This is a model I have been using to discuss the concept of ‘what is enterprise search?’ with practitioners. The concept is that ‘Enterprise Search’ is supported by a series of dedicated User Interfaces (UI) shown as blue boxes, used by virtually every member of staff in the enterprise on the left, to communities of decreasing size with specific business requirements (and increasing context and complexity) as you move to the right.
At the extreme ends denoted by the purple boxes (Company Intranet Home page for the masses on the left, niche domain application on the right), Enterprise Search has no dedicated UI per se, but is embedded within these applications, so queries are made to the Enterprise Search index ‘in context from other applications’.
The green triangles between the search index and UI is the reasoning saw tooth, representing simple traditional textual Information Retrieval (IR) needs (e.g. I need the investment proposal for asset x) and more complex natural language Question Answer (Q&A) needs that combine numerical and textual information (e.g. Who scored the first goal at the last world cup and at what minute of the game?). Critically, simple IR and complex Q&A may support both Look up and Exploratory search, hence the saw tooth across this continuum from left to right. Predictive and prescriptive analytics connect to this continuum towards the right, as prediction (what is going to happen) and prescription (what should I do about it) are just particular forms of an exploratory question.
Although the model shows a clear distinction between a dedicated Enterprise Search UI (blue boxes) and non-dedicated UI’s (Purple boxes) the difference between an Enterprise Search UI, Intranet Portal/Home Page and a domain application can become somewhat blurred, especially when canned queries (of some form) are used to deliver content (of some form) straight from the search index. However, in general the purple boxes are used to denote applications that mostly display, access and allow manipulation of content outside the enterprise search index, but have (some form) of search function.
This may help reconceptualise the debate about Enterprise Search deployments. There is always likely to be a need for the ‘common to all’ company A-Z and ‘Google like’ page, especially for routine tasks e.g. looking up the IT Request site, booking training, claiming expenses or finding out the times of the shuttle bus from one office to another. More ad-hoc or unusual tasks such as knowing how to report phishing emails, or the travel policy to certain countries are also well served by these mechanisms, where there is typically a right answer.
However, this is unlikely to meet all the search needs of the communities within the enterprise. Firstly, is the information overload challenge. Some enterprises have tens if not hundreds of millions of items in their index likely to cause search precision issues and information overload for almost any search. The use of context (scopes) to trim the amount of information returned for particular communities through dedicated UI’s with hard coded scopes of what content in the index is to be searched, can increase search precision for communities without necessarily the need for any additional functionality, disambiguation logic or ranking algorithms. Secondly, exploratory searches (Marchionini 2006) which are open ended where there is not necessarily a ‘right answer’ require more functionality and techniques. Finding analogues for example may require components such as a spatial GIS map, table, colour coded matrix or graph using underlying knowledge representations and text analytics. Users gathering information (berrypicking) may wish to export metadata of all search results, closer to a ‘library’ report than search results page. Faceted search techniques enabling both high precision and serendipitous browsing (an Interactive Information Retrieval (IIR) technique) may be of more value to these communities regularly performing exploratory searches.
Just how many of these UI’s are needed is likely to be defined by the type of enterprise, design, size and ultimately the business requirements. Too many can be maintenance intensive, inefficient and confuse users; too few and key business needs may not be addressed. Building a network of business focal points which ‘know the enterprise’ is likely to be key to a successful deployment, as is looking for synergies across these businesses and communities. Designing for serendipity should be a principle in UI design.
Think of the Enterprise Search UI as outwardly ‘heterogeneous’ but under the bonnet is ‘homogeneously’ architected. To only cater for the high volume queries ‘what is common to all’ with a single user interface for Enterprise Search could be considered in search requirement terms, a “tyranny of the masses”.