Soon shopping for a new dress online might feel as personal as buying one in a store.
At least, that’s the hope of the several startups and tech giants that are building AI agents, which are meant to assist shoppers in making online purchases, doing everything from surfacing personalised product recommendations to completing transactions on a user’s behalf. In the last year, AI-powered search engines like Google and Perplexity have launched these features, while in September, OpenAI announced a partnership with Etsy and Shopify to allow users to check out directly through ChatGPT.
But there’s also an emerging class of AI-powered e-commerce upstarts creating their own AI assistants, including shopping app Vêtir; price comparison platform Phia, co-founded by Bill Gates’ daughter Phoebe Gates; rental marketplace BNTO and OneOff, which recommends products based on what users’ favourite celebrities are wearing.
The pitch is that these agents can ask questions to understand a user’s specific preferences, which then allows them to make better recommendations than personalisation technology that solely pulls from a user’s shopping history. Plus, they provide the convenience of completing checkout for the user, too. For the retailers, a better shopping experience means consumers are likely to click purchase — and buy more when they do.
But building an AI agent specifically for shopping is complicated. For starters, the tool has to match inventory across different retailers to specific users’ preferences, and while people are already using AI-powered search giants to discover and buy products, shopping agents are not yet widely adopted. The bar is even higher for emerging platforms, which will have to compete with deep-pocketed tech giants like Perplexity, Google and OpenAI, as well as brands like Ralph Lauren.
Those most likely to stand out will create agents with unique features — such as identifying what people need in their closets without being prompted or generating avatars set where a person is most likely to wear an outfit. And despite their smaller footprint, these upstarts are still winning investor confidence: Phia raised $8 million in September; OneOff raised nearly $4 million in the last year; Gensmo has snagged $70 million in the last two years and Vêtir is planning to close a nearly $13 million round in the next month.
“There’s going to be a bit of an arms race,” said Juan Pellerano-Rendón, chief marketing officer at e-commerce software startup Swap, which launched a new platform in September that creates AI shopping agents for brands like the British menswear label Percival to offer on their sites. “It’s one of those situations where you need to move quickly; otherwise you’ll be left behind.”
The Case for Creating Your Own Agent
The new wave of e-commerce startups is creating AI agents with distinct capabilities the big tech firms don’t offer.
Search engines like ChatGPT or Perplexity can recommend products based on detailed prompts, such as someone looking for a flowy dress for a wedding in Napa Valley. But those results are often impersonal and informed by sources across the internet, from articles to YouTube videos, rather than a person’s individual taste. AI agents made specifically for shopping platforms, on the other hand, can help users understand exactly what they’re looking for by prompting them with the right questions to generate precise results, said Katherine Black, partner at global consulting firm Kearney.
“If I write a prompt that is 10 lines deep and I’m telling that platform all about my measurements, my size, what my mission is, it can give a fantastic recommendation,” Black said. “Most users aren’t going to necessarily write that type of query. If fashion-specific technology can get a user to share that information … then that can give them a real edge.”
These agents are also designed with each platform’s niche selling model in mind. Vêtir’s AI agent, for example, scans users’ virtual closets to identify gaps in their wardrobe based on upcoming events on their calendars, offering styling and product recommendations based on items users already own. Once a customer decides what they want, the agent then orders on the user’s behalf, said Kate Davidson Hudson, Vêtir’s founder and chief executive.
“People want smarter suggestions,” Davidson Hudson said. “They want speed and they want efficiency.”
Navigating Complex Technology
Most AI agents are built using open-sourced large language models from companies like OpenAI. But to create an agent for a particular platform requires additional engineering to ensure those LLMs can scan a site and register details like sizing and update when new inventory comes in, said Pellerano-Rendón.
BNTO launched an agent in September called Maya that can tailor recommendations based on whether a shopper is renting or looking to buy an item outright. Maya makes those recommendations through Alma, BNTO’s AI-powered platform that stores details about an item’s physical characteristics, such as fabric or shape, said Sixuan Li Pasinetti, BNTO’s founder and chief executive.
AI agents also have to be prepared for complicated interactions with users, going from answering questions to ordering goods and back again.
“There are a lot of moving pieces that, [when] siloed, are a little bit easy to achieve, but then once you integrate them all together into one experience, it has to dovetail in an efficient way,” she added.
Some are choosing to outsource riskier capabilities. AI-powered shopping app Gensmo built its own AI agents that can transpose products onto users’ avatars and then display those avatars into specific settings — like a party or dining al fresco — so shoppers can visualise how they might use the product. But for its agent’s checkout function — which will place orders across brands and retailers’ websites on users’ behalf — the company is partnering with software platforms that already have the infrastructure to protect users’ credit card information, said Ning Hu, Gensmo’s founder and chief executive.
“If you look at the implementation details of this, it’s actually quite messy,” she added. “In order to do it right, we need to be extra careful because it involves personalised information.”
OpenAI and Google are trying to make it easier for agents to complete purchases by introducing a set of rules that ensure those transactions are authenticated and authorised by users, which can be used by any brand or retailer.
Making Consumers Care
Perhaps the biggest hurdle for AI agents is convincing consumers to use them en masse.
“The consumer hasn’t fully demanded nor adopted this yet,” said Sonia Lapinsky, fashion lead at consultancy AlixPartners. Because of that, AI agents “have to be really good, and then it’s likely going to be a lot of marketing funds to get their name out there so people realise they’re really this good.”
Vêtir, for one, is enlisting influencers, including jewellery designer Jennifer Fisher and former fashion editor Kerry Pieri, to create content showing them using the platform’s agents to shop, and hopefully convincing their fashion-forward audiences that Vêtir’s tools are “additive” to their shopping experience and “not intrusive,” Davidson Hudson said.
Some startups plan to use their agents’ most advanced features, such as checking out on consumers’ behalf, as a customer retention driver by making them available to frequent users who have already grown to trust the platform.
OneOff, which uses AI agents to match images of celebrities and influencers’ style with products sold at retailers like Mytheresa and Net-a-Porter, is officially launching in October. After launch, it will slowly roll out a checkout tool — which places orders for users instead of them having to manually input their information on each site — to shoppers who spend the most time on the app, said Emir Talu, the company’s founder and chief executive.
Time will tell if consumers will actually use AI agents. But building them is a great exercise for emerging shopping platforms. The best move for startups right now is to “start to build a unique data source to train a model that can create differentiation in the long term,” Black added.
