Every Internet user has experienced the frustration of information overload. You type a few key words into a search engine and get hundreds of results, most of which aren't relevant. Chances are, you'll try to wade through a page or two, but odds are you won't go past the second page. It was that kind of inefficiency that Jim Smith, intranet Webmaster for the Controls Group of Johnson Controls was trying to avoid.
Johnson Controls is a worldwide supplier of automotive systems and facility management and control services. More than 33,000 people are employed in the company's Controls Group which is headquartered in Milwaukee. All of these individuals are connected to that division's corporate intranet repository via Advisor, a homegrown portal application that actually predates most existing portal products.
While developing Advisor in 1996, Jim Smith, the system's manager, chose a Verity (Sunnyvale, CA) search engine to complement the content management tools his team was creating. He wanted employees to be able to access and retrieve many different types of information from the central repository easily and quickly. While effective, simple text search had a number of drawbacks as the volume of information grew. Users were getting a lot of results that didn't apply to them. Additionally, the "documents" range from Word files to PowerPoint presentations and even complete programs, which can't be easily searched.
XML Expands Search Capabilities
In 1999, Smith prompted some modifications to the system. "The content management tools were already in place, I just had to apply them the right way," says Smith. So search logs were reviewed to see how people search for information and how the search engine responded. Instead of a structured database, Smith began to use a more flexible XML (extensible markup language) system. Because XML applies metadata (descriptions of data) to content, it allows for more relevant searches by attaching additional text that can be searched.
Using Verity's smart categorization functionality, Smith defined various content types. For example, there may be thousands of PDF (portable document format) files on the server, but only a fraction of those are brochures. A vocabulary of 15 discrete document categories were defined, and users who submit documents must select at least one of those tags in addition to other required metafields for comments or audience relevance. To further narrow searches, Smith defined four "roles" or audiences: sales, service/support, engineering/installation, and general users. In addition, documents are classified as either employee support, product or service information for internal use, or business/customer-related such as contracts.
Filtering Simplifies Searching Process
Using the updated Advisor system, users are able to search while the Advisor passively filters out irrelevant information. For example, when a salesperson signs on, he has a view that is customized for salespeople, including a home page that includes specific events like promotions or policies that affect only that group. The salesperson then sees only three choices: business, products, and employees. If the salesperson clicks products, he'll see Web sites and links that focus on product information that relates only to sales. When a document type, such as a form, is selected from the pull-down menu, the user will immediately find product-related forms used by salespeople. For even narrower results, a simple text query can be included with selections.
The challenge, says Smith, was to avoid becoming too detailed in creating the categories. For instance, the Controls Division has nearly a dozen different sales groups that are always changing. Trying to identify and customize for each one would be a time-consuming task and could create an "analysis paralysis" that prevents anything from being accomplished. Simply filtering out the nonrelevant content increases search quality exponentially, claims Smith, and the categories are fluid enough that they could be modified if any one got overloaded.
Smith also employs the Verity search engine in tracking the aging of documents. Using the search tools, monthly reports of files that are a year old are generated. The author of a document is notified that the file must be reviewed within 30 days or it will be "retired," which means it will remain on the server, but not recognized by the search engine. After 60 days of retirement, the document is "killed." By removing obsolete documents, the retirement process further promotes relevant results.
Unofficial early survey reviews conducted by the company indicate that 80% of the users reported that they were able to find information more quickly and felt the improvement met their needs. Employees who post content to Advisor needed some instruction in filling out the metadata form to fully understand its value.
Overall, Smith feels the implementation has been a great success and believes that many people are underutilizing existing software by not exploring all its features. "We did all this ourselves without any additional third-party tools for content management, just a few simple programs in Java."
Questions about this article? E-mail the author at EdH@corrypub.com.