Say “search engine” and most people think of Internet search sites like Google and Yahoo. But search includes other tasks such as analyzing document archives, finding customers in a database and identifying products in a catalog.
All draw on similar statistical and language-based techniques to identify relationships among texts. The important differences are in the applications that make those techniques usable for a particular purpose.
Mercado CSN (Mercado Software, 888/376-1400, www.mercado.com) searches for products on an e-commerce site. It extends beyond returning search results to provide extensive control over what visitors see, based on rules defined by site merchandisers.
Mercado lacks major parts of e-commerce functionality, including page design, content management and order processing, so it is not intended as a complete e-commerce platform. Yet, by controlling much of what appears on the screen, it can have a major effect on site results.
The foundation of a Mercado installation is an index file with information about the items to be searched. Mercado can import item information or use its own UniClass technology to extract, standardize and classify item attributes.
Like other language-based text engines, UniClass relies on thesauri, taxonomies and semantic analysis to interpret the inputs. Mercado provides reference bases for different subject areas, such as automotive, consumer electronics and food. Users can edit or extend these to fit their particular data better.
The index file contains links to the content that will be displayed on the Web page. The content itself is stored and managed outside of Mercado. The index also can include attributes from sources other than product descriptions, such as inventory levels or profit margins from an accounting system. An index file is typically updated in a regular batch process.
Once the relevant information is available in an index, Mercado can use it in many ways. One is to extract lists of categories and present these for users to browse. Another is to accept search words and return a list of results based on exact matches, partial matches, synonyms, value ranges, related terms and other connections identified through linguistic and category-based methods. Results are ranked using a relevancy score based on several elements, and users can adjust the element weights.
The list of results is transformed by user-defined rules before anything appears on a Web page. Mercado divides each page into several zones and can display different information in each. For example, there can be zones for actual search results, related products, recommendations, best sellers, banner ads and relevant documents.
Each rule specifies which zone it applies to, when it is triggered and what actions to take. Triggers can be based on search inputs such as particular keywords, on the contents of the search results such as a particular category or attribute as well as on the nature of the results such as the number of items found.
Rules also can be limited to particular customer segments based on customer identifiers imported from the Web site along with the search input. Mercado also can randomly assign customers to segments for A/B tests. In addition, rules can be limited to a particular department within a Web site, letting merchandisers for different areas operate independently.
Users define groups of items to change through the same search interface that site visitors see, working in an image of the actual site. This helps ensure the rule generates the desired results while avoiding the need to train merchandisers on a special group-building interface. But these group definitions exist only within a given rule, so groups that apply in multiple cases would best be converted into product attributes.
For organizations where managers must approve changes before they take effect, Mercado can generate an e-mail to notify managers when a proposed rule is ready and let the managers simulate the rule’s results before putting it into production. This simulation also presents a view of the actual Web site, with an option to select which customer segment is being simulated. Rules can automatically activate and deactivate during specified date and time-of-day ranges.
Management approvals, department-level rule creation and automatic activation are critical to administrators at large, complex organizations. Mercado has additional technical features to support large sites such as the ability to consolidate searches across multiple data types and servers.
Deployment of Mercado takes a fair amount of technical effort, from three to five weeks at a small site to 20 or more for a major company. Some of the work is modifying the underlying site to generate Web services requests that send information about each search or browse query to Mercado. The vendor provides a sample application to help users do this.
But most of the time involves revamping the site to take advantage of Mercado’s capabilities. This requires adding the different types of page zones and populating these with the XML outputs that Mercado generates in response to each request.
Mercado also produces a range of reports. Most focus on visitor behavior, such as lists of the most common search terms, browse selections, search and browse sequences, frequently viewed items and query trends. Other reports aim to help administer the system and identify areas for improvement such as lists of requests that returned few or low-relevance results, most-used business rules and load statistics. Users also can write custom reports.
Versions of Mercado are available for large consumer retailers, smaller retailers and business-to-business marketers. The software can be purchased to run in-house under a traditional one-time license fee or can be hosted by Mercado. Pricing for the hosted version starts around $3,400 monthly and is based on query volume. Mercado was founded in 1997, and its software has been purchased by about 70 brands.
