Back to top

Solr E-Commerce Case Study

Search plays a hugely important role in e-commerce. It is the main “salesperson” on the site. If the queries posed in the search box are not answered accurately and quickly, and results presented in a convenient way, then usually a sale is lost.

However, better search does not necessarily mean that a new search engine needs to be purchased. E-commerce search improvement is often simply a question of attention to detail. This brief case study, based on a recent engagement of ours, illustrates this scenario. The customer uses Solr, and is a world-renown brand in consumer electronics.

BACKGROUND

The customer conducted an internal study of search on their public-facing e-commerce website. Their aim was to find ways to improve the purchase completion rate, and to increase user satisfaction (to enhance brand reputation), as measured through feedback from customers. The study also reviewed and analyzed web log files.


Their main conclusions about what required improvement will sound familiar to anyone who has experience with important e-commerce search systems, and included:

  • The need to make site navigation easier, including the interchangeable use of search and browse
  • The importance of product presentation (descriptions, summaries, images, other supporting resources)
  • Site performance. Folks hate waiting in stores, either online, or on main street
  • The overall look, feel, convenience, ease of use, and general “flow” of the website

The website was not dedicated to e-commerce, so search requests made via the single search box could also be for customer support, detailed product information, and other reasons.

It was felt that the Solr-based search system had a range of challenges to address, but that core relevancy was the key issue. In addition to answering explicit user queries, the search function is also used to contextually serve product suggestions next to non-e-commerce web content. 

ATTENTION TO DETAIL

The customer approached Search Technologies for help, and to begin the project, a deeper evaluation of the search system was conducted using our Search Application Assessment service.

Following the publication of the Assessment Report, a range of improvements were agreed and implemented. The most notable of these were:

  • Improved support for the shortened names that consumers use for products (synonyms)
  • Limit the number of search results that appear; sometimes far too many are listed
  • Improved control over relevancy in general, and for key searches, precise control over the order in which products are listed
  • Better handling of searches for global products, ensuring that only those relevant to the local market were displayed
  • Improved differentiation between searches from consumers, and searches by retailers (for whom additional products are available) 
  • The introduction of “quick links” to guide shoppers to specific categories of products. This is especially important when the search clue is vague


The additional analysis conducted by Search Technologies discovered numerous other issues, and diagnosed the root causes of problems with relevancy, and the frequent occurrence of “zero results” events. Causes included:

  • A lack of control over the scope of the search – different content types were being thrown together without enough thought
  • Underutilization of phrase searching (implemented via query processing to improve precision)
  • Inaccuracies in the product taxonomy, which were leading to user frustration
  • Under-developed semantic resources – there was a significant scope for improving search through the addition of a range of synonyms to the system, picked from search log files


As is typical with many customers, a phased approach to the project was requested, the initial phase aimed at “quick wins” to ensure project momentum was maintained.  The following improvements were implemented in the first phase:

  • Introducing query logic to intelligently create phrase searches from multi word queries (QPL was used for this)
  • Further query processing improvements helped to differentiate product and product accessory searches from customer support queries
  • Make the “AND” search operator the default mode for Solr. This substantially improves precision. Where zero results are obtained, the search is automatically re-run using “OR” instead
  • Improvement to the existing query auto-completion logic, through the addition of synonyms and improved (targeted) spell checking 
  • Taxonomy improvements, resulting in accurate search navigation options 
  • The introduction of additional sorting methods (by price, etc). This required additional content processing, prior to indexing into Solr, to ensure that metadata was in the right place to support sorting on various properties
  • Changed query logic for “Best Bets,” to exclude content from non-best bets sources
  • The introduction of a more agile “relevancy boosting” approach, which takes input from the type of document 
  • “Exact Match” logic was added, based on titles and product display names. This refinement further improved precision
  • Support for on-demand crawling and indexing of advertising campaign pages, so that new campaigns were immediately visible through search.

IN SUMMARY

Human salespeople in physical stores benefit from continual training. The search box on an e-commerce site benefits from similar ongoing improvements. This customer can immediately see, and measure the impact on e-commerce sales that the above attention to detail has provided. 

0