Establishing requirements

1. A thorough audit of current and anticipated document types and volumes needs to be carried out. The rate of addition is important as the indexing of the documents needs to be carried out on a schedule that ensures the documents are available to be searched within an appropriate time.
   
   
2. Assessing the implications for server architecture, storage, application integration and network bandwidth is important, because the search will almost certainly be extended over time , and then IT scalability issues become critical. The most intensive server element is the indexing of the documents, but the ability to then serve up the documents rapidly as a user identifies them can put quite a load on servers and networks.
   
   
3. Determining the range of user requirements for search will assist in ensuring that the right information/metadata elements are presented in the search results page. Knowing how long a search engine took to do the search is one of the most useless bits of information ever presented on a desktop.
   

Implementation never stops

Ten critical success factors

1. Excellent project management
   From the outset, a broad-based project team and a senior-level sponsor are essential because of the range of knowledge and skills required to develop the initial specification and then to evaluate the products.
2. Knowing what there is to be searched
   The characteristics of the document collections that will be initially and potentially indexed and searched need to be audited to a high level of reliability. Just a few important documents in an unusual file format may remain undiscovered forever.
3. Rigorous testing as part of the product evaluation
   One outcome of the document audit should be a representative test collection of documents so that all the relevant functionalities of the search engine can be evaluated before any decision to purchase is made.
4. Understanding user requirements
   In any organization there will be a range of different search requirements from users, depending on the purpose for which they are searching and their previous experience of searching. These requirements are best identified though the development of search personas and scenarios.
5. Understanding the technology of search
   All search software products index documents, run queries against the index and then present the results of the search to the user. Understanding clearly how this is carried out is important in assessing the relative merits of each product.
6. Anticipating scalability and extensibility issues
   Most enterprise installations start small and then grow. The implications of scaling up the indexing process and extending the functionality (for example for multiple languages) need to be assessed at the outset and kept under continuous review.
7. Assessing integration issues
   The ease with which a search engine can be integrated into, or with, other applications (for example a database of clients) is not easy to determine in advance of the work being undertaken. A minor upgrade in either the search software and the other application(s) may give rise to a substantial loss of search performance.
8. Testing for usability
   Providing a high degree of usability of the search interface is essential, and should be based around the search personas and scenarios developed in the specification phase. A small change in the presentation of search results can have a significant impact on usability.
9. Meeting changing user requirements
   A search engine needs to be continuously monitored and tuned to ensure that search performance and changing user expectations and requirements are identified as quickly and unambiguously as possible, and feedback provided on the extent to which they can be met.
10. Appreciating that 'trust' is the best metric of search performance
    Raw computer power will, in itself, not meet user requirements for search. Effort has to go into developing and applying appropriate metadata, taxonomy and categorization schemes that enable the user to trust the search engine to either deliver the information being sought or confirm that no relevant information exists.