The data has spoken. It’s not enough for ecommerce site search to be “relevant.” Results must be attractive to drive revenue.
We analyzed over 609 million searches across 100+ leading ecommerce sites to measure click through rates for products ranked by traditional index factors versus a more personalized matching mechanism: attractiveness.
The difference is staggering — attractive products win nearly twice as many clicks as everyday “relevant” results…
But what does attractiveness actually mean?
First, let’s recap what goes into the relevance recipe.
When search applications are based on keyword matching algorithms, they capture textual index factors like keywords in the title, product description, reviews, category taxonomy, and attribute tags. They then calculate relevance by the frequency of keywords and the “weighting” of where these keywords appear (e.g. a keyword match in the title is given more points in the algorithm than in the description).
Notice that none of these factors have anything to do with the customer!
Although keyword-based search engines may have the ability to understand synonyms and use natural language processing (NLP) to extract the semantic meanings of user queries, these engines are missing some key ingredients that make product listings tasty for customers.
What Makes Search Results Attract Clicks and Conversions
In short, personalization.
A little longer: attractiveness represents the likelihood that an item will be clicked on, added to cart, and/or purchased for a specific search query entered by a specific user.
But let’s expand on this fully. Here’s what goes into a great attractiveness algorithm:
Clickstream data
What people click throughout their shopping journeys tells a smart search service like Constructor a lot more about a shopper’s likelihood to click or buy something.
We track behavioral data from individual users and the collective across visits and search queries. We capture the micro-interactions that happen between, including clicks, favorites, adds-to-cart (and removes-from-cart), next-clicks, paginated clicks, and more.
We also track what users scroll past (a.k.a. ignore), how deep they explore, how they refine search queries, and how they filter results.
Capturing all this data at scale, we can layer in these insights with relevance factors to tweak product rankings in a way that matches what an individual shopper is most likely looking for.
Purchase history
Of course, not all clicks turn to conversions. We can track-back sales to which search queries drove them, and which types of customers ultimately bought. This data can be mapped to individual shoppers and their own purchase (and return) history.
First-party data
Beyond clicks and conversions, what has an individual searcher told you about their preferences? Have they filled out profile data, clicked through email and SMS campaigns, saved products to favorites or searched for specific brands? What sizes do they typically buy (and return)?
First-party data is a powerful piece of the personalization equation.
User context
Factors like a user’s device type, geolocation, time of day, and even current weather are all relevant to what makes search results more or less attractive in any given moment.
This is useful under so many search conditions. Weather and climate impacts which apparel, accessories, furniture, food, and so many other product types are clicked or ignored. And when it comes to personal style, this is often influenced by a shopper’s zipcode (including how much they’re willing to spend) — all things your clickstream and sales data can inform.
Product factors
Let’s not forget product-level factors, both static (brand, price, features, materials, and other attributes) and dynamic (popularity, “newness,” seasonality, user reviews, available sizes, inventory status, discounts and promotions, etc.).
Most “machine learning” search applications only factor “popularity” to boost products based on recent traction. This means “the rich only get richer,” serving generally trending items for all searchers. Less trendy items simply don’t get seen, and won’t get sold.
By factoring all product factors in context with clickstream and user data, Constructor’s attractiveness algorithm does true “matchmaking.” By showing the right products to the right users, you can improve inventory turnover across the board (and give users a better overall shopping experience).
Keyword relevance
Of course, relevance is still relevant to great search results. Constructor incorporates keyword and semantic relevance into the overall attractiveness score. This is search, not browse, after all.
From Relevance to Revenue
Visitors who search are a VIP segment. Our research found even though searchers make up only 24% of site traffic (on average), they make up 44% of total site revenue and convert 2.5x higher than non-searchers.
But they convert best when search results are attractive, and we can prove it.
Are you leaving money on the table?