• March 28 2024

Introducing The Future of Retail Product Discovery: Multi-Search

Brendan O’Shaughnessy, Chief Commercial Officer

Anyone who knows me, knows I’m quite partial to navy blue polo shirts. Simple, smart casual styling that suits my coloring, and allows me to look professional without feeling overdressed. (Please don’t judge my fashion taste).

However, I also quite like collarless polos for a more casual look. The challenge is finding one online. Yes, Google understands what a collarless polo is, but try searching with that term on major sports brand websites… the results aren’t good.

On nike.com, a recent search for ‘polo’ returned 60 products. Searching for “collarless polo” presented me with 39 products. Much better I thought, until I started browsing the results. Of the 39 shirts presented, only 3 were collarless – a less than 10% relevancy. All the rest were polos with collars.

On another major sportsbrand.com website, the results were either better or worse depending on your point of view. This site’s search engine didn’t recognize the term collarless (worse), so it ran a search for polos instead, delivering 224 results with 100% relevancy (better). But it couldn’t deliver me what I was looking for.

Why is it so hard? 

The genesis of our frustration with e-commerce search lies in outdated systems. Traditional keyword-based search engines rely on literal matches, incapable of interpreting the nuances of human language and intent. Poor or inconsistent data classification across e-commerce catalogs, coupled with a lack of context in the search algorithms, which have largely remained unchanged for a quarter-century, compounds the difficulty. So, whilst the status quo has been in place for a long time, the market is about to see the arrival of new search options that can address this challenge.

Today’s consumers demand a search experience that is not just accurate but also intuitive. The search above; collarless polo, could have been solved using current Visual AI Search technology.  Allow shoppers to upload an image of the product they seek, and if the site has similar products in stock, they’ll get very relevant results (80-90%, not 10% as per my example above). But despite being around for many years, Visual Search is not as widely deployed on e-commerce sites. Whilst Visual Search is proven to deliver better results, merchants mostly consider the adoption rate (often <5% of consumers) and overlook the fact that it drives up to 4X conversion and 2X AOV when compared to keyword search.

For those of you who don’t already know, Visual Search works so well, because it’s based on a vector search engine that harnesses machine learning to transform images into numerical vectors (a string containing a lot of numbers). This allows the system to perform a K-nearest neighbors (KNN) algorithm, mapping out visually similar products accurately and efficiently. The result is not just a list of products but a curated selection that mirrors the shopper’s vision.


Vector search however is not limited to images. Vectors can also be created from text, such as product title, description, price, color, or any other combination of attributes that best describe a product. Better still, when you combine image and text vectors with keyword search capabilities, you’re able to create a new category of search engine  – a hybrid search engine.  This Hybrid Search significantly improves the relevancy of text search and supports the ability to search concurrently with text and image. In layman’s terms, this means being able to search for a specific product, a similar product, or even a complementary product, from one search engine. 

This evolution from an unimodal to a multimodal search platform is only the beginning of the coming transformation. This new hybrid search capability becomes even more powerful when it is coupled with a configurable Large Language Model (LLM) and advanced product recommendation algorithms. Such a system can support a range of product discovery experiences far surpassing the limited scope of a traditional search bar. Imagine allowing your shoppers to write their own product recommendations, whilst viewing a particular product: “Show me a bag and matching shoes for this dress”. Or for them to be able to seamlessly move from search to shopping assistant when appropriate.

So, where are we at today?

At ViSenze, we have spearheaded the development of vector search technology and solutions that power Visual Search, Tagging, and Recommendations for over 10 years. So, in mid-2022 when we first started looking at how we could build a better search engine, our vector architecture was our natural starting point. The journey since then has been driven by meticulous customer problem analysis, extensive research, steady evaluative testing, agile development, and a lot of great work from many colleagues. 

March 15th marked a quiet but significant milestone with the release of Multi-Search, extending our visual search service to incorporate textual inputs, and enhancing the relevance of visual searches with unparalleled precision.


Rolling this first Multi-Search service (R1) out to our existing visual search customers is underway and we can’t wait to share the results with you over the coming months. 

What’s Next?

The excitement builds as we approach the release of Multi-Search (R2) in early May. This isn’t just an iterative improvement but a radical departure from basic keyword searches to a fully-fledged hybrid search engine optimized for e-commerce. Our solution will be accompanied by a suite of web widgets, allowing for rapid deployment that will support multiple shopper interactions that you can integrate throughout your online shopper journey. All are supported by API and SDK for a tailored fit to any e-commerce platform.

The month of May will also see the unveiling of Multi-Search R3, integrating our Hybrid Search and Visual Embeddings with the latest in conversational AI technology, Chat GPT 4.0. This will empower an interactive Shopping Assistant and Product Finder service, bridging the gap between search and purchase with ease.       

It’s an exciting time to be in the commerce search game. ViSenze’s Multi-Search is not merely a new tool; it’s a harbinger of a new era in e-commerce. It heralds a future where the disconnect between shopper intent and search results is finally overcome. Shoppers will no longer have to wade through irrelevant options, or repeat their search attempts; rather, they will be presented with results more closely matching what they looking for, be it a collarless polo shirt or a matching accessory for their latest purchase. For merchants, the implications are transformative: heightened engagement, reduced bounce rates, and escalated conversions, culminating in robust revenue growth.

We’re just getting started with the next generation of search technologies, and if you want to join us on this journey, find out more about ViSenze Multi-Search and how it can help grow your business, Request a Multi-Search Demo!

About the Author

Brendan O’Shaughnessy is the Chief Commercial Officer at ViSenze. Brendan has over 30 years of experience in sales and business development across a range of industries including media, technology, and consulting.