How Emerging Tech Uplevels The Ecommerce Product Discovery Experience

Search and product discovery has evolved immensely since its early days, continually shifting to keep pace with ever-changing customer expectations and advancements in technology.

What started with keyword-based engines, which were limited in their ability to return attractive products, evolved to include vector search. While the most innovative vector search engines widened the net of relevancy, they still didn’t solve well for attractiveness —the idea that search isn’t just about showing “relevant” products, but about showing products that are most likely to be purchased by a given user in a given context. 

Thanks to quickly evolving AI and machine learning (ML), clickstream-based approaches came to the forefront, allowing retailers to present highly personalized results to end consumers. And nowadays, product discovery vendors are advancing even further, experimenting with emerging tech like transformers, which allow for an even deeper understanding of what users are looking for when they search, browse, and interact with product recommendations.

At Constructor, we recently released a new white paper, “Forward. Fast. The Future of AI in Product Discovery,” that explores the impact of  emerging technologies in enterprise ecommerce.

Here’s an excerpt from the white paper that explains why transformers, large language models (LLMs), generative AI (GenAI), and clickstream data are so appealing for large ecommerce brands.

Transformers 

Transformers are advanced algorithmic models trained on large ecommerce data sets. They’re built on deep learning architecture with a natural language processing (NLP) model to weigh the importance of different words in a sequence. 

Prior to their advent, leading AI translation methods relied heavily on recurrent neural networks (RNNs). These RNNs processed each word sequentially within a sentence. However, significant relationships among words often exist, regardless of whether they are sequential. Transformers use a mechanism known as attention, enabling models to recognize the relationships between words — irrespective of the distances between them — and to identify the critical words and phrases within a passage.

And because of the way they’re built, transformers can better comprehend the context of each term in a query, making them useful for processing complex natural language search queries (i.e. “find me hiking boots under $150 that come in women’s size 8”). 

Why they matter in ecommerce.

The more ecommerce data you train your transformers with, the more fine tuned your base model becomes to your specific use cases, regions, unique retailer attributes, etc. The outcome is an iterative process that continually evolves based on the depth of insights ingested, enhancing the precision of the overall product search and discovery experience. 

And thanks to having a deep understanding of a query in the context of clickstream, the use of transformers allows ecommerce companies to return more attractive and personalized results across their entire suite of product search and discovery tools — every interaction, every time.

This is different from vectors, which increase the number of products returned for a query, but don’t have much understanding of whether a product is attractive nor ensure user-level attractiveness.  

Large language models (LLMs)

Large language models (LLMs) are a category of AI models that have been trained on massive amounts of text data to perform natural language understanding and generation tasks. 

These models are also typically based on deep learning techniques, particularly transformer architectures. This allows them to mimic an understanding of the world from which the data comes, which has great implications for large ecommerce companies that send LLMs large amounts of domain-specific data.   

Why they matter in ecommerce.

Training LLMs on massive amounts of ecommerce queries and clickstream creates a model that can mimic what people really want in a given situation.

As an example, through the clickstream, the model can start to understand that when someone searches for “butter,” they mean a product like salted butter, not peanut butter. Or, when they search for “pants” in the U.S., they mean trousers, but when they search it in the UK, they may mean underwear. This can all be learned from the clickstream, which allows you to see what people buy versus what they scroll past in each of these contexts, to fine tune the system.

Clickstream data

Clickstream data is first-party behavioral data that can be used to understand shopper behavior onsite along with how the popularity of products changes over time. 

With clickstream, you go beyond simple segmentation, which relies on third-party data and retargeting. Because you’re collecting customer data anonymously in real-time, you provide shoppers a hyper-personalized product search and discovery experience that surfaces products that are attractive to them as early as the first session.

Clickstream data also simplifies the near-impossible task of layering your business KPIs on top of user-level personalized search results, allowing your team to further optimize product rankings sitewide. 

Why it matters in ecommerce.

Enjoying fully automated personalization on a user-level across your entire ecommerce site has tangible business benefits, directly boosting conversion rates, AOV, and RPV. 

Ranking products in the context of clickstream is also extremely important as ecommerce companies continue to grow sales channels, expanding into marketplaces. 

Since marketplace products aren’t always proprietary, companies don’t always have direct control over third-party catalog data. There’s often misalignment pre-purchase, as merchandisers take different approaches with recommendation pods and collect clickstream only from the website, disregarding the store and other channels. Pair this with the fact that shoppers are getting better at describing what they want, moving beyond simple keywords to search.

Leveraging AI-native product discovery tools that position you to excel in this new reality and allow you to merge data across channels is paramount for business success. With Constructor, companies can do just that. 

Powered by our Native Commerce Core™, our platform offers a holistic approach to product discovery that helps ecommerce companies track omnichannel user behavior from one dashboard, acting as a single source of truth to solve inputs and outputs of the entire product discovery experience. 

Whether customers are searching, browsing, or getting recommendations, all products learn from each other, leveraging the same clickstream data to provide one unified experience for site visitors and ecommerce teams alike. 

Generative AI (GenAI) 

At a high level, generative AI (GenAI) refers to the use of AI models to generate content, responses, or suggestions. It goes beyond traditional rule-based systems by leveraging ML algorithms — often based on neural networks like transformers — for language understanding. 

In the context of large amounts of clickstream data, it provides an even more tailored product search and discovery experience, where shoppers can quickly and easily find attractive products in one session — something that sounds easy in theory, yet is difficult in practice. (A recent study found that less than a third of U.S. consumers describe their most recent experience finding products on a retail website as “enjoyable.”) 

This is where product search and discovery technology based on GenAI becomes so appealing. 

Why it matters in ecommerce.

There are a mix of customer-facing and back-end GenAI use cases in enterprise ecommerce that provide tangible business value while automating redundant merchandising tasks, ultimately saving them time, improving productivity, and allowing them to focus on strategic initiatives.

One such use case is to create dynamic landing pages based on customer on-site behavior, such as a collection of gifts for runners. Every time a customer clicks and interacts with your site, they’re essentially voting for products they want to see. An AI-native search engine like Constructor’s pays attention and dynamically re-ranks products to show the most attractive items site-wide. 

This technology can be integrated into existing search, such as having a pop-up with options to search intelligently, or an intent hub for conversational commerce interactions. You can also have an AI search toggle or generate AI search suggestions within the search bar. 

GenAI can also be connected to your product catalog and used to improve your search infrastructure. This is possible with Attribute Enrichment. Using a mix of AI plus new innovations in machine vision and text classification, our models tag your products with new, relevant attributes and categories automatically, leveraging clickstream to prioritize the attributes that are most important to your buyers in real-time.

And finally, you can also leverage GenAI to create personalized and engaging experiences for different types of users through conversational commerce. This is seen through Constructor’s recently released AI Shopping Assistant. 

Shoppers enter retail sites with an intent. They might not know exactly what they want to search for or where to browse, but they often have an intention — starting a new hobby, figuring out what to cook for dinner, planning for a camping trip with kids, etc.

Overall, use of conversational commerce shortens the time it takes for shoppers to go from goal to reality, allowing them to shop all in one place and effectively addressing a fundamental problem in the way people shop: taking hours to do something simple. 

The final word

Understanding how emerging tech plays a role in building a holistic product search and discovery experience is just the tip of the iceberg. See what else there is to consider — including what Constructor CEO, Eli Finkelshteyn, predicts will happen post-ChatGPT — in order to confidently stay ahead of nascent ecommerce technologies and ensure your platform is future-fit.