two core customer groups—buyers and sellers—the company has to think about implementing technology that can benefit both.
Tech Tactics: How Carve Designs' AI Quiz Simplifies Swimsuit Sizing
On the seller side, the company already uses AI to flag low-quality image resolution, which often deters consumers from buying a product. However, it does not yet offer generative AI tools for creating product listing descriptions.
Rajashekkar said the company continues to evaluate how AI can benefit sellers’ storefronts and sales on Etsy, both through new tools and through education on existing tools like Gift Mode.
“We want sellers to have the agency over their shop, and we see our job as shepherds in helping them get there,” Rajashekkar explained.
Using AI as a ‘partner’ internally
Rajashekkar said that while Etsy has some fun, front-facing use cases for consumer experiences powered by AI, it has also helped make internal processes more efficient.
“We don’t want to use AI just for the sake of AI,” he said. “We use a lot of AI within Etsy for helping us be more productive—for developing software, for creating content in some areas, for marketing, for helping us come up with new ideas.”
Etsy has worked to democratize ML and AI for its employees, which it calls Admins. For instance, Rajashekkar said the company has equipped its programmers and developers with GitHub Copilot, which uses generative AI for code completion. While the developers still create Etsy’s features, he explained, Copilot helps them move through the journey faster, which, in turn, allows the organization to test software quicker.
“Really, the purpose of a co-pilot is not someone who’s taking over driving something completely autonomously. The co-pilot is someone who’s helping you assess at every step of the way, and that assistance happens at different places within our organization,” he told Sourcing Journal.
So far, bringing generative AI to all different teams within the organization has caused “the exploration piece of this [to go] through the roof.”
“Maybe in 2016 or 2017, in order to do machine learning, we really needed people who were experts in this domain, so they had to be trained in this domain; they had to spend many years in school in this domain,” Rajashekkar said. “But now, we have these intelligent systems accessible to us, which opens up this doorway for a lot of exploration to happen.”
Data from a December EY survey shows that 67 percent of employees in the U.S. have concerns about losing out on promotions if they don’t know how to use AI; similarly, 66 percent of workers said they fear falling behind if they decide not to use AI at work. Etsy’s approach of bringing AI directly to its teams—and encouraging trial and error—could help quell those fears among its workforce.
What’s coming next?
As Etsy’s employees continue to use AI, the company has plans for how it can use AI to enrich its mission. While it does so, Rajashekkar said it will continue upgrading its existing AI systems.
“We will really take the concept of iterating to heart. It’s not about the first product [we release]; we really want to learn from it,” he said. “Learning is a pretty big piece in how we build products.”
Already, it uses the technology for fraud and risk, all pragmatic use cases it plans to expand in the future. But it also has aspirations to continue developing use cases to evaluate quality of goods.
It won’t stop there, though. Rajshekkar said Etsy is currently running an experiment that would allow generative AI systems to describe an image. It has also started testing summarization of product reviews with AI so customers don’t have to parse through dozens of reviews to understand other shoppers’ general sentiment about a product.
“We’re trying to bring AI to every single piece in the products that we build,” he said. “Most of the places that we are leveraging AI are really on the backend side. We are integrating really seamlessly in the experience and making sure that it’s working [and that] it’s still true to our mission.”
From Waste to Wealth: (Re)vive's AI-Powered Revenue from Returns
Most Popular
Comments / 0