For a company whose mission is “keeping commerce human,” it’s a bit ironic how well Etsy has leveraged artificial intelligence and machine learning to drive better search results for its users.
Etsy operates a two-sided online marketplace that connects millions of creative buyers and sellers around the world. It has become the global destination for unique and creative goods made by independent sellers, connecting nearly 100 million buyers with over 7.5 million sellers and over 120 million unique products for sale (Source: Etsy). Etsy’s sellers generated over $13 billion of gross merchandise sales on its platform in 2021 (Source: Etsy). A simple snapshot of the homepage offers a glimpse into the variety of products that are offered on its site, ranging across custom clothing, jewelry, art, home goods, toys, craft supplies, and more.
The Challenge of Organizing the Long Tail of Creative Products
The characteristics of the Etsy marketplace create a uniquely difficult “search engine” challenge, more-so than other e-commerce marketplaces like Amazon. Etsy’s products, which are often hand-crafted by creative individual sellers, are often not the mass-produced goods that can be easily organized and tagged. In addition, the reasons a product may resonate with buyers often fall outside of traditional search terms and product listing information. While Etsy describes its sellers’ collections of millions of unique items as the “foundation of its competitive advantage,” it is equally critical that its buyers can find those items quickly and easily (Source: Etsy).
Ultimately, Etsy understood that delivering highly relevant search results to its users, taking into account their specific aesthetic preferences, would be core to its platform’s success. The company states in its 10-K:
“With millions of items listed on Etsy.com that don’t map to a catalog or a stock keeping unit (‘SKU’), our challenge is delivering world-class search and discovery technology that surfaces the right unique product to the right buyer at the right time in order to drive sales and buyer satisfaction.” (Source: Etsy).
This created a very complex technical challenge for the company – building a search engine that could deliver relevant results for its buyers from its 120 million products across a wide variety of queries, both highly specific and very broad.
Artificial Intelligence as a Technical Approach
The company turned to machine learning and artificial intelligence, teaching its platform concepts such as “style,” in order to better respond to searches. The team quantified 43 different styles of goods that had traction on the site, including “romantic,” “geometric,” “minimal,” “tropical,” and “midcentury” (Source: Fast Company).
Importantly, Etsy could not rely on a traditional, off-the-shelf AI algorithm, since seller’s often do not reliably convey a product’s style in their descriptions. In addition, standard approaches to image recognition could not accomplish this alone, since the concept of “style” often captures more than just colors, patterns, or other visual cues. Instead, Etsy built its own customized model, using a blended approach of both text-based analysis and image recognition (Source: Fast Company).
This approach stretched Etsy’s IT infrastructure. The company created a new department dedicated to Data Science and Machine Learning and hired over 50 engineers dedicated to tackling these problems (Source: Etsy Data Science & Machine Learning). Eventually, it also shifted all of its on-premise, localized datacenters onto the Google Cloud Platform within a single, cloud-based database to conduct its AI training and testing (Source: The Next Platform).
As a sign that its new AI model was working, the company found that purchase patterns of certain products aligned with the team’s expectations – for example, “tropical”-styled items saw spikes in sales in summer months, and “romantic”-styled items saw spikes in sales around Valentine’s Day.
These capabilities not only improved the core functionality of the Etsy platform, but they also opened up new opportunities for customization and targeting. For example, the platform can now deliver customized search results to different users who search for the same item, incorporating other information about the individual, including: (i) previous search history, (ii) demographics, (iii) time of day, (iv) device type, (v) the page on Etsy, and (vi) any other information available (Source: Springboard). This customization creates for more relevant results and helps convert more potential buyers.
The Impacts on the Business
The AI/ML approach has driven significant improvements in product discovery for the platform. Approximately 95% of search purchases are made with a result from the first page of search (up from 85% last year), and dead-ends are less than 1% of searches (Source: Etsy).
Better search results have led to better sales, which in turn attracts more unique sellers, feeding the company’s powerful flywheel. In turn, this increase in value creation has led to greater value capture – Etsy increased its marketplace transaction fee from 5% to 6.5% in April 2022, a reflection of the “added value for sellers.”
With nearly $13 billion of gross merchandise sales running through its platform and $2 billion of annual revenue delivered in 2021, the company’s current market capitalization sits at nearly $15 billion (it reached a high of $37 billion in December 2021) (Source: Google Finance). As the creator economy continues to expand, and more individuals seek to bring creativity to the e-commerce marketplace, Etsy offers these sellers a chance to reach an incredible number of engaged buyers at scale, with contextualized, relevant search results that sellers cannot find elsewhere.
Indeed, AI/ML have been core to the company’s noble mission of “keeping commerce human.”