OPTIMIZING MAGENTO SITE SEARCH
The difference between a sufficient site search experience and an excellent one can significantly increase conversions. Consider some baseline statistics: on average, customers who abandon e-commerce sites only need about 8 seconds before moving on. Because roughly half of e-commerce customers go to the search function first to find what they’re looking for, a poor search experience translates to lost revenue.
Magento Community and Enterprise both offer flexible search options, but there are areas where these can be improved on with configuration, development and/or third party tools.
As in external SEO, site search is built foremost upon editorial content. Ensuring that products, categories and all other relevant content types are written with relevancy and keyword density in mind is key in exposing those elements to the customer in any technical scenario. Of secondary (but still vital) importance is the manner in which content is structured, for example with product attribute configuration and typing. Some guidelines to keep in mind when setting up the initial catalog:
- Call out all possible search variants in product attributes (i.e. color, size, style, brand)
- Expose a subset of relevant attributes to faceted search filters
- Weight attributes by importance when determining relevance
While relatively simple from a technical perspective, displaying the search box in a prominent position, with relevant user cues, can help to guide customers to the search function before abandonment. Sites that hide the search field behind a magnifying glass or text link are losing the majority of users who could have otherwise converted. Some best practices include:
- Displaying the search box as an “oversized” field at the top (right) of the page.
- Include grayed-out text prompts (like “enter keyword to search”) that clear on click
- Use a rich autocomplete function that returns a truncated “top 10” inline, as the user types.
Bearing in mind that each step in the customer funnel is an opportunity for abandonment, the design of any interstitial pages between the entry point an the PDP (such as search results) must be executed in a manner that surfaces relevant information and, when possible, provides new information intended to move the customer forward. Likewise, jump directly to a call to action whenever possible.
- Handle the single result case by redirecting to the PDP (i.e. in SKU searches)
- Present “buy now” buttons directly in the result set
- Handle the “no result” case with relevant product suggestions and search guidance
- Return social media markers in results (# likes, # comments)
- Show video badges in results
- Incorporate merchandising tags in results (Sale, free shipping, etc…)
Displaying the most relevant results early on in the result set is most often (but not always) ideal. In addition to the aforementioned guidelines related to content generation and attribute weighting, There are some out-of-the-box configurations that can significantly improve relevancy, in some cases:
- Leverage configurable products for variations… returns a single base product vs all color/size/style variants.
- Implement search by “fulltext” rather then by “like”… this enables magento to return a matching score based on multiple keywords vs. returning an unscored list of product matches.
Leveraging dictionaries, language algorithms and synonym mapping can assist customers with typos and general keyword issues. For example, it’s relatively straightforward to implement search maps that can interpret plural forms and common misspellings, which can significantly improve perceived search relevancy. Some best practices to consider include:
- Consider manually configuring synonyms for high volume searches.
- Implement (porter) stemming (gives->giv) in the index creation and call.
- Run a spell check against known dictionaries and propose suggestions when relevant.
- Generate a list of ignored words (i.e. “the” and “a”) and exclude them from the query.
- For multilingual customer bases, consider implementing a translation map and offer translated content as a preference.
In cases where there are many search results, consider implementing filtering in a manner that enables customers to refine the result set by attribute. For example, when searching for “shoes”, propose facets like color (“green”), style (“athletic”) and brand (“Nike”). Normally, this is configured in the content generation process, the only caveat being that out-of-the-box Magento does not support a multi-select faceting interface (i.e. both “green” and “athletic”), so consider an extension or development project to remove the interstitial pages.
When the occasion arises where it is beneficial to push technically irrelevant products into search results, for example, to inform users of a new line or to clear out aging inventory, it is often desirable to preform inline merchandising. Some examples include:
- Configuring custom landing pages for high volume searches, where appropriate.
- Including inline banners by query, above (and/or below) the result set.
- Attribute weighting by margin, sales initiatives.
On occasion, for example in industries that require a high degree of customer education, including multiple content types (besides just products) in the result set can be beneficial. For example, a “grout” query on a home improvement site could return a how-to article that then links to multiple relevant products. This is generally achievable with most technologies that leverage an independently generated search index. If appropriate, consider:
- displaying search results in a tabbed format ,by type, still emphasizing product results.
- including various content types, including cms, categories, products and blog posts
Because customer preferences are not always predictable and they indeed change, various search vendors have developed “learning” search solutions that use insight gleaned from user analytics when determining search relevancy. For example, if a product on the second page of a result set is consistently purchased against a given query, it makes sense to promote it based on user behavior to optimize conversions. Likewise, certain other markers of relevancy may be retrofitted into a product’s ranking, including:
- Search-to-click (i.e. how many times a product is clicked on for a given query)
- Search-to-cart (i.e. how many times a product is added to cart for a given query)
- Search-to-wishlist (i.e. how many times a product is added to wishlist for a given query)
Currently, some of the most exciting innovations in site search are responding to the user’s migration to mobile and, thereby, away from the keyboard. The most popular mobile operating systems allow users to dictate search queries to sites using spoken word, but what we’re finding is that, in these cases, some users are entering queries as questions and phrases rather than keywords… for example “find me a pair of inexpensive blue shoes” rather than just “blue shoes”. Given that over half of e-commerce traffic already originates from mobile devices, it behooves site administrators to consider incorporating a natural language processor (or service) into the query algorithm in an attempt to derive meaning from these types of queries,
In search, as well as in all other areas of Magento, optimizing for performance is a key metric in conversions While we will cover performance, in general, outside of the scope of this document, its important to consider speed and availability when implementing a search solution. Consider leveraging a search-optimized architecture (outside of MySQL) to return results, particular when tasked with managing large catalogs. Some examples include SOLR and bigtable.
A number of companies have engineered external search tools that integrate with magento that, in one form or another, propose solutions to many of the aforementioned challenges. Magento CE comes with a baseline MySQL search engine, which is good enough for smaller catalogs. EE ships with a SOLR option that enables attribute weighting and improves performance. The external vendors have generally focused on behavioral weighting and, to some extens, natural language processing. Some vendors include: