The Past, Present & Future of E-Commerce

Online shopping has undergone a revolution in the past decades, evolving from basic keyword searches to personalized recommendations. However, even current personalization efforts fall short of delivering the instant, tailored experiences today's consumers demand. Looking ahead, the next frontier lies in hyper-personalized shopping powered by embeddings. In this article, we will explore the past, present, and future of e-commerce personalization - from early search to merchandising algorithms to the coming age of AI agents. We will chart the path retailers must take to provide the deeply customized shopping experiences needed to succeed in this new era. By examining key innovations in finding and recommending relevant products, we will uncover what the future of truly individualized commerce looks like.

The Past

Just a decade ago, the landscape of e-commerce looked very different than it does today. In the early days of online shopping, rudimentary keyword searches were the primary way that customers could find products on retail websites. E-commerce was still in its infancy, and the user experience was simplistic at best. Amazon set the standard with its search bar being the most prominent feature on its minimalist homepage. Customers would type in a product name or description and hope for reasonable results. The burden was on the user to hunt for items using the limited search tools available. Even as late as the mid-2000s, searching for products online remained an imprecise process, full of frustration when the right items weren't surfaced. During this early period, the focus for e-commerce companies was optimizing their search algorithms and building out their product catalogs. Platforms like eBay and Craigslist relied on simple searches to connect their buyers and sellers. The priority was mainly on establishing inventory rather than curating a personalized shopping experience.

It wasn't until recommender systems, predictive analytics, and other more advanced technologies matured that e-commerce sites could begin moving beyond just search. But in the early days, search reigned supreme as the digital shopping experience was limited by the constraints of available technology. E-commerce retailers focused their efforts on helping customers find what they wanted through improving keyword searches, not yet envisioning the personalized shopping journeys that would eventually become commonplace. This stage of e-commerce is defined by David Sandstrom, (Chief Marketing Officer, Klarna) as "People Searching For Goods”. Companies that succeed in this stage of e-commerce are the ones who empower their search engines with semantic search. The old keyword search mechanism became insufficient due to problems such as word ambiguity, synonyms, misspellings, and more. Furthermore, keyword searches were not able to understand the context and return relevant content. Therefore, companies then utilized LLMs to improve their search accuracy. LLMs can turn any data into vector representations. This provides a semantic understanding of any data, which removes the points of failure that occur because of traditional keyword searches.

The Present

Even though we have come a long way the search era is coming to an end. Now, there is no doubt personalization is a non-negotiable requirement for e-commerce success. Shoppers demand tailored experiences and will abandon brands that fail to provide them. Research shows that consumers expect brands to demonstrate they know them on a personal level. 71% of the consumers expect personalization while 76% of them think they experience getting frustrated if they don’t receive personalization and that costs more than $500B in avoidable user churn. (Accenture Newsroom)

It seems buyers now want to spend time on shopping applications just like they are walking through a shopping center and exploring products that they were not looking for. Again, in the words of David Sandstrom, this brings us to a new state which is “Goods Searching For People”. The personalized commerce revolution has already taken hold in Asia, transforming user expectations. Research shows that in the Chinese model, 80% of the purchases are driven by recommendations(Nielsen, 2019), and this ratio was almost 10% just a couple of years ago. There is also a rapid increase in social commerce usage in Asia because users have been receiving better personalization at those applications. Furthermore, applications like TikTok made people more addicted to personalization, and users are now used to finding what they are looking for without searching or even noticing what they are looking for.

Value of social commerce sales worldwide from 2022 to 2026, (https://www.statista.com/statistics/1251145/social-commerce-sales-worldwide/)

Now, as this shift reaches other markets, retailers face a critical juncture. To keep pace with rising consumer demand for hyper-personalized shopping experiences, brands must leverage next-generation technologies. Those who fail to adopt these advanced personalization capabilities risk disappointing customers and losing business to savvier competitors. However, retailers who quickly implement personalized vector searches can gain a competitive advantage. By providing the customized product discoveries and recommendations users have come to expect, these retailers are poised to attract and retain loyal customers, outperforming those still relying on outdated methods.

While some e-commerce giants like Amazon and Klarna have experimented with personalized feeds, even they lack instant, hyper-contextual personalization. Hobbled by legacy systems, most platforms - giants included - can only achieve basic segmentation-based personalization, reaching just (10%) of their potential. Across Western markets, the share of recommendations driving purchases remains closer to 20% rather than 80%, indicating ample room for improvement. To improve these numbers, retailers need to get personal at a granular level if they want to drive conversions and loyalty. Currently, the most advanced personalization leverages LLMs too. As we mentioned LLMs are helping us to turn any data into vector representations and we can conduct vector search on top of them. Rather than relying on keywords, vector searches utilize embedded item data to deliver hyper-contextual results. This allows for personalized product feeds that dynamically adapt in real time based on individual user signals. Shoppers see items matched to their interests, not generic suggestions. The rise of embeddings has profoundly impacted the personalization algorithms powering e-commerce recommendations.

Previously, most systems relied on simple collaborative filtering approaches. These matched users to similar behavioral profiles, recommending items based on what similar users purchased. This entire system relies on an idea that argues if User A and User B like Item X, then it is more likely for User A to show interest in another item that User B liked previously. However, collaborative filtering has major limitations. It can not account for context, requires extensive user histories, and often leads to homogenized, repetitive suggestions. The algorithms also fail to capture the nuances of a user's interests. Embeddings enable more advanced, contextual algorithms by encoding latent representations of items and users. Product embeddings contain semantic information about an item's attributes and similarities. On the other hand, User Embeddings takes personalization to the next level by generating unique embedding vectors for each user to reflect interests across diverse contexts. These vectors encapsulate the user's unique interests based on their browsing behavior, clicks, likes, and other interactions. By inputting these dense vector representations into neural networks, retailers can support personalized ranking, contextual recommendations, and advanced natural language processing. Or they can simply leverage User Embeddings to hyper-personalize their existing vector search results in real time. As users interact with products, their embedding vectors dynamically shift, delivering hyper-contextual results tailored to their intent at that moment. Then by retrieving the most relevant items by comparing them against the product embeddings, they can provide personalized feeds. This facilitates the exploratory, serendipitous shopping experience that consumers have come to expect. User Embeddings create the ideal symbiosis between user signals and vector search that powers next-generation personalization. Without embeddings, most personalization algorithms lack the semantic understanding of products and users required for true individualization. Leading retailers are already augmenting their product feeds with this technology to remain competitive. The brands that fail to do so risk disappointing customers and losing mindshare. As personalized vector search becomes the norm, User Embeddings provides the ideal solution to keep pace with user expectations.

The Future

Stepping into the future, it appears that the future of e-commerce is headed toward AI-powered shopping agents that can help consumers find the right products, make recommendations, and even complete purchases. In physical stores, shoppers have always enjoyed the assistance of salespeople to motivate and guide their shopping journey. Consumers appreciate having that human touch. Some e-commerce platforms like Klarna are trying to replicate that experience online with live chat support. However, providing human assistance simply doesn't scale online when millions of shoppers visit a site daily. Instead, AI-powered shopping agents offer a more scalable and cost-effective solution. These virtual assistants can offer consumers customized guidance and product suggestions using the latest advancements in natural language processing. But for these AI agents to be truly useful, they need to build an intimate understanding of each shopper. This is where User Embeddings becomes critical. By creating unique personal AI memories that capture every user’s individual preferences and interests, retailers can power intimate, human-like AI shopping assistants. As e-commerce grows, intelligent virtual shopping agents built on User Embeddings have the potential to transform the online retail experience. Consumers may enjoy shopping from home even more than visiting physical stores. User Embeddings help create the ideal foundation for next-generation AI that can provide a personalized, conversational commerce experience that consumers will love.

Heads Up

While delivering hyper-personalized experiences with AI is promising, privacy remains the most crucial consideration. As companies collect more personal data to power recommendations, they must place privacy first. Upcoming regulations and increasing consumer awareness will hold brands accountable for how they use private information. This poses a major obstacle to personalization efforts. However, User Embeddings offers a compliant-by-design solution that preserves privacy. Because User Embeddings operate at an anonymous level without requiring any personally identifiable information, brands can avoid regulatory violations or misusing consumer data. With User Embeddings, e-commerce players can build intimate user profiles that fuel accurate personalization while fully respecting user privacy and consent. Overall, this anonymous approach will be key for brands seeking to provide customized experiences in an ethical way that earns consumer trust. User Embeddings allow retailers to advance personalization while putting privacy first.

Takeaway

As we reach the end of this exploration, it is clear that User Embeddings and vector search are pivotal technologies for retailers looking to provide hyper-personalization. As the personalized commerce revolution expands globally, brands must evolve to meet rising consumer expectations. Those that quickly adopt vector search will gain a competitive edge. By leveraging these advanced technologies, retailers can finally deliver the hyper-personalized shopping experiences that consumers demand. Shoppers will reward with loyalty the brands that understand their unique preferences and provide customized product recommendations. On the other hand, retailers that fail to implement User Embeddings risk disappointing customers and losing out to more innovative competitors. With user privacy top of mind, anonymous User Embeddings enable ethical personalization that builds consumer trust. For Western retailers hoping to succeed in this new era of individualized e-commerce, the time to deploy User Embeddings is now. Companies that leverage these technologies will be poised to attract and retain customers, outperforming outdated business models. In short, User Embeddings is the key to unlocking next-generation personalization and winning in the personalized commerce revolution.