Search Done Right

What makes enterprise search challenging (but also very rewarding) is its open-ended nature: the behavior of a search engine cannot be easily specified. Predicting and specifying the behavior of a search application for every possible query is not humanly possible. Unlike a document management system or a database, the performance of a search engine is hard to measure and quantify. Speed and coverage are important factors, but the primary function of the application highly depends on qualitative and subjective factors.

This open-ended nature comes along three dimensions: what, when and how.

1. What Data is Searched

What would seem easy to specify in theory is often in practice one of the biggest challenges. One would expect enterprises to know the location of the “good” information that they are producing, but in my experience this is the exception rather than the rule.

2. When Data is Presented to the User

This is the old “relevance” question. A lot of attempts have been made to quantify this factor, the TREC Conference being the most well known. But most relevance models rely on very controlled environments and fall short of representing the complexity of the enterprise environment.

3. How Data is Presented

This aspect of search is often ignored. Search results do not have to look alike in a long monotonous list. Even in a list, what is presented for each search result is critical, from the titles and the snippets (most of the time extracted using heuristics to give the most insightful snapshot of the document) to the overall organization of the search results (clusters, spotlights, meta-data navigation.) Ultimately, search results can be the information and not just link to it. Presentation possibilities then become endless, as open as the information itself. Here is an example of search results as information from querying “recommendations” within the 9/11 Commission Report:

Here is another example of search results as information from querying “Paxil side effects” on a medical information site:

This open-ended nature makes it hard for many organizations to understand, tackle, and deploy search. That in turns explains the current state of the enterprise search market – it is highly fragmented and confusing. Customers don’t know what to ask, what features are needed and which vendors they should look at.

As a result, enterprise search often ends up lower on the CIO priority list than it should be. There is also the “illusion” of commoditization. “Let’s get search, let’s get Google”, is often heard from those who don’t realize that Google, as in Google.com, is hundreds of smart PhDs optimizing a search application for one specific environment: the Web, an environment in many ways more uniform than the enterprise. And the unwillingness to tackle the unpredictability of the search process translates into its reduction to a basic function: “enter keywords, return list of results”, ignoring its most critical aspect: how useful or relevant the information returned is, should or could be to the human user?

This unpredictability should not discourage IT managers and buyers from controlling search deployments. On the contrary, it should trigger new levels of craftsmanship in project management. The “what” and “how” highlighted above can be specified to a decent extent by undergoing a careful examination of the data space and expert crafting of specifications (going beyond a list of industry buzzwords or vendor-touted features that may or may not have addressed your true needs.)

The “when” and “how” can be evaluated by adapting methodologies commonly used in science to a specific enterprise environment. Known and blind use cases can be created to evaluate vendors. The dynamic of deployment also needs to be revised. Search is not just a piece of software that is configured in the bowels of IT, deployed when ready, used for a time and then phased out. It should be seen as a long term living application, deployed quickly and improved in phases based upon end-user feedback. There is no limit as to how good and useful a search can be, but modest goals, early rewards and especially, valuable user feedback, can be obtained through quick deployment.

I strongly believe that search is opening a new era of open-ended enterprise applications, finally making plausible the promises of “Intelligent Applications” never fulfilled by the “expert systems” created twenty years ago. Its unpredictability and open-ended nature are what makes search extremely valuable. You don’t only get what you put in it. You get indefinitely more. This is what forward progress and productivity increases are all about.

This entry was posted on Friday, February 23rd, 2007 at 3:20 pm and is filed under Uncategorized. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.