InformationWeek

CRM Makes Online Shopping Personal

Charles Waltner

January 29, 2001

Personalization tools help customers with purchases and let merchants cross sell

By Charles Waltner

Illustration by Philippe BehaIExplore Inc., a Chicago travel company with an extensive Web site for adventure-travel advice, products, and services, has developed enough information to keep any globe-trotting soul happy. But with more than 50,000 pages of dynamic content and 1,056 separate trips just for hiking, trekking, and walking, visitors to iExplore needed a tour guide just to find their way around.

To keep visitors from getting lost in iExplore's labyrinth of information, the company decided to make its world of adventure travel a little smaller. Rather than force visitors to navigate the site from pole to pole seeking information on possible travel destinations, iExplore turned to Art Technology Group Inc., which has customer-relationship management software that drives the automated personalization of Web-site content for individual shoppers.

ATG's Dynamo suite lets iExplore deliver content to site visitors based on their particular interests, helping customers find the content they want with far fewer clicks and in less time. "The Dynamo server provides customers with information that they might never have found," says Bill Bernahl, iExplore's chief technology officer.

Web-site personalization has been around since 1998, and today the market has as many as 40 vendors, including ATG, BroadVision, Data Distilleries, E.piphany, Manna, NetPerceptions, Personify, and Verbind, each offering its own take on driving automated Web-site personalization. That's proving a timely development for Web merchants. As sites get larger and larger, personalization programs can ease customers' shopping experiences by getting them the information they want quicker and more efficiently. The technology also offers an effective way for merchants to up-sell or cross-sell by providing offers attuned to visitors' interests.

Personalization software generally addresses three areas: product sales, the look and feel of Web-site content, and ad generation. The applications create individualized content, or at least content narrowed to certain customer categories, based on customer data, including clickstream activity, registration information, third-party demographics, purchasing patterns, and other information stored in databases or derived in real time from visitor activity. Though technology varies, most personalization tools use mathematical algorithms, business rules, or more advanced forms of artificial intelligence to determine what information customers want.

Still, industry analysts warn that although most personalization software technically works, it's difficult for a computer, or a person for that matter, to accurately understand and anticipate the interests of a customer. Customers can grow weary of crudely targeted offers, says Sharon Ward, director of enterprise business applications at Hurwitz Group. People's buying habits often are far more eclectic than any software can anticipate. As a result, "personalized" offers often miss the mark and can lead to more customer disdain than satisfaction, Ward says.

"It's not so much about technology as about strategy, tactics, business processes, and understanding customers," says Scott Nelson, a research director at Gartner. Partly for these reasons, the service costs of setting up and maintaining personalization software can run two to five times the cost of the software, he says.

Bill BerhahlPhoto by Blair JensenBut companies such as iExplore are finding that personalization tools, with some careful thought behind them, can provide far greater benefits than drawbacks. IExplore is one of the more adventurous Web sites using Web-personalization tools to boost sales and customer satisfaction. CTO Bernahl says that almost all areas of the site can be customized for each customer in one way or another.

The ATG software, which iExplore installed last February, uses an array of business rules to direct content to customers. Content pushes include specific links, including those to an iExplore topical expert; specific content, such as information on Zambia or kayaking; and merchandise, such as books and clothing for travel in different parts of the world.

IExplore draws on information primarily from customer-registration profiles and real-time customer activity to decide what content to provide. A customer, for example, might be interested in exploring Africa via a safari. By clicking on "safari," the Dynamo server can provide a link to an area in the kayaking section, among other content, that provides information on kayaking safaris in Africa, a type of travel that the customer might not have considered but could nevertheless find appealing.

Such anticipation of customer interests helps reduce the number of pages site visitors must sort through and increases the likelihood that customers will find what they need, Bernahl says. In bottom-line terms, the ATG personalization engine increases iExplore's effectiveness and reduces the amount of information that the site must provide over its toll-free telephone lines, a far more expensive channel to maintain than the Web site, he says. Bernahl, however, hasn't conducted a return-on-investment study to determine how much the Dynamo personalization saves in toll-free telephone costs.

In general, Web-site personalization applications are divided between software that uses business rules to guide personalization and software that relies on mathematical analysis of database information to determine the most likely interests of shoppers. Pricing varies, but the sophisticated technology generally isn't cheap.

The Dynamo platform used by iExplore, for example, averages $250,000, says Rich Caplow, director of product marketing at the software vendor. But that product isn't just for personalization; it also includes an application server and a commerce server. E.piphany, meanwhile, charges $250,000 and up for its products, including service and support. E.piphany relies on Web technology such as C++, HTML, and Java to run its application. The product uses mathematical algorithms known as Bayesian networks, which analyze data stores for predicting the interests of customers.

"It's a horses-for-courses issue," says Andrew Dorward, director of personalization for the technology unit of Bertelsmann E-Commerce Group, the parent company of online retailer BOL .com. BECG is a subsidiary of European media conglomerate Bertelsmann AG. Dorward says one type of personalization engine isn't best for all circumstances; companies must pick the type of customization most appropriate to their sales and marketing goals.

BOL.com currently uses collaborative-filtering technology from NetPerceptions; however, the company is evaluating other personalization tools as well. "Some technologies are better for certain types of personalization goals than others," Dorward says. Rules-based personalization, for example, provides direct control of what types of information site visitors see, but it doesn't scale well, he says, because the rules need manual updating. On big sites, that can become onerous.

An analytic personalization approach, such as collaborative filtering, is excellent for media products and for items that inspire browsing and choices based on personal taste. It doesn't do much for directing purchasers to particular items, but it provides general encouragement to buy and an alternate way to navigate a site, Dorward says.

Collaborative filtering is perhaps the most widely known type of personalization software because of Amazon.com Inc. This software takes information from other customers and compares it with the actions of a particular shopper, then provides content based on overlapping interests. For example, if someone clicks on a CD by a particular artist, a collaborative-filtering engine would then offer a handful of CDs bought by other people who also purchased that same CD. Collaborative filtering can also draw information from customers' product ratings.

NetPerceptions pioneered the filtering mechanisms made famous by Amazon. Bertelsmann was also an early partner of NetPerceptions, deploying the technology in June 1998. Dorward wouldn't disclose how much BOL.com paid for the NetPerceptions product, but he says return-on-investment analysis indicates the personalization software paid for itself in a few months. The assessment is based primarily on the growth of sales before and after the software deployment.

Andrew DorwardPhoto by Douglas LevereDorward notes that NetPerceptions could make some improvements, particularly with its reporting mechanisms. But small shortfalls are easy to live with because trying to build a sophisticated Web-personalization mechanism in-house is foolhardy. Says Dorward, "That way madness lies."

Web merchants are beginning to benefit from the relative maturity of Web-site personalization tools. The Dynamo server, for example, is in its fifth version, and Art Technology Group has more than 450 customers, including such established retailers as Blockbuster, J. Crew, Kmart, and Target. Many vendors are making solid strides in providing more sophistication to their products. ATG added a "scenario" server to its suite. The product keeps track of customers' actions over time, building a broader base of knowledge about the individual and driving proactive marketing, such as prompting new offers at the site or triggering E-mail advertising. And NetPerceptions is expanding on its core collaborative-filtering technology to include rules-based decision making.

Steve VanTassel, VP of commerce solutions at NetPerceptions, says the rules-based engine provides merchants with more options for controlling the type of content driven by collaborative filtering. For example, a merchant can write a rule that limits the types of items prompted by collaborative filtering to only the two most profitable CDs; or an E-retailer can tell the system to show an item for up-selling to all customers shopping for CDs.

The capabilities of Web-site personalization software are outstripping the abilities of companies to understand and implement the technology, given all the other challenges they face in running a quality site, says Sheryl Kingstone, a program manager at the Yankee Group. "We don't see a lot of companies using the technology to the full extent of its potential," she says.

Still, Web companies don't necessarily need to use the most elaborate personalization tools to gain worthwhile benefits for their sites. Some sites find lighter-weight personalization tools are equally effective.

Jack Kiefer, CEO of BabyAge.com Inc., a Secaucus, N.J., distributor of baby products to specialty stores, says the $30,000 or so that NetPerceptions wanted for its software was more than his company could afford. Instead, BabyAge turned to ResponseLogic Inc. and its core product, Adapte.

Though Kiefer won't say how much Adapte cost because his company was a ResponseLogic beta customer and received an unusual discount, he notes the product was far more affordable than the NetPerceptions tool and took only two weeks to deploy. Adapte draws on the order history of BabyAge's customers and their ZIP codes to individualize content.

Interest in certain types or styles of baby products is closely tied to geographical area, Kiefer says. People in the Midwest, for example, rarely buy wooden high chairs, while customers in the West want more rugged strollers for activities such as running. Products emblazoned with Mickey Mouse sell well in the East, but people in the Midwest prefer clothing and toys with Winnie the Pooh graphics, Kiefer says.

BabyAge uses rules-based drivers from Adapte to design the home-page content that targets a customer's geographical area and assumed product preferences. "We want to make the front page as efficient as possible," Kiefer says. BabyAge also uses the rules drivers to promote ancillary items fitting the same geographical tastes. If a customer buys a crib, for example, Adapte knows to offer a bedding set with Winnie the Pooh designs to a customer in the Midwest.

Kiefer says Adapte also offers strong reporting tools for tracking click-through responses on the customized, rules-driven offers, making it easy for his company to determine which offers are working and which aren't. "Tools like Adapte give you so much more power to merchandise," Kiefer says. "And they should only get better as the years go by."

Illustration by Philippe Beha
Photo of Dorward by Douglas Levere
Photo of Bernahl by Blair Jensen