Twelve ways in which enterprise search is different and difficult

These are the first twelve that came to mind – there are many more! This is why you need a good search support team, information management and search management strategies, extensible technology from a vendor you can talk to and a range of training and mentoring schemes.  And maybe some assistance from me on the lessons I have learned over the last twenty years of making search work?

  1. There can be massive amounts of structured and unstructured content (no one will know the true figure!) very little of which is curated for quality and findability on a consistent basis across the organisation because there is no incentive to make it findable. There might be metadata but it is inconsistently applied and so  could be more misleading than no metadata at all.
  2. Organisations have relatively narrow subject interests so there will be a substantial number of related documents which may all have very similar relevance scores that even high-quality snippets may fail to disambiguate.
  3. Content can be in multiple, and often mixed, languages, containing a substantial number of acronyms, alphanumeric (e.g. product) designations, technical terminology and idiosyncratic ways of referring to applications, departments, projects, products and customers.
  4. The ability to track down other employees is always a very important role of search (possibly 60% of all searches) and yet the employee names can be shortened, mis-transliterated or belong to employees who have left the organisation.
  5. The information is stored in multiple applications in text, data, image and video file formats, with PDF files presenting a particular challenge in text extraction. Many of these applications have an embedded search application.
  6. Content is security trimmed and there is no way to know whether information you are seeking has deliberately (for what reason and by who?) or inadvertently been removed from your access permissions.
  7. Recall is just as important as precision – the organisation’s entire information history needs to be at the disposal of all employees, subject of course to protective marking policies.
  8. There are multiple points of failure at collection, index, query and ranking level, none of which will be apparent to, or fixable by, the search user.
  9. Because of the rate of change in an organisation personalization based on past search histories and/or assumptions about the roles and interests of a user can be misleading, especially as many employees have multiple roles.
  10. Personalisation and security policies play havoc with analytics as it is not apparent from the query logs what information was delivered to the user, especially if it is not clicked so there is no audit trail.
  11. Employees may be searching on behalf of one of the many teams they are members of so their own search history, ‘recent documents’ and current role will be poor guides to search intent.
  12. Failure to find is not an option as it puts the organisation, its customers and the careers of its employees at risk.

These peculiarities are why enterprise search is a good example of a wicked problem

Martin White