How do you solve a problem like enterprise search?


The final KMLF meeting of 2014 was a panel discussion about enterprise search. Brett Matson, Managing Director of enterprise search provider Funnelback, revealed some interesting insights that I wanted to capture in this interview. Brett used to work in the search industry as an engineer for the CSIRO, before starting Funnelback in 2006, so he knows a bunch about search and how people use it.

1. Where Google has succeeded, the enterprise still struggles. According to APQC research, only 53% of firms rate their enterprise search as effective or very effective. What do you think is happening with the rest?

The 53% in the APQC research relates to federated search, which is the idea of having a search engine query third-party search engines and combine the results. This is generally ineffective because it’s difficult to rank heterogeneous results against a common baseline, and you also can’t use tools such as faceted navigation.

However, there is a general dissatisfaction with enterprise search technology and it’s these kinds of statistics that have led Funnelback to adopt a new approach. We believe the top three causes of dissatisfaction are:

  • Lack of Ongoing Investment in Enterprise Search
    Organisations too often consider enterprise search to be a fixed-duration project, rather than an ongoing investment. Users change, needs change, and information changes. Search engines can cope with this to an extent, but the can’t offer the same effectiveness as a human assessing business needs. Organisations need training and guidance on where the quick wins are in ongoing investment in optimising enterprise search. Google only succeeds on the global Web because it has thousands of people doing this constantly.
  • Re-inventing the Wheel
    Enterprise search budgets are often consumed in developing complex search interfaces and applications from scratch, instead of focusing on configuring off-the-shelf solutions to fit business needs. An enterprise search product should have pre-assembled, best-practice templates for a range of different search requirements (e.g. eCommerce search vs. events search vs. courses search) to allow implementation effort to focus on the business aspects of the solution.
  • Only Solving Part of the Problem
    Most enterprise search products attempt to rank information with respect to its relevance to a text query, but ignore other significant factors that detract from search effectiveness.

A holistic enterprise search solution should include:
Bird’s-eye view metrics of all content, showing where it’s stored (e.g. web vs. enterprise vs. social media), how much exists in each repository, how old it is, missing metadata, poor quality titles, duplication, accessibility metrics, and the link graph. This provides information managers with a means to prioritise organisational investment in managing information, and thereby enhancing search effectiveness.
Intelligent guidance on how to make content more visible/findable. Search engines generally attempt to hide the internals of their ranking systems and this makes it difficult for customers to learn how to make content more findable. An enterprise search engine should use its internal ranking knowledge to show content authors why pages rank the way they do and provide guidance on how to increase each page’s findability.
The ability to surface and promote content based on user context with simple rules such as “User is in Department A”, “User is located in New Zealand”, “User is in the finance industry”, “User works for LexisNexis”. These rules can then be overlaid to form more sophisticated rules, without the need to create rules for every distinct possibility. Funnelback goes even further by allowing these rules to be applied to anonymous users by looking up their IP address in an internal database and inferring information based on the organisation that owns the IP address.

2. Do you think an enterprise taxonomy is important?

We don’t often come across organisations that have enterprise taxonomies. If done well, they can provide benefit to many aspects of knowledge management, including making better use of technologies such as text analytics. However, it’s a significant investment that can fail to pay off if implemented ineffectively. The two main problems are:

The taxonomy is not aligned with how the organisation uses information. This can be a result of the taxonomy classifications not reflecting the way people attempt to find information. It can also be due to an onerous system for classifying content.
Information systems, such as enterprise search tools, being unable to leverage the taxonomy effectively.

Instead, we recommend organisations invest in a behavioural taxonomy. This includes assessing:

What are people commonly searching for? (i.e. the top search queries)
What information is being accessed? (i.e. which search results are being clicked?)
Which searches produce no search results?
Which searches cause people to click through to the second page of search results?
What feedback are people providing in response to a search query (i.e. using a feedback form on the search results page)

It’s this kind of information that informs smart decisions around:

What information needs to be created that doesn’t currently exist?
What search promotions (i.e. best bets) would save people time?
What hard-coded search auto-suggests would make people more productive by preventing them from needing to see a search results page?
What synonyms would help people find what they’re looking for?

Following the Pareto principle, if you can ensure that 20% of searches work effectively then 80% of the information needs are addressed. This small amount of effort pays large dividends.

3. After I’d read Weinberger’s book, I was convinced tagging and faceted search in the enterprise was in our future. So, I was surprised to hear you say at our KMLF meeting that most Funnelback users don’t tag documents. Why is that?

I suspect it’s a cultural issue rather than a technology one. The ability to tag content in enterprise search is powerful because it’s not limited to a single data store; users can tag content whether it’s in an EDRMS, file share, social media channel, intranet, or third-party website. Funnelback provided a means to do this, but users didn’t see any immediate value in tagging content, so chose not to.

In hindsight, we would have benefited from complementing the technology with a cultural program of educating users on the benefits of tagging and sharing, and optionally gamifying the experience to reward and incentivise.

On the other hand, in the last 10 years, faceted search has evolved from a technology used almost exclusively on eCommerce sites, to an out-of-the-box feature in every Funnelback deployment, including intranets, databases, and enterprise search. It’s popular because it’s intuitive and provides immediate value.

4. What are the first important steps for any large organisation embarking on a new search strategy?

The first question every organisation should ask is:

Who are the stakeholders affecting the success of our organisation and what information do they need to maximise our success?

At a more practical level, this includes questions like:

What are the personas in our organisation? (i.e. the archetypes that represent the different roles)
What information do they need in order to maximise productivity and make better decisions?
What are our customer personas?
What information do they need in order to maximise engagement and have a positive customer experience?

Without asking these questions, organisations sometimes assume that searching everything with a single query (access controls permitting) is the answer. Sometimes it is the answer, but it can be a more complicated and costly exercise than necessary. For example, do users want to use an enterprise search tool to search their own email, or would they prefer to use the search on their mail client?

5. When an organisation proceeds with implementing an enterprise search solution, what sort of management and maintenance resources, if any, should the organisation be prepared to commit?

At a minimum, search analytics should be checked monthly with the following questions in mind:
What information is needed that doesn’t currently exist?
What search promotions (i.e. best bets) would save people time?
What hard-coded auto-suggests would make people more productive by preventing them from needing to see a search results page?
What synonyms would help people find what they’re looking for?

We also recommend having an expert conduct an annual or bi-annual health check to assess all aspects of the search system, including crawl scope, ranking quality, search filters, etc., as well as ensuring the system is aligned with changing business needs.


— Thanks for your time, Brett!


Brett Matson has contributed to a scientific paper on document-level security and developed Funnelback’s Contextual Navigation system.


Post a Comment: