By Andy Narayanan, Vice President, Intelligent Commerce at Sentient
As product catalogs continue to get larger and larger, e-commerce sites need to provide shoppers better ways to navigate through their inventory.
One of the most common methods applied right now is product tagging. Generally, product tagging operations are completed by huge teams of interns or crowdsourcing providers.
Those people are given instructions and definitions and set off to tag various characteristics of every individual product in a site’s catalog. In essence, these tags add additional metadata to a product database which in turn give users more ways to find they’re looking for.
However, for products selected primarily by their design versus their specifications (think fashion items versus lumber or car parts), this approach falls short.
In addition to being expensive and time-consuming, large-scale tagging operations simply give users more words and categories to use when searching for a product.
In essence, they add more columns to a spreadsheet. They don’t make search more personal. They often make search more less useful. They’re limited because some things simply aren’t easily described. Not to mention that maintaining these alternate taxonomies can be an absolute nightmare.
AI-Powered Visual Search
Thankfully, new search technologies are providing an alternative to tagging that’s both more efficient and more intuitive. One such technology? Artificial intelligence (AI). AI-powered visual search looks at both manufacturer metadata and, importantly, product images to help users on their buying journey. It doesn’t require a user to interact with a spreadsheet, but rather lets them browse images of products they like until they find one they love.
The easiest way to show the difference is for us to highlight three shortcomings legacy tagging operations suffer from and show how AI handles each. We’ll start with a big one: subjectivity.
Problem 1: Subjectivity
No matter how specific the instructions you provide to the team doing your tagging, the whole process is subject to minor discrepancies that can substantively affect product search. For example, is a particular shoe tan or is it brown? Is it snub toe or round toe? Is the heel height short or medium? Those small differences can keep users from finding the product that’s right for them.
And that’s just subjectivity on the tagging end. Buyers aren’t a monolithic group either. One shopper might search for a “black dress” but what if she searches for a “cocktail party dress” or an “elegant, flowy black dress”?
Most search functionality can handle the first query, but the others might be more difficult. Not to mention that your definition of “elegant” may differ from the users. In the end, it’s simply impossible to account for the different vocabulary every visitor to your site might use.
With AI-powered visual search, you avoid these issues. AI doesn’t need subjective tags to help users find what they’re looking for. It doesn’t require buyers to match your nomenclature.
After the first click, it looks at the image itself to suggest similar options. Each subsequent click trains the AI, in the moment, as the user is shopping.
It understands how products are similar to each other based on hundreds of dimensions and vectors and uses those to suggest products that are not only similar to what the user has clicked on so far but also helps surface products she may never have found. It finds patterns after every click, refines on its own, and gets the user to the item they really want.
Problem 2: New facets and new products
E-commerce sites get new products all the time. And while those products will come with images and manufacturer metadata, they won’t come with the specific tags you use on your site. That means your tagging operation will have to hand-label characteristics for each and every one.
This can be a huge pain. Whether you’re using crowdsourcing or outsourcing, turnover for these operations is generally quite high. You may need to retrain interns or hire whole teams to add tags to those products. That’s not to mention that if you want to add to a new characteristic for your product catalog (such as event-based queries like “cocktail party” dress), you’ll need to tag every item in your catalog again. That’s expensive and it takes a lot of time.
AI visual search avoids that issue. It’s trained on your original product catalog and understands new images you add to your inventory. New products don’t need to be tagged, they just need to be uploaded.
This lets you not only add items to your catalog quickly, but lets users find them the moment those products are on your on site. And it saves you both time and money as you never need to employ a team to label those new products or facts for you.
Problem 3: Lack of nuance
You can only tag so many characteristics of your product catalog. Attributes like color and style, while subjective, are two simple examples.
But what about the subtle, yet difficult to describe traits that draw users to particular products?
A shopper may not even be able to articulate those qualities. She might have a preference for a certain shade of green, a certain placement of a designer logo or a certain slope of a hemline on a dress. Those small distinctions in products, especially products that are design-based, simply can’t be accounted for with legacy approaches.
This is another area where artificial intelligence excels. As the user browses through product images, she trains the AI in the moment to understand her individual sense of style.
AI-powered visual search understands the almost ineffable similarities between the images she selects. It looks at hundreds of vectors of a product image to do this to finds patterns which, in turn, helps the buyer narrow her search and arrive at the perfect choice, not just one that’s close enough.
AI Makes Search Easy, Intuitive, Fun
Product tagging and faceted search have been around for quite some time. And, indeed, there are things those approaches do well. But they require consumers to shop through a spreadsheet. They ask the user to interact with a database, understand a seller’s terminology, and dig through a catalog, category by category, page by page.
AI-powered visual shopping isn’t like that. It’s intuitive. It’s easy. And it’s actually fun.
AI more closely mirrors what it’s like to shop at a store, with the help of a fluent and helpful salesperson, someone with years of experience who can pick up on subtle preferences. In a very true sense, AI can actually pick up on someone’s sense of style. And that’s a pretty incredible thing.