The Search is On
Optimizing Search for the Customer Service Environment

When it comes to customer service, search is fundamental. Contrary to common perception, the biggest reason search fails is not ineffective search engine technology. The real problem is that we tend to ask the wrong questions—if we ask a question at all. Both call center agents and self-service customers suffer from the following two common search "worst practices":

Second-guessing the answer. At times, someone searching a knowledgebase already knows the answer or is reasonably certain of the correct solution. But this is rare. Most often, and particularly in self-service, the user does not know what the correct solution will be. Writing content from an "answer-centric" point of view and optimizing the search engine to find and retrieve answers is a recipe for failure. Content needs to be written from a "problem-centric" point of view—clearly stating the problem, issue or question that the content addresses, and encouraging users to enter question-oriented queries or what they know about the problem.

Searching for topics instead of key issues. This second worst practice is more subtle, but can result in responses that are significantly off-target. Executing searches for topics, or "keywords," usually results in a flood of vaguely relevant content. The problem is a lack of specificity on the part of the searcher. The result is predictable but no less frustrating—the search engine returns all the content available about the topics or keywords queried.

The solution is to write content with key issues specifically spelled out and in the language of users. Users should be encouraged to elaborate on the issue. Even one or two extra terms—"my voicemail PIN is being rejected"— can turn a vague keyword query such as "voicemail" into a search that will return highly relevant content.

Differentiating Structure in Content
Given that maintenance efforts and costs increase directly with the amount of authoring structure, finding the best balance is no trivial task. At a minimum, there should be enough structure to set aside a designated field for the key issue, problem or question in each article along with the ability to tag the content with more general topics.

Most search engines can specify different priorities, or "weights," for different fields or content zones. This is a useful way to increase ranking when the search query finds a match within the zone set aside for key issues. Take care when authoring and maintaining content to use this field or zone exclusively for the one or more key issues, questions or problem statements that the article addresses.

Now we get to the topic of "topics" or content tags. Tags are inherently limited, as a tag says only that the article is "about" that topic and not to what extent the topic is a key issue. For this reason, topic tags are only marginally useful in searching. However, they can be extremely useful for filtering out irrelevant content. Consider situations where the same issue can be encountered in a number of products, yet the resolution is different for each product. A good initial search will find all relevant content. A simple mechanism to filter those results based on the available tags present in the result set provides a second step to aid the user in finding the most relevant solution.

Even if the article we need is ranked at the very top of the search results, it might as well not exist if the user cannot recognize its relevance. Every search engine has some way of "selling" the user on why the search engine "thinks" a particular article is relevant. The best way to communicate relevancy is to go one step further and indicate whether the match occurred inside the "key issue" zone or was simply mentioned in the general body of the article.

Testing Search
Typically, testing starts with awareness of the content and selection of a particular item that should be retrieved by a given query known to match that item. Then a variety of searches are executed that more or less specifically match the item’s title or some text in it. If an item is ranked with a high score when the search is a close match, and is then ranked lower as the queries match less exactly, the search is considered to be working properly.

Then the system, its content, and search engine are rolled out to the support users where the complaints begin: "I can’t find anything relevant to my question." "There are too many search results."

Testing must be done with queries that express the issue a user is reporting so that testing does not just become an exercise in our "worst practices." Include only the symptoms users could know about or questions that could arise from their attempts to use a product or service. Use the terms and language of users, not your knowledge experts. When possible, use queries taken from logs of Web self-service tools or verbatim customer requests from call logs. Rather than trying to prove that particular articles can be retrieved with the right query, evaluate the search behavior using real-life inputs.

Search is an important tool, but even installing the best search engine will likely fail the needs of customer support if its implementation is based on something other than a set of best practices as discussed here. By implementing search that reflects the way your users think and talk, you will be able to achieve the kind of success experienced by KANA customers worldwide.