model skull with inner nerves showing
While search technology is showing signs of improvement, vendors still over promise when it comes to how much the technology alone can deliver PHOTO: jesse orrico

Whenever I start an intranet strategy project, it doesn't take long before someone raises the topic of delivering personalized content to employees. Fellow contributor and ClearBox Consulting founder Sam Marshall wrote an excellent summary of the benefits and challenges of personalization from an intranet perspective. These same benefits and challenges apply to search personalization. 

At the outset it seems easy. Just provide communications stories, policies and access to applications relevant to Region A to employees in Region A, as well as all the essential content from headquarters. But then inevitably a manager asks for Region B communications stories as they are working on a joint project with Region B and wanted something to chat over with their colleagues. Once the exception floodgates open, the requests will continue ad infinitum. 

With an intranet it is usually fairly easy to locate Region B stories by parsing the URL. If all else fails, then use the search application. But what happens when the search experience is also personalized?

Weak Signals Works for Google, but Enterprise Search?

Over the years Google and Bing have heavily invested in using weak signals to optimize the search experience. When I searched for [Portsmouth] recently, Google offered suggestions for Portsmouth, New Hampshire. The exception was [Portsmouth FC], which is the soccer club of Portsmouth, Hampshire, close to where I live. I clicked on [Portsmouth FC] and was presented with some useful links. When I queried [Portsmouth] again, none of the suggestions related to Portsmouth, New Hampshire. Impressive.

This leads me into the world of what Gartner calls "cognitive search" and Forrester calls "insight engines." According to the Cognitive Computing Consortium, 18 high-tech companies developed the definition of ‘cognitive search’ in mid-2014. Not a single search manager participated.   

Why not? The fundamental premise is sound: use a range of weak signals to improve the quality of the search experience. These signals could be office location, role, product range, documents opened in the last month and many more. These weak signals are especially difficult to define and incorporate in a global business where multiple languages, date formats and many other factors need to be normalized in the process of creating personalized results. 

In addition, employees perform multiple roles during the day: how will the search application know which role the employee is performing and deliver high precision results?

Precision Leaves Out Innovation and Serendipity

Very little of the promotional literature from the search vendor community reflects any understanding of what it's like inside a major corporation working across multiple countries and product lines. KMWorld published an article discussing the claims of these vendors without a single quote from a customer. 

When you look at a plot of word query frequency for an enterprise, what immediately strikes you is the precipitous fall-off after the first 100 or so terms. It is common to find that the most common query terms are people looking for applications. When speaking with a search vendor, show them this chart and ask what signals they would use to improve result relevancy for these top 100 terms, and then the next 100 terms.

Vendor PR often limits its focus to the "high precision" results of cognitive search. For certain queries this is important, but well-curated promoted content often can produce the same results. My concern about this focus on precision is that it does not support the equally important role of explorative search, including the requirement to be able to find information to learn about a subject. 

In these cases, the employee may not have any idea what terms to use. In the classic model by UCLA Professor VI Emeritus Marcia Bates, people search to find information and then use that information to optimize the next stage of the discovery process. If they don’t know what they are looking for until they see it, how can cognitive technology support serendipity and innovation?

(Read "Researching Serendipity in Digital Information Environments" and "Reimagine Enterprise Search to Unlock Deeper Information Discovery" to understand the value of exploration in search.)

Cognitive Search Resource Deficit

The single major cause of poor quality search is not the incumbent technology but the lack of skilled information professionals. 

Historically, search vendors have been reluctant to identify the team skills needed to optimize the performance of their software in case potential customers take fright at the implications. This is certainly the case with cognitive search. One vendor claims to be able to incorporate 100 different weak signals, a seemingly significant benefit until you ask how these affect content ranking and who decides who sees what in their searches. Does role carry more weight than location? If someone submits a query in German should the search engine assume they are located in Germany? 

Companies need the ability to balance the benefits of such tools against the total costs of migration and implementation. As it is, the continual push of the benefits of technology without quantifying, or better yet illustrating, the resource commitment required is helping no one.