Semantic search for ecommerce site search – do you need it?

Posted on January 19, 2017 by


Semantic search as per Wikipedia seeks to improve search accuracy by understanding the searcher’s intent and the contextual meaning of terms as they appear in the searchable database, whether on the Web or within a closed system, to generate more relevant results.

Some of the examples of how this may translate in ecommerce site search for a fashion retailer are, search queries that read like “Dresses for spring” or “Best selling moccasins in 2016 April” or ” Trendiest shirt for a cocktail  party”. You get the drift. So will it make an impact on your conversion rate, if your search engine has the potential to show accurate search results to these search queries

  1. Data: In our cumulative analysis of search query data across clients across verticals, less than 0.07%, meaning 7 out of 10000 search queries, were semantic in nature. Semantic search becomes relevant if the revenue generated from the x% who convert from the 0.07% is considered as significant to your business and also offsets the increase in cost for this added ability . Even sites like Amazon, Walmart and Target where 1% of their search volume could lead to millions of search queries, do not place a heavy emphasis on Semantic search for many reasons. Some of which could be ROI, customer experience or latency in the display of search results, that could increase search bounce rate. Convinced and want to  SIGNUP NOW?semanticwalmarttarget
  2. Frictionless experience: As suggested in our article best practises in site search, consumers expect to be fed. They are spoilt and prefer to have predictive suggestions that match intent, hence almost 99.07% of your audience is more likely to click on a suggestion or type a short query, vs. a fully type a semantic query. If you notice a trend on certain contextual search queries or terms (e.g., cheap, under, newest etc.,) from the database, that can be managed without a semantic system, but creating a rule set or synonyms libraries to tell your search index what products are potentially relevant for that query.amazon ss target ss walmart ss
  3. Consumer search behaviour: When it comes to visitors who choose the search funnel, most of them already show intent or interest in purchase. Hence search queries are typically tied to products (dresses, shoes, beds etc.,) or combinations of products and tags (Maxi dress, Nike Shoes, Single beds etc.,).
  4. Vertical: Semantic search is usually more common in a contextual database vs. a product database. Most visitors who visit your online store are looking to buy products. And unless you are selling something that is not easily understood to the end customer, most of your visitors know what to search for
  5. Performance vs Price: Lets say that over 1% of your search volume consists of queries that are semantic in nature and you need a site search engine to deliver results. As a product manager this leads to leveraging an NLP system than can process or cache results for a semantic query in less than 50 ms. Anything longer than that, is going to slow your results page and that can lead to bad search experience. Secondly, there is an increase in cost to build an maintain a system that can not only process these queries with accurate results, but also show these results to your visitors like any other search query. Is that increase in overall cost, going to be offset an increase in revenue driven by semantic searches?

SIGNUP NOW, join a free plan or a free trial to know whats best for you. 


Posted in eCommerce Site Search
Tagged in , , ,