Despite existing purely to help shoppers find what they’re looking for, the search bar on e-commerce sites isn’t known for yielding great results.
A recent search for “black straight leg jeans” at one well-known retailer returned a pair of blue jeans among the first three products. On another site with a luxury focus, baggy wide-leg jeans appeared among the top three.
“The search world was really optimised for ultra-large corpora [i.e. bodies of information],” said Malte Ubl, who previously led desktop search at Google and is now chief technology officer at Vercel, which powers e-commerce storefronts. “That’s what those algorithms work at.”
Finding that one right product in a retailer’s inventory can require a different solution. To address the challenge, companies are upgrading site search with AI ranging from computer vision to large language models.
Retailers such as Revolve and Vestiaire Collective have taken matters into their own hands, internally enhancing their search functionalities with new AI capabilities. But there’s also a rise in tech providers such as Constructor, Lily AI, Vantage Discovery and Nosto promising to improve site search with AI and drawing customers like Tapestry and Sephora.
These search-as-a-service start-ups are gaining traction, said PitchBook e-commerce analyst Eric Bellomo, in part because retailers aren’t meeting consumers’ expectations. When Deloitte Digital surveyed consumers and companies last year, 79 percent of brands described search and discovery on their websites as good or excellent. Only 63 percent of consumers said the same.
“There’s a gap between consumer and brand perceptions of the experience, meaning there’s some level of investment necessary to bridge that gap,” Bellomo said.
The opportunity is great enough that search-as-a-service start-ups have attracted $1.7 billion in venture-capital funding since 2019, according to PitchBook, with search edging out other “pre-purchase” technologies like personalisation and price optimisation in deal count last year.
While every company may have its own approach, the goal is the same: to make the search bar where shoppers go to find — and buy — what they want.
Though how important search is to a fashion retailer can vary. When the Baymard Institute, which studies online user experience, recently tested how shoppers locate products on different apparel sites, just 10 percent used the search function, and only resorted to it after failing to find what they wanted with the site navigation. Baymard surmised that the smaller product catalogue and relatively navigable category taxonomies made search unnecessary.
Vestiaire, on the other hand, says 40 percent of its orders derive from text searches, while J.Crew’s chief information officer, Danielle Schmelkin, recently told The Business of Fashion for a case study on creating the perfect e-commerce site that a significant number of its shoppers rely on search, particularly those intending to make a purchase.
How Site Search Is Getting an Upgrade
To help improve the experience of those shoppers, J.Crew tapped Lily AI, which as co-founder and chief executive Purva Gupta put it, lets retailers “speak the language of their customers.” By that she means a retail merchant might describe a colour as “oxblood,” but a shopper would just call it “red.” Search relies on the information attached to an item, including any tags and labels, to find things. If the shopper and site aren’t using the same terms, it can make search ineffective.
Using AI including machine learning, computer vision and large language models, the company extracts product attributes from item imagery and enriches the product data with the information to make the item more discoverable. It can help not only with on-site search but also making it more visible on Google, according to Gupta, who said they generally see a lift of 5 percent to 25 percent in clicks, impressions and conversions on products.
The same approach can also let retailers surface products if a shopper looks for a style or trend, like “quiet luxury” or whatever niche aesthetic is bubbling up on TikTok. It just requires creating the labelling data to ensure a product appears in that search.
“We are listening to trends all over and then we are sending those to our brands and retailers,” Gupta said. “They can just enable whatever they would like on their sites.”
The advent of generative AI is paving the way for other advances, too. The large language models powering tools like ChatGPT are probabilistic machines good at predicting a user’s intent and factoring in context. That makes them ideal to apply to product search, according to Nigel Daley, co-founder and chief operating officer of Vantage Discovery, a generative-AI search platform. The shopper is trying to describe what they’re looking for and the retailer has to predict the best matches among a large number of products that may be labelled in different ways.
“This is the perfect use case for AI to clean up and understand your product catalogue based on the visuals that you provide and some messy amount of structured and unstructured data,” Daley said. (Though he doesn’t believe AI chatbots are the ideal interface.)
Daley and his co-founder, Lance Riedel, came from Pinterest, where Riedel was instrumental in helping to launch shopping, which has become an area of success for the company. The challenge on Pinterest was that much of the content, rather than being just text, was visual and style-oriented, and there’s so much of it that the platform needs to learn what individual users want to see.
Vantage Discovery enables visual searches and personalised search as well, meaning the results users get are tailored to their preferences based on factors like their browsing activity on the retailer’s site and past purchases. You can “thumbs up” or “thumbs down” results to get matches more aligned with your tastes in the future. That way, if a customer is always buying activewear, their search for “blue shirt” could turn up casual or gym T-shirts rather than a blue button-up for the office.
Daley said the company is currently working with some large retailers but wasn’t able to disclose which ones. He also said some luxury brands are testing the technology as well, since the style component is so important to that shopper. The company introduced an app for Salesforce’s Commerce Cloud last month, and according to Daley, will be offering integrations on more large e-commerce platforms soon.
Do It Yourself
While AI’s powers are helping to drive the growth in search-as-a-service, PitchBook’s Bellomo said these start-ups do have some competition.
“The availability of open-source [AI] models that can be fine tuned presents a competitive alternative,” he said.
In theory, brands can integrate large language models into their search on their own, but it isn’t easy to pull off. Many retailers use off-the-shelf e-commerce systems with standard search bundled in. Overhauling search on your own requires resources and technical expertise. That’s not something every fashion company has, though technology partners like Vercel can help.
Still, some companies are doing upgrades themselves. This week, for instance, Vestiaire Collective unveiled a new search engine that, in its words, “translates keyword searches into image pattern recognition,” allowing it to visually spot matching results as well. Vestiaire said in a release that the new search is delivering “significantly more accurate and relevant items,” and leading to more sales of products featured in the “similar items” widget on its site.
Revolve, having used a third-party search platform for years, recently introduced on its FWRD site a new AI search algorithm it developed internally. On the company’s August earnings call, co-founder and co-CEO Michael Karanikolas said it was “more about understanding general aesthetics and broader concepts that traditional search struggles with.”
The new search was outperforming the old third-party technology, he added, driving higher revenue per search at a lower operating cost. The results were good enough that they started testing it on the main Revolve site as well.
For now, if all goes well, consumers may hardly notice the difference. The search bar is one of those features that stands out most when it’s not working.
In the meantime, companies are hard at work trying to make it more than a last resort, and ideally maybe even the preferred way for shoppers to find the items that best match what they have in mind.
“I think the consumer is going to have a whole different shopping experience in the near future,” Lily’s Gupta said.
