Consider the numbers: According to the Harvard Business Review, it can cost six to ten times as much in marketing and advertising costs to acquire a new customer as to retain an old one. And for e-businesses, those customers don't amount to much: In March 2000, The Industry Standard reported that first quarter revenue earned by online retailers from each new customer averaged a mere $24.50.
In the following quarter, however, those same customers brought in an average of $52.50 each, and that figure remained consistent through each successive quarter. Clearly, maintaining long-term relationships with customers is the way for businesses to start seeing real returns.
Unfortunately, the same study revealed that less than 40 percent of e-commerce customers were repeat purchasers. Online businesses are now faced with what is, for many, a brand new challenge: How to better serve the needs of customers so that businesses can establish lasting relationships?
The Personal Touch
Personalization and customization software was developed with customer retention in mind, and it's a solution that increasingly more Web businesses have begun to embrace. According to research conducted by eMarketer, as of June 2001, 56 percent of U.S. e-businesses had incorporated some level of personalization.
Perhaps the most obvious candidates for these technologies are B2C e-commerce sites, where personalization makes true one-to-one marketing possible. By analyzing and addressing each customer's specific needs, retailers like Amazon.com can more accurately target smaller market segments for promotions and product cross-selling.
Beyond retailing, however, a wide variety of sites benefit from personalization. For example, B2B e-commerce sites can use personalization to deliver customized news, tools, or advisories tailored to the specific relationship established with a business partner.
Content-driven sites like Slashdot.org use personalization to encourage repeat visits by letting users select the topics that most interest them. Content with an international audience can be especially beneficial; Planetarabia.com demonstrates how such sites can automate localization by storing users' language preferences.
Similarly, personalization engines are useful when you're delivering content to mobile devices, where pages must be tailored to varying screen sizes and interface capabilities. By letting users choose their own media preferences, sites that offer rich media content can avoid deluging customers with data formats and plug-ins their browsers can't handle. (C-Net already stores such preferences for its streaming broadband content.)
Personalization can also let users custom tailor the interfaces of Web-based applications to suit their own work habits, as they've grown accustomed to doing with desktop applications. Portal sites like Excite.com offer great flexibility, though sometimes even a change as simple as modifying a color palette can improve a user's experience on a site.
Tough Choices
The decision to add some level of personalization to your site may seem like a no-brainer. Deciding just how you should implement those features, on the other hand, can be much more difficult.
Should you deploy an existing product, or build your own? While the most basic personalization features can be created fairly easily using tools like Java or PHP, integrating a prepackaged solution is probably more cost-effective if you require more sophisticated features.
A wide range of commercial personalization solutions are available, including options for most of the popular Web application servers, such as IBM WebSphere, ATG Dynamo, Oracle 9i, and BEA WebLogic. Content-management mainstays like Vignette and BroadVision have strong offerings in the personalization space as well.
The downside to many of these products, however, is that they can be extremely expensive. Pricing for BroadVision typically begins at $300,000, while solutions from ATG and Vignette start around $100,000. As with most enterprise-class software, however, pricing is often negotiable. For example, Microsoft Commerce Server 2000, which costs $8,499 per CPU, is also available in a rentable configuration.
Open-source alternatives to these products are beginning to appear, but so far, few products compare to the commercial offerings. As of this writing, none of the most prominent open-source application and content management servers—including Zope, ArsDigita Community System (ACS), and Enhydra—had shipped a personalization option, even though all three had something in the works.
When evaluating any of these solutions, remember that development and support costs can factor heavily into the total ownership cost for a given product. A product that supports open standards is often much more easily integrated with third-party add-ons. And products that support a proprietary scripting language (rather than a Web mainstay like Python or Java) have a much steeper learning curve, potentially requiring expensive training and hands-on support.
While weighing all of the available choices can be overwhelming, consider your options carefully before you commit. Remember, user profiles and customer data are seldom seamlessly transferable from one personalization server to another. Switching products after deployment can be difficult, because it often means starting over. Before arriving at a decision, it may be helpful to first understand how personalization works.
Targeting Users
Depending on the level of customization your site requires, the actual process of targeting specific users can be fairly complex. At the core, all personalization engines must provide three basic functions: profile management, user segmentation, and content targeting.
Profile Management. The most basic concept of personalization is that each user can be uniquely identifiable, and you can therefore record specific data about each user. The resulting data store is generally known as a user's profile, and it might contain any number of information pieces, ranging from the user's name, to the last ten items he or she purchased, to the time of day the user most often accesses the site. Personalization packages generally provide software to store, retrieve, edit, and analyze user profiles.
Note, however, that while every profile represents an individual user, storing a profile doesn't necessarily imply gathering any personally identifying information about that user. In fact, many personalization engines let you generate completely anonymous profiles. These are created the instant a user first accesses the site, and stored either in cookies or by passing identifying codes in URLs. Anonymous profiles don't divulge any user information except their behavior on the site—which is often the most important information.
If, at some point, a previously anonymous user chose to register with the site, that user's anonymous profile could be upgraded to a personal profile. This new profile would retain all of the usage information previously gathered, but it would augment whatever personal data the user opted to disclose.
User Segmentation. Once you've begun to gather information about user interests and habits, next you must begin to assign those users to segments. Only then will you be able to target particular segments for marketing, promotions, or customized content.
Defining user segments can be tricky, as it's really a process of representing business rules in a form that the personalization engine understands. Correct representation of those rules often requires input from non-technical staff; therefore, if your business has complex user segmentation requirements, pay special attention to the user interface of any personalization package you're considering. The user segmentation portion should let you edit user classification rules in a straightforward, intuitive manner, without sacrificing flexibility.
Once your segmentation rules are defined, the personalization engine must employ analytics to determine which users fit into which categories, based on the information that's stored in their profiles. This process may operate based on explicit rules, data intuited from usage patterns, or both.
Content Targeting. Finally, to target certain content to the user categories you've defined, you must also classify the content itself. Which information is appropriate for which types of users, and when? This process usually involves assigning metadata to individual content entities and flagging them for targeted delivery.
On an e-commerce site, this might be as simple as assigning product cross-sells. The process could be more involved for other kinds of Web applications: To target an article for a content-driven site, you might have to generate an XML-encoded abstract describing the article's contents.
Explicit Language
Different personalization engines will implement these three functions to varying degrees, depending on the type of personalization required. Generally speaking, package prices increase as the complexity and sophistication of your demands increases. The easiest (and often the most cost-effective) solution is explicit personalization. Using this method, a site adapts based solely on well-defined, consistent rules.
In one implementation of this technique, the personalization engine first checks for an existing profile for each visitor. If it finds one, the engine might be preprogrammed to automatically populate the site's homepage with a personalized greeting. A different greeting might be issued if the site notices that it's the user's birthday, or the look and feel of the page could change based on the user's preference settings. Further notifications could be delivered based on the user's content channel preferences, and so on.
Another standard rules-based practice is product cross-selling for e-commerce. A user who purchases an electronic gadget from Outpost.com, for example, might be offered the opportunity to buy AA batteries as well. A user who has purchased an ink jet printer in the past may be offered periodic discounts on cartridges or paper.
All of these effects are the result of explicit choices the user has made in the past, which then trigger pre-defined responses from the personalization engine. If, on the other hand, you require a level of customization such that models of user behavior grow and change over time, you'll want to investigate packages that support intuitive personalization.
Use Your Intuition
Rather than merely reacting to a well-defined set of rules, advanced personalization engines develop user profiles based on information intuited from customer behavior on the site. This lets a site adapt to and target even customers whose usage patterns weren't foreseen by its designers. There are several ways to achieve this; the method that's most appropriate for your site depends on your business needs, and your software capabilities.
Content-Based Filtering. As mentioned earlier, the user-targeting process involves categorizing content, generally by assigning it some form of metadata that describes its subject matter. When using a content filtering approach, the site stores metadata from previously accessed pages into a user's profile. Over time, the personalization engine is able to recommend other content based on the user's track record. For instance, a customer who purchased several Anne Rice novels from Amazon.com might receive recommendations for other products coded with the keyword "vampire."
Collaborative Filtering. Another way to generate personalized recommendations is to let your user base do the research for you. You can achieve this by using a form of mathematical analysis, known as an approximate nearest neighbor search, to compare a given user's profile to those of others who may have performed similar actions. For example, if User A has bought some of the same products as User B has bought, then User A's purchasing history might yield other products worth recommending to User B. Amazon.com makes effective use of this technique.
An alternative collaborative filtering technique involves user-contributed ratings. In this scheme, users rank given products or content based on their own preferences. Over time, statistical analysis will reveal users whose preferences run along similar lines. The DVD rental site, Netflix.com, employs this type of ratings system.
Usage Tracking. The most sophisticated personalization engines generate complex usage profiles for each visitor, based on his or her behavior. The primary way to achieve this profile is through log mining, where Web server usage logs are regularly processed to reveal visitors' movements within the site. Having a click-tracking server collect usage information in real time can generate more immediate responses; but such solutions are difficult to scale, as they can be very CPU-intensive.
Usage tracking software is complex, often involving such advanced analysis techniques as neural nets, fuzzy logic, or genetic algorithms to divine patterns in user behavior. Consequently, though this type of personalization system has perhaps captured marketers' imaginations more than others, in practice it's probably the least-often deployed.
It's possible to simulate usage tracking with a rules-based approach, where the personalization engine reacts to specific navigation patterns. (See the review of ShoMaster in this issue.)
Measuring Success
High-end software packages might provide more than just one or two of the personalization methods I've described; and some implement all of them. But don't deploy a given technology simply because it's there. Let your business needs drive the techniques you use on your site. Clearly define your assumptions, goals, and expectations for personalization before you begin.
When deploying a personalization engine, remember that this type of expenditure seldom leads to measurable short-term financial gains. Evaluating the total return you can expect from your investment can be difficult, as hard statistics are rarely available.
One way to measure the success of your deployment is by testing the assumptions made during the content targeting and user segmentation phases of development. Have they been validated by your user's actions since you implemented personalization?
For example, how often do users purchase products that the personalization engine has recommended? Are you seeing an increased number of repeat visits from each user? Are users taking advantage of the customization opportunities?
Ultimately, however, the success of personalization efforts should be measured relative to your business goals. Personalization enables a one-to-one marketing philosophy, in which total share of each customer's business is emphasized over raw market share, and every customer's lifetime value is maximized. When deployed effectively, personalization and customization servers can help strengthen your company's relationships with its customers or partners, making them essential components of any forward-thinking customer service strategy.
Neil is the senior technology editor for Web Techniques.
more
