Making it personal

There's an arms race on the web personalization space as vendors of rules-based, collaborative filtering and self-learning artificial intelligence applications jostle for position. The goal: to have the best technology to deliver the right offer at the right time to the right customer.

But despite all the talk about how best to personalize a customer's website experience - through if-then scenarios, by comparing new customers' behavior with past customers', or through AI for predictive analysis - many sites are taking a more old-fashioned tack. They're talking to their customers.

Catalog apparel retailer Lands' End Inc., after flirting with collaborative filtering, a personalization technology based on a complex statistical algorithm that makes product recommendations based on the purchasing patterns of other consumers who behaved similarly at the site, launched a new feature last month called My Personal Shopper.

My Personal Shopper shows customers different outfits side by side and asks them which one they prefer. It also asks customers what type of clothing they're looking for, for what use, and what colors and fabrics they like. Then it takes that information and makes recommendations accordingly.

"It's a totally new way of recommending products online," said Terry Nelson, the Dodgeville, Wis., company's e-commerce marketing manager. "We put the customer in control of the experience. They don't have to buy something from us for us to make recommendations to them. They tell us what things they're interested in, what their preferences are, and we map our products to them."

That effort doesn't stop with My Personal Shopper. Lands' End also uses virtual models on its site that customers can customize to their looks and measurements, to get some idea of how the clothes might look on them. And it's even rolling out scanning stations across the country, where customers can have their exact measurements scanned by laser and uploaded to the site.

"The Internet allows us to be more personal and not just focus on mass marketing," Nelson said. "We want to continue to be innovative and set the bar higher for personalization."

Lands' End may be innovative but is certainly not alone in its concept of personalization. Online educational toys retailer SmarterKids.com Inc., of Needham, Mass., like Lands' End, uses only "referential" as opposed to "inferential" personalization. It creates individual toy stores for children its customers buy for, based on surveys the customers fill out about the children.

"We refer to what the customer has told us and personalize accordingly," said Al Noyes, executive vice president of sales and marketing at SmarterKids.com. "We don't infer that every customer is the same and present people with offers based on what customers with similar buying habits have purchased. Instead, we refer to all the information a parent has given us and personalize the site according to their specific needs, because every customer is different. The parents seem to get it."

SmarterKids.com invites parents to register profiles for their children, based on the child's birth date, grade level and gender. It also asks for a focus area, such as a particular school subject for an older child or developmental skills for a younger child.

In addition, parents can determine their child's learning style by answering a series of questions. Smarter Kids.com then creates an individual storefront for each child with items that fit the child's profile.

SmarterKids.com's personalization technology has been developed in-house, using Microsoft Corp.'s SQL Server database look-up tables and Access Server pages. Personalization software vendors that depend on monitoring clickstream data to build customer profiles may deride such technology as primitive, but it's worked for Smarter Kids.com. About 170,000 children have been profiled at the site, and their parents spend about 60 percent more than customers who haven't profiled their children.

Perhaps more important, as at Lands' End, customers are able to control their own experience at the site. That's a more effective, if less sophisticated, use of personalization, according to Laurie Windham, CEO of San Francisco-based e-business consultancy Cognitiative Inc.

"Effective use of personalization is about building a better relationship with the customer, not trying to fool the customer into doing something they didn't want to do," Windham said.

Part of the way dialogue-based personalization builds those relationships is by acknowledging the privacy concerns of customers.

"Asking users to volunteer information is far more preferable than inferring information based on clickstream data," said Jason Catlett, president of Junkbusters Corp., of Green Brook, N.J., which makes privacy software. "It's better from the privacy perspective and much more likely to be accurate and useful."

But even advocates of this kind of dialogue warn that too many forms and surveys can be counterproductive. Martha Rogers, a partner at marketing specialist Peppers and Rogers Group, of Stamford, Conn., advocates the "golden question" approach, where asking just one question of a customer can yield tremendous insights into what offers to make to that customer. Rogers uses an example of a pet store site asking customers if they buy a holiday present for their pet; the answer shows how much they spoil the pet.

Still, collaborative filtering, rules engines and advanced artificial intelligence applications for Web personalization are by no means dead. Online vendors still find them useful for introducing customers to products that they may not be looking for or to help clear product overstocks and move slow-selling products. They also enable the vendor to give customers special offers, such as free shipping, to entice them to follow through on the purchase.

Internet superstore Buy.com Inc. has experienced a "significant" conversion rate increase since deploying E.piphany Inc. collaborative-filtering-based personalization software last month to set up cross-sell opportunities in the books, games and movies departments, said Travis Fagan, Buy.com's vice president of customer support.

Personalization has been so successful that Buy.com, of Aliso Viejo, Calif., plans to expand collaborative filtering into other areas of its site, such as computer hardware and software, consumer electronics, and wireless devices, eventually setting up cross-category recommendations.

But the site hasn't forgotten about customer dialogue, either. Earlier this month, Buy.com rolled out a form of customer dialogue known as suggestive selling at its site. Customers answer a series of free-form questions in natural language, and Soliloquy Inc.'s Expert determines which product is best suited for their needs.

For now, Buy.com is using the dialogue-based technology to help customers find the right notebook computer, under the brand Notebook Expert. It plans to extend the technology to other products where the buying process can be complicated and the customer needs expert advice.

"You start off with a fairly large suite of options, but, based on what you select, the field gets smaller and smaller until you're left with five choices," Fagan said.

Suggestive-selling engines like this are catching on all over the Internet. Some use free-form responses, some multiple-choice questions. DealTime.com Ltd., a comparison shopping site, last week announced a partnership with Digital Jones to provide a dialogue-driven product recommendation engine on its site for consumer electronics products.

Separately, Nextel Corp. last month added AskJeeves.com's Jeeves Advisor product to its Web site to help customers pick the right cell phone and calling plan. Jeeves Advisor also powers a solution that helps shoppers at Nike.com pick the right pair of athletic shoes and helps shoppers at eTown.com find the consumer electronics products that are best suited to them.

Buy.com is still looking at more ways to increase the dialogue with its customers. "Giving each customer a 'personal zone' is very powerful and is something we're looking at," Fagan said. "We'll have a storefront that's the same for everybody, but parts will be tailored to customers' own needs."

But collaborative filtering and other personalization technologies will have their place at Buy.com as well.

"For products that have a degree of differentiation - like notebooks and PCs - it's important to understand customer preferences and position an offering that's tailored to their needs," Fagan said. "For other categories, like entertainment, that are based more on behavioral preferences than product attributes, it makes sense to offer suggestions to the customer based on whatever data you have available."

Finding the right mix of personalization is something that Buy.com and other online retailers continue to strive for. "Your customers should have the opportunity to use the technology, but it shouldn't be crammed down their throats," Fagan said. "Finding a balance is a million-dollar, probably billion-dollar, question we're all trying to answer."

Here's how different types of personalization systems might generate recommendations based on a customer's purchase of a John Grisham book.

AI-based system will analyze customer's entire clickstream, including links and items clicked on; check shopping-cart contents and the customer's buying history; and combine this information with broad data.

Interactive forms system will launch a questionnaire that might relate to other titles and authors and whether customers enjoy films based on favorite books. System will centralize data for use in other customizations.

Rules-based system will consult databases based on preset rules looking for potential cross-sell items, such as a DVD for a movie of a Grisham book, and new or promotional items.

Collaborative filtering system will check buying patterns of other customers who have purchased the same book and information on related authors and genres.

Personalization has been so successful that Buy.com, of Aliso Viejo, Calif., plans to expand collaborative filtering into other areas of its site, such as computer hardware and software, consumer electronics, and wireless devices, eventually setting up cross-category recommendations.

But the site hasn't forgotten about customer dialogue, either. Earlier this month, Buy.com rolled out a form of customer dialogue known as suggestive selling at its site. Customers answer a series of free-form questions in natural language, and Soliloquy Inc.'s Expert determines which product is best suited for their needs.

For now, Buy.com is using the dialogue-based technology to help customers find the right notebook computer, under the brand Notebook Expert. It plans to extend the technology to other products where the buying process can be complicated and the customer needs expert advice.

"You start off with a fairly large suite of options, but, based on what you select, the field gets smaller and smaller until you're left with five choices," Fagan said.

Suggestive-selling engines like this are catching on all over the Internet. Some use free-form responses, some multiple-choice questions. DealTime.com Ltd., a comparison shopping site, last week announced a partnership with Digital Jones to provide a dialogue-driven product recommendation engine on its site for consumer electronics products.

Separately, Nextel Corp. last month added AskJeeves.com's Jeeves Advisor product to its Web site to help customers pick the right cell phone and calling plan. Jeeves Advisor also powers a solution that helps shoppers at Nike.com pick the right pair of athletic shoes and helps shoppers at eTown.com find the consumer electronics products that are best suited to them.

Buy.com is still looking at more ways to increase the dialogue with its customers. "Giving each customer a 'personal zone' is very powerful and is something we're looking at," Fagan said. "We'll have a storefront that's the same for everybody, but parts will be tailored to customers' own needs."

But collaborative filtering and other personalization technologies will have their place at Buy.com as well.

"For products that have a degree of differentiation - like notebooks and PCs - it's important to understand customer preferences and position an offering that's tailored to their needs," Fagan said. "For other categories, like entertainment, that are based more on behavioral preferences than product attributes, it makes sense to offer suggestions to the customer based on whatever data you have available."

Finding the right mix of personalization is something that Buy.com and other online retailers continue to strive for. "Your customers should have the opportunity to use the technology, but it shouldn't be crammed down their throats," Fagan said. "Finding a balance is a million-dollar, probably billion-dollar, question we're all trying to answer."

Giving customers more control over their own shopping experiences through dialogue-based technology can have an additional benefit beyond improved personalization. It can also ease customers' privacy concerns.

While personalizing based on stated user preferences doesn't automatically resolve all online privacy issues, it does account for one key tenet of a good privacy policy: not acquiring any data on customers without their consent. It can be the first part of establishing the trusting relationship between the business and the con sumer required for successful e-commerce. Companies still have to use the information properly, not resell it to third parties, and must give their customers access to that information, according to experts in the field.

"Talking to the customer is not enough," said Martha Rogers, a partner at e-marketing specialist Peppers and Rogers Group, of Stamford, Conn. "You have to have a privacy policy in writing that cannot be violated."

And that privacy policy must include provisions for consent, disclosure and access, said Jason Catlett, president of Junkbusters Corp. and an outspoken privacy advocate. "It's not enough just to have a privacy policy; it has to be a fair policy," said Catlett, in Green Brook, N.J. "You don't collect information without the customer's consent. You don't give out that information. You let [customers] see that information on demand."

Some companies have even built software products that offer personalization with privacy. New York-based YouPowered Inc., launched three months ago, has a desktop application called Orby and a server application called SmartSense that map a site's privacy policy to Platform for Privacy Preferences standards.

Using Orby, customers can share as much personal information as they want with sites that have installed SmartSense, which is a simple, rules-based personalization engine that enables users to set one of four privacy levels. Customers have access to information gathered at the site at any time.

But YouPowered's software has been slow to catch on. To date, there have been only 2,000 downloads of Orby, a free product, and only four sites are using the SmartSense beta. This month, the company released a free download version of SmartSense, called SmartSense Consumer Trust, to spur adoption.

Whether this technology will ever catch on is debatable. Until now, it hasn't, Catlett said. In fact, another company with a similar offering, Enonymous.com, suspended operations just last month after it could not raise enough new funding.

"All of these companies that try to be 'infomediaries' between consum ers and sites have a chicken-and-egg problem," Catlett said. "If they don't have the consumers signed up, companies aren't interested, and if they don't have the companies, they can't interest the consumer. There's a problem of getting critical mass."

When Amazon.com was the first e-tailer to rocket to international prominence, many analysts pointed to the site's pioneering use of collaborative-filtering-based personalization as one of its keys to success. In fact, for some time, personalization vendors meeting with eWEEK Labs would often describe their products as providing "Amazon-like" capabilities.

Several years have passed since Amazon first deployed this technology, and, despite lots of hype and extensive development efforts by very prominent software vendors, personalization technology has not come very far.

One area of development that has received a lot of attention recently is the use of AI (artificial intelligence) systems to predict customer likes and needs, but the results have not matched the buzz. AI-based personalization pioneer Open Sesame was recently acquired by Allaire Corp., which intends to embed the technology in its Spectra content management system. Web site operators interested in exploring Open Sesame's AI technology for personalization are now out of luck unless they are willing to deploy Allaire's Spectra.

The fallout from the Open Sesame-Allaire deal has pushed AI-based personalization onto the back burner and allowed the focus to return to the original personalization technology: interactive forms that directly ask customers what they like. This method of personalizing Web sites has many benefits, the most notable being accuracy. In addition, site developers can easily build their own interactive forms. However, if a site operator does a poor job of implementing forms-based personalization, it can become the electronic equivalent of the annoying salesperson who accosts customers the moment they enter a store.

Another widely used technique is rules-based personalization, which is embedded in leading e-commerce packages such as IBM's WebSphere Commerce Suite. Rules-based systems have the advantage of being very simple — if the customer does X, give the customer Y — and can be set to follow common-sense patterns. However, rules-based systems require lots of management because sites need to have their rules keep up with customer buying trends.

While it hasn't changed that much, collaborative filtering is still one of the most popular and effective technologies for personalization, led by companies such as Net Perceptions Inc., which was behind Amazon's original system. This technology can misfire when directing customers to specific purchases, but its ability to extrapolate buying patterns based on similarities among customer purchases can be extremely effective when applied to a large customer base.