« Our Top 3 Google Search Appliance Tips | Main | Dummies Books Take On Search »

August 04, 2008

Enterprise Search Dead? Or just misunderstood?

Search experts including Steve Arnold in his Beyond Search blog and Leslie Owens of Forrester Research have written the obituary for 'enterprise search', that monolithic 'one size fits all, everywhere' scheme that leading search vendors have pushed for years.  Others, including Tony Byrne, publisher of CMS Watch, differ.

While there's no question that all of the major vendors were pushing this flawed model - along with the magic beans of single-shot relevancy, I don't think we've ever really seen a full-blown, 100%, all-encompassing 'enterprise search' implementation of the sort that Steve, Leslie, and the other analysts have pronounced dead.

What we see in successful implementations is what we'd have to call distributed search. Most companies have anywhere from 2 to 5 search solutions, typically from several vendors. The public web site may use a Google Search Appliance; the Intranet uses FAST and/or SharePoint; Corporate Legal uses Autonomy's Zantaz; and Customer Support may use InQuira. And of course each of the divisions and web sites throughout the company will have Endeca, DieselPoint, a few Lucene/Solr sites, and maybe an old Perl script written in 1998.

The real trick is to glue these technologies together not into a single giant searchable index, but to combine them together logically so the user does not need to know where to look for specific content.  We, like many others, call this Federated Search, and have written about it any number of times in our newsletter.

Once you look at a federated search solution there are a couple of options. Honestly, most companies use the first; but in reality the second is probably better in the long run.

The first scenario is to select a primary entry point and thus, a primary search technology. Most internal users start will start from a corporate intranet, so used the example above, let's assume this primary search engine is FAST/SharePoint. The primary FAST index has most of the common content indexed; but when a user enters a search, the search application calls out via federated search to the public web's Google Appliance, to Legal's Zantaz; to support's InQuira, and to all of the other authoritative sources. Each of these sources sends back one or two high-confidence results, displaying them as suggestions rather than as results.

For example, in a Financial Services company, a query on the intranet for "retirement plan" will have authoritative results from both the products area - the company markets retirement services, after all - and from the HR area, since its employees have retirement plans of their own. Rather than have these two perfectly valid results 'fight it out' for top positioning, why not suggest both options, depending on the users intent. Or provide dynamic facets navigation to let the user provide more context. And consider the user's context as well in order to make a 'best guess': is the user in sales, or in research? This ensures the user finds the best context without having to know exactly where to look and without having to read his or her mind.

The downside of this first type of approach is the user still needs to navigate manually to the 'primary' search site. A researcher who spends her day on the R&D portal would have to leave her primary portal to search for content corporate-wide. As Steve Krug says in his books, "Don't Make Me Think".

This leads to the second approach that we see very few companies actually using: each portal used by a large number of employees - or partners and customers for that matter - should do its own federation to other authoritative corporate sources, and display federated results just like the primary search did in our first scenario above. This insures that anyone searching on the corporate intranet will find relevant content, even if their primary portal is not the main corporate intranet.

These methodologies both provide the kind of enterprise-wide search that has eluded us for so long in the old model: search once, find everywhere.

Enterprise Search is Dead - Long Live Enterprise Search.


TrackBack URL for this entry:

Listed below are links to weblogs that reference Enterprise Search Dead? Or just misunderstood?:


Interesting points, and I don't disagree.

That being said, I've never been a fan of any federated solution, whether it is Dogpile or Verity Federator or the federated web part in MOSS - they all seem janky to me.

The first approach you suggest, with the primary search engine kind of smells like sidebar ads. What most users say they want is google.com in the enterprise. Period. For people like us who know that pagerank is broken in the enterprise, this is an opportunity.

The "one shot relevance" is what pagerank does for the web. I interpret your post as saying that "one shot relevance" is dead. You may be right, but it smacks of defeat. I think one shot relevance still has some life in it, but it is a fundamentally hard nut to crack. Once cracked, it has the potential ooze goodness all over our portals.

The comments to this entry are closed.