This article on generative AI in search is part of our Vogue Business membership package. To enjoy unlimited access to our weekly Technology Edit, which contains Member-only reporting and analysis and our NFT Tracker, sign up for membership here.
The internet has successfully provided shoppers with everything, everywhere, all at once. Now, the problem lies in narrowing that down.
Tech giants are turning to generative AI to change e-commerce search and discovery tools. The hope is that people will eventually be able to talk to computers the way they talk to sales associates instead of navigating via the restrictive phrases we’ve all been trained to use when conducting an internet search query.
For example, instead of “orange puffer coat women’s”, the future of search in the AI era might sound something like “I’m looking for a bright ski jacket for my upcoming trip to Whistler.” In addition to improvements to static, one-way search queries, in which a shopper plugs in words and the e-commerce site spits out (hopefully) matching products, brands are starting to experiment with ChatGPT-style, two-way dialogues whereby the shopper can iterate on previous comments. For example, they might respond to results for the Whistler search with, “No, I’m looking for women’s coats, and I don’t like yellow.”
“We have lived in a world of keyword search, but a profoundly different experience is possible because of generative AI,” says Raj De Datta, co-founder and CEO of e-commerce personalisation company Bloomreach, which acquired AI-powered shopping concierge Radiance Commerce last month. “The quality of what comes back is dramatically different.” Critics point out, however, that generative AI requires large data sets and cannot (yet) match human intuition or problem-solving skills. Customer distrust remains high even as adoption rises.
The retailers that are likely best positioned to benefit immediately are those that have larger product catalogues.
This month, Amazon began rolling out a generative AI-powered shopping assistant called “Rufus”, starting with a test for US customers on its app and accessible through the search bar. In a blog post announcing the technology, Amazon VP of search and conversational shopping Rajiv Mehta, said that “generative AI is going to change virtually all customer experiences that we know.” Google now enables retailers, such as Victoria’s Secret, to build and deploy conversational chatbots on their own sites and apps. Salesforce’s new Einstein Copilot for Shoppers is an AI assistant that retailers can plug into their digital storefront or messaging app. It is informed by existing customer data, such as location, preferences and past purchases.
Rent the Runway is now testing an AI stylist for all customers, and Kering has been testing “Claire”, an assistant that returns product suggestions from Kering’s suite of brands. “Fashion as an industry serves to benefit from AI to narrow the endless aisle problem of e-commerce,” said Rent the Runway CEO Jennifer Hyman in a recent earnings call announcing the pilot.
Half of retail executives are prioritising AI-driven personalised product recommendations in 2024, according to Deloitte research. The thinking is that this will solve a lot of problems, enabling shoppers to search in increasingly natural ways and enabling retailers to better match their queries with a relevant available product. Approximately 75 per cent of consumers report having issues with search in the last six months, says Carrie Tharp, VP of industries at Google Cloud; for many, it affected their willingness to shop on certain sites or led to cart abandonment, she says. Search abandonment cost US retailers more than $300 billion in 2021, according to research commissioned by Google.
This is not the first time that online retailers have attempted to improve the search-and-discovery process. Already, they have used machine learning to improve search results and recommendations while deploying chatbots to enable back-and-forth dialogues. The difference is that the generative AI tools are more “intuitive” and don’t take as much training to provide better results. While chat-based generative AI tools aren’t the same as talking to a trained associate, they are seen as far more flexible and less potentially frustrating than a chatbot whose outcomes are limited to what has been narrowly pre-programmed. It’s also the latest development in 2024’s “practical tech” trend.
“This is not like Web3 where it is a conceptual shiny object,” Tharp says. “Search is a current problem and friction for all consumers and websites. The old search box is very constraining, and it became the standard for how you interact.” Tharp, like many in the tech world, believes that improvements to AI will “fundamentally change the future of search and discovery on websites”.
Conversations replace keywords
Generative AI is now powering search results that are trained on large language models (LLM) to better interpret what people mean. Most tools can understand both images and text, and the recommendations aren’t only based on manual keywords (“taxonomy”) that a retailer has determined or pieces that a retailer is trying to prioritise, but rather the best results based on real-time data. Most retailers are likely to begin by using this type of generative AI to improve discovery because it upgrades an existing customer experience: type in a query to get results.
In addition to more intuitive search results, the next step after search is a new generation of chatbots that don’t require retailers to programme all the possible responses to questions. In the past, retailers used something called a “decision tree”, and the answers were all planned and programmed, something called a “deterministic flow”. Going forward, brands are looking at using generative AI to generate more open-ended responses that are still able to incorporate brand tone and be limited to a brand’s products. This is especially useful for customers who are looking for something for a specific need or occasion but don’t know what they want. “If I am clear about what I am looking for, the internet is a great place to shop, but people find just about 20 per cent of products online. For the other 80 per cent, why do they go into the store? A big reason is they want to talk to somebody,” De Datta says.
Generative AI, which can create content in the form of images, videos and text, is a new field for fashion and retail that promises to aid creativity and business processes — but also introduces new considerations.
While manual taxonomy is still useful because of LLM and multimodal AI (meaning AI that recognises both words and images), the degree of dependency on the retailer or brand to describe their products is less than ever with generative AI tools, he adds.
Amazon’s shoppers can ask assistant Rufus questions such as, “Are these true to size?” and follow-up questions such as, “Are these comfortable?” and Rufus will generate answers based on listing details, customer reviews and community Q&As, Mehta says. Rufus is designed to answer questions related to a range of needs, including occasions, purpose, broad category (“clean beauty”), product comparisons (“lip oil versus lip gloss”), recommendations and product details (such as “is it machine washable?”).
Rent the Runway customers can search common fashion terms or use cases, such as “Miami vibe” or “Clambake in Nantucket”, to find recommendations. In the earnings call sharing the news, Hyman said that Rent the Runway is “uniquely positioned to be a significant beneficiary of AI” because customers frequently interact with the company and because of its extensive data on customer behaviour and product insights, and that the company would continue to iterate on this technology. The majority of Rent the Runway subscribers review 10 or more items each month, Hyman said.
Macy’s has also already witnessed improved results using AI-informed search; since it has already used Google search tech, the retailer was able to “literally check a box” to turn on its generative capabilities, Tharp says. In addition to a reduction in “null results” (from, for example, searches with a very long description), Macy’s saw a 2 per cent increase in conversion, 1.3 per cent higher revenue per visit and significant improvement in relevance. “The larger the catalogue and more categories, the more business value that relevance creates,” Tharp says.
It also helps businesses whose in-store experience is important — and hard to recreate online. Victoria’s Secret, for example, already uses Google’s AI-powered visual product search, allowing people to upload an image and find a similar style. It plans to build on this with more advanced search capabilities that use generative AI and a generative AI chatbot, says chief customer officer Chris Rupp. This mimics the experience of going into a store and asking an associate to find a bra similar to one they are already wearing. “If you go into a Victoria’s Secret store, you can show an associate your bra strap and an associate could tell by the strap and hardware what type of bra it was and get you to that section,” Tharp says. “This is the most similar.”
The retailer plans to further enhance search and recommendation capabilities alongside introducing a generative AI chatbot, says Rupp.
Overcoming the barriers to uptake
Generative AI requires large data sets, which can be a challenge for smaller companies. However, experts believe there are ways around this. De Datta says that reviews, in particular, can be a great source of data on products, in addition to social media posts and product descriptions or images.
Retailers without large data sets can also use tools designed for e-commerce that pull data from other shopping sources. Xgen AI, which just came out of stealth last week, enables what is dubbed “composable AI” specifically for e-commerce, meaning that retailers can pick and choose from various options to “compose” their own AI search tool. It is already used by those including Reformation and Khaite. Companies can select from 22 proprietary AI models to essentially “plug and play” AI capabilities without needing to know how to code, says Xgen AI CEO and founder Frank Faricy. Xgen AI also enables retailers to test results in a sandbox, meaning that it can test how various options, such as enabling it to understand other languages, impacts results. “If someone says, ‘I want to look like Barbie’, even with zero relationship to the customer, brand or products, you can get a result,” Faricy says.
While search results might better “read” a vague sentence and conversational chatbots might sound more familiar, that doesn’t mean they yet have efficient human intuition or problem-solving skills. A recent search by Vogue Business for “winter white trouser suite”, with the word “suit” intentionally misspelt, returned multiple results for suits that can be worn in winter. A white one was part of the results, but it wasn’t the first result; therefore, a store associate might have been better able to interpret that “white” was part of the “winter” phrase — i.e., “winter white”. Another search — “What should I wear to Paris Fashion Week” — returned less helpful results: strapless floral dresses. (The rental retailer also invited customers to share feedback on the quality of their AI search results.)
It must also be trained on brand voice, yet still needs guard rails and oversight. A test of one luxury retailer’s generative AI chatbot, currently in pilot, recommended a perfectly nice piece — but it wasn’t sold by that company. One person intentionally tricked a ChatGPT-powered bot into offering a car for $1. Brands also need to ward against those who intentionally try to trick the bot into untoward content.
There is still considerable distrust of AI, especially when it seems to mimic humans. That’s why many brands have taken to explicitly labelling their chatbots as an “AI assistant” instead of pretending that a human associate is on the line. Trust in a brand drops 144 per cent for customers who know a brand is using AI, Deloitte reports. Victoria’s Secret plans to “keep a close eye” on customer feedback and AI performance metrics through social listening, focus groups, surveys and analysing trends, Rupp says.
Utilising the tech also requires a shift in customer behaviour. Even though Google, for example, has significantly advanced beyond stilted keywords, a generation that grew up on the internet might find it unnatural to subsequently translate their internet searches back into human-like correspondence. “We have turned our brains into keyword generators,” De Datta admits.
Tech executives advise companies to test now, comparing the current period to the development of search in the 1990s. Already, 70 per cent of retail executives are looking at implementing generative AI this year, Tharp says. Some anticipate that if big tech is a first mover, the customer will come to anticipate these tools on smaller retailer sites — the same way that Amazon trained customers to anticipate free and fast shipping and Google made users accustomed to sophisticated search results.
“You have to look at the major platforms and experiences that touch the consumer because as they change their experience, then the expectation of retail will elevate,” Tharp says. “If one retailer had started this journey two years ago, I would tell you that traffic wouldn’t naturally start using it because they weren’t used to it. But as we bring it into more platforms and social media platforms start having it, then you will see consumers drive that expectation.”
To receive the Vogue Business newsletter, sign up here.
Comments, questions or feedback? Email us at feedback@voguebusiness.com.
