Customized Web Interactions -- The Next Piece of the Puzzle

Fastwater Rapids vol. 1.3, 6Sep98

As the web continues developing, it has changed from being a place focused on information distribution to one focused on interactions. Companies are now using the web as a place to engage with their customers, partners, and the industry at large. As this happens, a wider range of business goals is brought to bear on the web. In the first issues of Rapids, we've examined the issues around using the web for interaction and relationship building. We've taken a look at what it means to collect information on customers via the web and connect that to other information systems. And we've started analyzing how you can track the interactions on your website and understand what this data means. The next step is to close the loop and use this understanding to help your site fulfill your business goals. Oftentimes, this means building a site that adapts to visitors, is easy for them to navigate, and supports rich interactions between your organization and those visitors.

Not long ago, building any kind of dynamic website involved significant custom software development. However, now these projects have become much more practical. Vendors offer development platforms including web application servers and content management systems. And the tools for customizing and extending systems have improved, now including both client side and server side technologies. In the near future, we expect to see greater use of XML and its related stylesheet (XSL) and processing (DOM) standards. Making the right choices about these technologies is important -- comparable to choosing a database vendor. However, these are fundamentally IT issues: to be decided on considerations such as scalability, implementation difficulty, ease of deployment and so forth.

In addition to those choices, you also need to solve the hard problem of deciding what type of interactions you want to support on the website and how it should respond to different visitors. These decisions involve a broad range of business and marketing issues and extend to disciplines beyond IT.

In this article, we look at the steps you need to take to support guided interactions driven by a visitor profile. Using this approach, you distinguish between different types of users and direct them towards particular content, interactions, and presumably outcomes. There are other approaches to building dynamic and adaptive websites and we'll discuss those in future issues. For example, you can build a website that is a frontend to a database application. A reservations section of an online travel agency illustrates such a site: it's adaptive, but responds to travel requests that the visitor makes, rather than attempting to guide the visitor towards particular travel choices. Or, you can use collaborative filtering tools such as those offered by Net Perceptions or Firefly Networks. These tools are used to build sites that recommend products such as books or music to a visitor based on the choices of other visitors with similar behavior.

Each of these various techniques for supporting rich interactions with your website visitors is appropriate for different purposes. After we look at the details of how you can direct visitors based on their profiles, we consider where use of this technique is appropriate.

Guided Interaction based on Visitor Profile

The approach we examine here is one of:

understanding the interactions you want to support and the desired outcomes of those,
gathering information about visitors to your website,
selecting content to present to those visitors in support of your desired interactions.

Some examples of the interactions you can support this way include:

Delivering custom content. As part of an effort to attract visitors to your site, you can provide a custom news service. Visitors indicate areas of interest and you then present news articles in those areas.
Implement various selling techniques. For example, you can offer a user a discount if s/he looks at a product three times without buying. Or you can present content intended to cross-sell or up-sell.
Provide self-service support. By directing a visitor towards the support pages for products owned by that user, you can help that user efficiently find product support and thus reduce support costs and increase customer satisfaction.

There are a number of companies offering tools that can be used to build these kind of websites. You can do most of the work yourself, starting with a web server and a backend database and implementing the interaction logic on top of those. Another way is to use a content management system that supports dynamic content delivery. In preparation for this article, we spoke with representatives from two of the leading vendors offering such systems: Marty Connell, a product manager for BroadVision, and Erik Josowitz, VP of Product Marketing for Vignette Corp.

There are three main steps to making use of one of these systems:

Content Categorization. The content that can potentially be delivered must be organized into a set of categories. These categories are then referenced in order to select appropriate content for a particular visitor.
Gathering Visitor Profiles. In order to direct content to specific visitors, you need to learn about those visitors. This learning is stored in a visitor profile.
Create Associations between visitor profiles and content. These associations define the rules used to select content for custom delivery.

Content Categorization

Content is selected for display to a particular visitor using rules that work on the visitor's profile and the set of possible content. As we'll see, there are varied approaches to defining those rules. But in general, the rules work a lot better if they don't refer to specific pieces of content but instead provide criteria for selecting the content.

For example, a common strategy to draw visitors to a site is to provide customized news to those visitors. Connell of BroadVision showed us a demonstration of such a site and several of our Network Economy Practice Reports describe sites using this strategy. At these site, news content is developed or obtained via syndication. The content is organized by topic, e.g. press releases for various industries. A user who indicates interest in particular industries can then be shown appropriate new press releases. If the rules for selecting the press releases directly referenced the releases, those rules would need to be revised every time a new release was made. Instead, by referencing the topics for the releases, the rules can select new releases as they are made available.

Content categorization consists of creating a taxonomy of the possible content topics and then tagging each individual piece of content to indicate the category or categories it belongs in. This gives the rules a way to select content without referring to specific pieces of content.

Today, you'll most likely have to create the taxonomy and do the tagging by hand. Any particular piece of content may be belong to several different categories and human judgment is needed to decide which of those categories are appropriate. But, in the future, it may possible to automate this using indexing and abstracting tools coming from the knowledge management community.

Gathering Visitor Profiles

In order to respond to a particular visitor, you need to create a profile for that visitor -- a place to store the information you gather about the visitor. There are basically three ways to gather profile data: self description, observation, and integration with other sources of customer data.

Self Description

This is the simplest way to gather visitor data and requires the least integration work with other systems outside the website. Visitors are prompted to fill out forms describing themselves demographically and psychographically. You then store their responses for later use.

However, beware of turning these forms into a barrier for visitors: requiring visitors to answer lengthy questionnaires and asking for information before the visitor has reason to trust you. A better approach, described to us by Tim Chandler for Bay Networks during research for an upcoming Network Economy Practices Report, is to incrementally gather this information. Start with a relatively small number of questions that aren't too personal. Then, as the visitor comes to trust you and your site, ask for more information. Chandler described the approach as " Ask a customer for information only when you're ready to use that information." By asking for information "just in time", you incrementally build a relationship with your customer and you enable the customer to understand why you're asking for personal details. This makes it much more likely that the customer will provide those details.

Observation

Observation comes in two flavors: long-term statistical analysis of web usage and activity, and on-the-fly observation of an individual visitor. Analysis of overall web usage can be used to drive new web designs and to ensure that users in general are seeing the content that you want them to see. Observation and tracking of a single visit and a single visitor can be used to direct content to that visitor. For example, you may want to show a visitor who has seen several pages on one product information about a related product. Vignette provides access to this information by keeping track of the content categories that have been displayed to a particular visitor. BroadVision provides similar access via their "moment-to-moment" data gathering.

Other Sources of Customer Data

In many cases, the website isn't the only way your customers interact with you. For example, if the customer has previously bought products from you, that information is stored in some other backend database. Knowing what products the customer has already purchased could be very useful in determining what content to display next. As Adina discussed previously in Rapids, integrating with these backend databases is very challenging. To date, we find that many sites are delaying this integration work and focusing instead on explicitly entered and observationally gathered visitor profile information. However, we expect to see these other sources of customer data assuming greater importance over the next year.

Creating Associations

Categorizing your content and gathering visitor profile data aren't much use in themselves. The important step is creating associations between the visitor profiles and your content. These associations can then be used as rules to select content for a particular visitor. For example, the profile for a visitor may indicate interest in a particular product. This could be the result of the visitor explicitly telling you of his/her interest. You might have observed that the visitor repeatedly viewed descriptions of the product. Or you may have integrated with your backend sales database and thus know that the user has already purchased this product. By creating an association between visitors with interest in one product and the content for a related product, you can then show the visitor promotional literature for the related product.

So, creating these associations is key. This is exactly what successful sales and marketing people have done in the past: understand their customers and decide what to show them when. The trick is figuring out how to:

Do this on the web
And do it at the scale required by the web.

BroadVision's approach is to provide tools that support the existing process that marketers use. BroadVision lets business people define rules governing the content displayed to particular visitors. These rules specify a set of criteria that a visitor must meet (e.g. has looked at product page three times without buying) and an action that determines the content to display (e.g. show the discount offer). There's no magic involved here. This is what marketers already need to do: understand their audiences and choose appropriate messages for each audience. To date, this has been too time consuming and difficult to apply to customized web interactions. The focus for BroadVision's tools is making this easy enough that it can be done on the web.

Vignette takes a different approach, aiming the rules at the taxonomy of content categories. Rather than creating audience segments using visitor criteria and then choosing content to delivery to the segments, Vignette's tools focus on associations between content categories. As a visitor uses the site and is exposed to content from various categories, the system records the number of times content from each category was seen. Marketers then draw associations between the categories recorded in the visitor profile and other content categories. Josowitz described a hypothetical online grocer. When a visitor shows interest in beer, the grocer may choose to guide the visitor to pages featuring pretzels. The Vignette system makes it easy to make such associations between content categories and drive custom delivery based on these associations. Thus the task with Vignette is to correctly define the content taxonomy and then create associations across that taxonomy.

Challenges

Any implementation of dynamic web content delivery based on visitor profiles will face a similar set of challenges, regardless of whether you use a system from Vignette, BroadVision, or a different technology base. Among these challenges are:

Maintenance of the content taxonomy. The categorization of content is central to any of these systems. As your base of content grows, you will inevitably need to revisit the set of categories. And as you change the set of categories, you may need to recategorize existing content.
Audience Segmentation. These systems offer no magic pill. The hard work of defining audience segments and deciding what messages to deliver to those segments remains.
Testing. Building and defining a dynamic and customized website is considerably more difficult than building a simple static website or even a website that functions as a frontend to a database application. You will need to be careful not to underestimate this effort.
Performance. Until recently, dynamic customized websites simply weren't practical because of performance issues. New generations of hardware and software have helped to address this, but you will still want to carefully monitor your site performance. And you may want to use a mix of static and dynamic pages, rather than building a completely dynamic and custom site.

Another piece of the puzzle

It's clear that a significant investment is required to manage your content, build up visitor profiles, and create the profile-to-content associations necessary in order to provide customized content and direct visitor interaction. When is this investment worthwhile?

Here are some guidelines for deciding when to apply these tools and techniques:

Meeting visitor information needs. Website communication is sometimes used for persuasion and sometimes for providing information. Persuasion is hard and these techniques are not a panacea. Providing needed information is more straightforward and easier to support. For example, compare informing a reseller about new product developments to implementing a cross-sell strategy. Tracking the product lines sold by the reseller and then keeping the reseller informed about those product lines is a natural and effective use of these tools. Correctly judging when a browsing visitor is receptive to a cross-sell pitch and then targeting that pitch appropriately for the visitor is much more difficult.
Practical to gather visitor profile data. In some instances, you could provide very effective and customized content if you could access certain information about your visitors, but that information is just too hard to get. In these cases, these techniques aren't practical. For example, consider American Airlines's website, one of the highest profile sites built using BroadVision's system. This site is integrated with American Airlines's backend frequent flyer databases and can thus deliver custom information both about frequent flyer accounts and about flight specials of possible interest to visitors. Asking users to re-enter flight data would be entirely unworkable. Thus, for American Airlines to provide this kind of custom interaction, they needed to make the investment to integrate with their backend databases. Lacking this integration investment, the system would not be practical.
Measurable results and outcomes. Implementing a site that uses these techniques requires substantial investment of capital and effort. You should be sure that you're able to define and verify the results you expect from such a system. Don't assume that a customized web experience is, of course, superior and worth the investment. For example, providing up-to-date product information to increase reseller effectiveness is likely to be a worthwhile investment. And one whose results you can measure. Implementing a complicated cross-selling and up-selling strategy based on a visitor's browsing behavior may work well in some industries, such as catalog clothing sales, but won't be effective in other industries.

We've been looking at the pieces of the puzzle that, when put together, will let you build a complete system to support interactions on the web. Today, any complete system will involve a combination of manual work and a variety of tools. One of the areas you'll need to address is how to guide visitor interactions on your website, and a rules driven approach based on visitor profiles is often appropriate.

Over the next weeks and months, we will continue our coverage of the emerging tools, and more importantly of the issues and problems you need to consider as you make plans to take full advantage of the web.