RFP CHECKLIST: Enterprise search
- By S. Michael Gallagher
- May 18, 2006
Unlocking the information at your agency may require a robust enterprise search platform. But before seeking a solution, plan ahead. If your RFP
doesn't really describe what you need, valuable data could be left hanging in the breeze. Consider the following when you prepare an RFP:
- Specify how much data you'll need indexed. Low-end, departmental systems can typically handle tens of thousands of documents; midrange enterprise search starts at around 300,000 documents and heads into the millions.
- Specify the types of content you'll want indexed. Almost all search engines can handle intranet sites, most standard office automation document formats and Adobe Portable Document Format. Fewer can index images, video and audio content, or information lodged in enterprise applications.
- If you want search results to include information from the Web, e-mail, enterprise application data and reports, and records within databases, you'll need support for 'federated search.' Federation describes the passing of search queries to other applications and remote services, as well as the aggregation of results across all sources. Determine which data stores are supported by search engines 'out of the box,' and which require additional integration or some other workaround.
- Consider compliance with related application standards, such as Defense Department 5015.2 and NARA regulations for records management. If you're using search for records retrieval, you may need to specify that vendors support baseline functional requirements for federal records management programs.
- Determine how quickly users must get search results to maintain an acceptable quality of service. To do this, consider the types and volume of queries you'll be executing. Depending on the type of data you're indexing, a small amount may require high performance. Many e-commerce sites, for instance, have less than a gigabyte of data but need to serve up as many as 150,000 queries per second. Larger data stores, such as those at research libraries or intelligence/law enforcement agencies, may process fewer queries, but the computational resources for each could be very large, especially across multiple data repositories or where data analysis is involved.
- How will users navigate to their answers? While most people are used to Google's search box and pages of ranked results, you may want content organized taxonomically to help narrow searches before they begin, or have results clustered by related concepts to help find answers quicker. If the information is geographically based, you may want it overlaid on a map.
- Make vendors demonstrate search precision. Enterprise users (as opposed to general Google users) don't want hundreds of pages of results. They want exact answers to their queries. Ask search vendors how they tune their engines for better precision, and you can use search logs to improve how you index content (or how users write their queries).
- Don't overlook security. Ask what authentication and auditing controls are supported.