Making Search Work Effectively: E-Commerce Design Patterns

Problem Summary

Solution: 

1. Understand the keywords your customers use to describe your type of products This needs to cover all levels of description from the product type/category through to the individual product and then on to the product features/benefits.
2. Match the structure of content metadata and search engine settings This means either tuning the settings of your search engine to suit your content metadata or changing your content metadata to suit your search engine. 
3. Review and, where necessary, enhance content metadata  To work effectively, your content metadata must correspond to the keywords your customers use to describe your type of products. 
4. Review and, where necessary, enhance the synonym-matching capability of your search engine . 
5. Create the search index and check search performance Indexing is done automatically by all the major search applications for e-commerce applications but benchmarking will require measurements of search performance; these benchmarks, in turn, should have been designed to achieve target levels of search conversion. 

Rationale:

1. Understand the keywords your customers use to describe your type of products What are the most popular keywords used by customers searching for that new TV for the World Cup? Flat screen TV? LCD TV? Plasma TV? No, according to Wordtracker it's actually 'hdtv' and knowing this will help you make search on your e-commerce site work more effectively than on your competitors' sites.  
2. Match the structure of content metadata and search engine settings Contemporary search engines are excellent at indexing the content of your e-commerce site and then matching this index to search keywords. They still, however, need help in understanding that certain words have particular meaning and that words with some meanings should be treated differently in processing a search request (see blog for more details). To schieve this, content needs to be structured (either with tags on the page or directly within the content management system) and the search engine needs to know where to find different kinds of content during indexing and what to do with it during search processing.
3. Review and, where necessary, enhance content metadata  Most of the problems of ineffective search arise from poor metadata. If you search for 'mobile phone' on the Debenhams site, 49 out of the 50 search results (as of 2nd June 2006) are for handbags.  Digging deeper, it turns out that all these handbags feature mobile phone pockets and hence, the search engine could be argued to be working perfectly. From the customer's point of view, however, it suggests that search on the Debenhams site is broken and shouldn't be trusted in future. Metadata needs to distinguish between product categories and product features, for example.
4. Review and, where necessary, enhance the synonym-matching capability of your search engine . E-commerce content is often highly consistent in its vocabulary - customers aren't! If your site always uses the term 'TV' then your search engine will assume you do not have a 'television'. Less obviously, fridge is spelt with a 'd' but refrigerator isn't, so there is potential for mispelling of both keywords. Some search technology solutions have pre-populated synonym tables but few will be perfectly suited to your product range.
5. Create the search index and check search performance Online Retail 2004 revealed massive room for improvement in search performance. Leading e-commerce sites, on average found only 43% of the products that perfectly matched the search keywords. 82% of search results were redundant. Through this research we identified search fidelity, search redundancy and search tolerance as useful benchmarks for assessing search performance.

Author: Mike Baxter | Blog entry on this pattern | Last modified: 1st June 2006