If you are an e-marketer, you know building and sustaining a profitable business on the Internet is not an easy task.
Because banner ad click-through rates hover at less than 0.5 percent, it’s evident that mass marketing techniques on the Web fall far short of effective. Even the traditional segmentation techniques of direct marketing are ineffective at best, with the average sales conversion rate lingering between 2 percent and 3 percent.
The good news, however, is that we can do better. Given the personal interactivity of the Web and the technologies available today, e-marketers can achieve more effective results for their efforts – even profitability.
According to industry press and research reports, the solution is personalization. Personalization is the use of technology to offer products that are uniquely relevant to each individual visitor, creating a dynamic Web experience that is much more effective than typical static merchandising techniques. A virtual panacea, personalization claims to do everything – increase stickiness, drive sales and create a more pleasant experience for the customer. Does it really work?
Yes, albeit in varying degrees. Jupiter Communications, New York, reported personalization techniques such as collaborative filtering-based suggestive selling, like that used by Amazon.com and CDNow, increases sales by 52 percent. Jupiter attributed this significant rise to increased sales conversions, larger order sizes because of cross-sell and higher customer retention rates. Other industry studies have also attributed a 50 percent increase in conversion rates from the use of collaborative filtering-based personalization.
To further lend credibility to these assertions, Jeff Bezos, founder/CEO of Amazon.com, publicly attributed much of Amazon’s success to its use of collaborative filtering technology. Given all of this, it’s easy to see why e-marketers are clamoring to add personalization technology to their sites in an attempt to increase sales and the bottom line.
Personalization is hot – sizzling hot. Taking advantage of the buzz, many artificial intelligence and profiling technology vendors have repositioned themselves as personalization companies and are hawking their software and application service provider services. However, all this noise makes choosing the right solution a daunting task. How then do e-marketers distinguish the real from the hype and choose the best solution?
There are many forms of personalization – ranging from simply adding a person’s name to a Web page to sophisticated technologies that recommend the right product to the right individual at the right time. The four main technologies that affect the botton line include: rules-based customization; profile-based targeting; neural network-based modeling; and collaborative filtering-based recommendation. Not all solutions are ideal for every e-business. It is important to know what is right for your organization.
Rules-based customization. This technique uses logical statements to determine what content to include on a particular Web page. For example, a simple rule on an electronics merchant’s site might be: “If the visitor adds a camera to the shopping cart, add a coupon for a roll of film to the next page.”
Rules-based customization is a feature in most customer relationship management programs. These tools enable the e-marketer to create rules and monitor their effectiveness. These tools are fairly basic and require only modest investments in both hardware and software, and are fairly easy to implement. On the downside, manually modifying these rules in response to their effectiveness can be laborious and, because these rules are applied universally to all visitors, rules-based customization falls short of providing a truly individualized experience in which each visitor’s unique attributes are taken into consideration.
Profile-based targeting. This method uses market segmentation techniques borrowed from direct marketing to build demographic and psychographic profiles for each visitor and uses these profiles as the basis for determining which offers should be targeted to whom. Because of the relative ease with which demographic and psychographic profiles can be built (e.g., you can ask visitors to answer a series of questions during registration to determine their age, gender, ZIP code, etc.), a number of profile brokering services can be used in conjunction with ad-serving technologies to deliver your marketing message to your target segment.
However, as banner ad click-through rates attest, targeting on the basis of demographic and psychographic attributes is no more effective when applied to online personalization than it is when applied to offline direct marketing. People with similar demographic profiles can have nothing in common in terms of taste and preferences. Therefore, treating these people as one segment, and marketing to the segment with the same product, is at best a hit-or-miss approach.
Neural network-based modeling. This modeling technique uses artificial intelligence technology to “model” visitors on the basis of their click-stream and purchase behavior and uses these models to predict how a given visitor would respond to a particular marketing message. Each time a visitor’s behavior deviates from what the model predicted, the technology “learns” from this encounter and modifies its model of the visitor accordingly.
The effectiveness of the neural network approach depends largely on the design of the particular neural network; and as yet, there is no standard method for designing a “good” neural network for a particular task. Thus, it would appear that making a network design is more of an art than a science. To date, very few sites have used neural network-based modeling for personalization; therefore, there is insufficient data available to ascertain its effectiveness.
Collaborative filtering-based recommendation. This technique uses proven statistical correlation techniques to find other visitors with similar patterns of behavior and uses the behavior of these “like-minded” peers as the basis for its recommendations. Of the three techniques mentioned above, collaborative filtering is most similar to neural network-based modeling, in that its recommendations are based on a visitor’s actual behavior, not demographic and psychographic attributes. Therefore, these two models tend to be far more effective than profile-based targeting. The key difference in these two approaches has to do with whose behavior is used as the basis for personalization.
Neural network-based modeling personalizes based on the behavior of a visitor’s past behavior; collaborative filtering-based recommendation personalizes based on the behavior of the visitor as well as her like-minded peers. Therefore, it excels at recommending unexpected, yet delightfully relevant products. This technology also allows cross-category recommendations so that a book purchase can be a relevant basis for recommending a particular type of clothing. The downside of collaborative filtering is that it requires behavioral data gathered from a relatively large number of users to be effective. Collaborative filtering has been proved to be extremely accurate. Many of the top-tier e-tailers, such as Amazon.com, have found this technique to be the most effective.
In summary: Rules-based customization is best when you wish to program what a visitor sees in response to a pre-defined choice. Profile-based targeting is best when you wish to reach a demographically or psychographically defined market segment. A well-designed neural network is best when there is not enough behavioral data for collaborative filtering to be effective. Collaborative filtering-based recommendation is best when you have behavioral data about a large number of users.
