Exploiting Your Website's "Find" Analytics

By Larry Harris

The analysis of your customers' product "find" experience can be your compass to give you the directions you need to improve the performance of your e-commerce website. Without it, you are shooting in the dark trying to make improvements. The problem is that, given the amount of data, tracking each shopper's search and navigation steps can be overwhelming, making it hard to mine the key nuggets of insight that can lead to steady improvement.

Much like search itself--in which the interesting cases are the two extremes: too many results and too few results--analyzing website data is about looking at the two extremes: what is working poorly and what is working well. We will explore how properly analyzing this data can enhance the shopping experience, improve important site metrics (such as conversion rates) and even be used to attract more traffic to the site from web search engines.

IMPROVING THE FIND EXPERIENCE
The find experience refers to all the ways in which a shopper can find the product or products they are looking for on your site. The two common methods are search and navigation. Search refers to the process of typing a description into a search box and having the search engine return a list of relevant products. Navigation refers to clicking on links to narrow the set of products to a more manageable subset. It includes both top-level navigation, which allows the shopper to traverse the taxonomy into which the products have been classified, and search refinement navigation, which refers to the navigation links provided after a search to allow the user to further refine the results set.

In many websites, these two methodologies work completely independently. Products can be found either by search or navigation, but not in combination. A search ignores your navigation context and a navigation click ignores your search context. What many websites have found is that the more tightly these two can be integrated, the better the find experience. Search refinement navigation is one expression of this integration that allows navigation within the search context. Similarly, you can allow search within the navigation context.

SEARCH ANALYTICS
Common search analytics show the most frequent searches or the most frequent individual words used in searches. The most useful data of this type comes when the words are clustered into conceptual categories. The clustering should take into account both stemming and synonymy. For example, the terms "Womens," "Women's" and "Ladies" should all be counted as the same term rather than as separate terms.

These clustered reports are always interesting, but they don't provide the guidance that is needed to improve the site. The most useful reports are those that show those words that are the most problematic. These are rarely the most frequently used words and they may be combined in searches with other words, which can make it hard to identify the words that are really causing the problems. You want a list of words that "failed" every time they were used in a search. In this case, failure is a search that resulted in no products found or would have resulted in no products found were it not for search relaxation. (Search relaxation is a technique that avoids a "no products found" message being shown to the shopper by ignoring a word in the search. For example, a search for "black beaded dress" may return "black dresses" rather than returning no products at all.)

Once these problematic words are identified, the behavior of the system can then be improved. This normally takes one of three possible forms: adding synonyms, defining a site search or an early warning.

Adding Synonyms: This is the most common reaction. For example, if the word "plus" is identified as a problematic word, looking at the actual searches would indicate that it is used as part of the phrase "plus size." By defining "plus size" as a synonym to an appropriate size range, the search behavior will be improved.

Site Search: Users will often enter searches like "size chart." The proper response to such searches is to redirect them to an appropriate page on the website. Unfortunately, many websites interpret this as a product search and try to return relevant products. Such searches will appear as problematic words and, once discovered, can easily be defined as site searches to yield the desired behavior.

Early Warning: Other problematic words are really searches for products that you don't have. For example, searches for "flag shirts" started appearing after 9/11 and were identified as problematic words on many sites. Agile companies were able respond to this early warning by ordering flag shirts and selling them over the website.

NAVIGATION ANALYTICS
The analysis of top-level navigation is primarily to understand where shoppers are getting lost. The problem with navigation as a find mechanism is that it presents all your products in a single hierarchy of how you (and perhaps not your customer) like to organize your products. For example, in searching for a screw, a shopper may not think to click on "fasteners," the category under which screws are classified. Once the shopper descends another branch in the hierarchy, it becomes a case of "you can't get there from here." Looking for points at which shoppers stop descending and back up the hierarchy can give insight as to where they got lost.

The analysis of search refinement navigation is primarily to understand which categories and product attributes they find the most useful in refining the result set. For example, a search for "sweaters" might return a number of search refinement navigation links, such as collar style, sleeve length, material or price range. Once many such product attributes have been defined, there is often contention for space on the screen. Analysis of how many times each attribute is clicked on can help determine which attributes should be shown and in what order they should be presented.

IMPROVING SITE METRICS
Conversion Rate: It has been shown in many websites that integrating search and navigation, and fine-tuning the search based on problematic word analysis leads to significantly higher conversion rates. Conversion rates are affected because shoppers don't buy what they can't find and any improvement will produce a corresponding lift in conversion. In addition, conversion can be affected by the order in which the products are presented. In general, presenting the more popular products that meet the search criteria first yields a higher conversion rate.

Average Order Size: Average order size is impacted more by cross sells and upsells than by the product find mechanism. Of course, if you don't make the sale of the first product, cross sells are irrelevant.

A/B Testing: In general, the only way to really know whether a change to the site impacts a metric, such as increasing conversion rates, is to employ an A/B testing methodology. Comparing the metric before and after making a change doesn't work because the month-to-month variation in the metric is too high. An A/B test is one in which the traffic to the site is simultaneously split into 2 groups: A and B. Group A is the status quo website and group B is the site with the change implemented. Since the test is run at the same time, seasonality is removed from the equation.

IMPROVING SITE TRAFFIC
Search Engine Marketing: Search analytics should provide you with your conversion rate on a word-by-word and a search-by-search basis. This will give you some insight into not only what words are worth buying, but also as to how much to pay in the bidding process for each word.

Search Engine Optimization: The web search engine crawlers are often poorly suited for crawling an e-commerce website. Experience has shown that just crawling the product pages yields poor results from a search engine optimization point of view. Once again, website analytics can be used to tell you what words and searches you are effective at converting and therefore what content should be exposed to the crawlers. By focusing on exposing a few things very well, it is often possible to get better results than simply trying to expose everything.

Larry Harris is vice president and general manager of Bedford, Mass.-based EasyAsk, a division of Progress Software Corp. He can be reached at (781) 280-4653, or via e-mail at larry@easyask.com.