September 5, 2000 Search Tech Goes Retail: How the Next Generation Will Transform E-commerce by Joe Gold, onechannel.net

Forrester Research minced no words in its June report, "Must Search Stink?" Despite the critical role a search plays in most sites, little effort goes into selecting or evaluating search tools. As content grows, the report says, search becomes ever more important and "e-commerce sites are wasting content and cripple functionality without decent search capabilities."

Forrester analyst Paul R. Hagen says that the way to fix the search problem is to create a context, adopt good content-management practices, and design interfaces to guide customers. Sadly, many sites fail in even these basic tasks.

A second generation of Internet search tools
The second generation of search is underway, trying to dissect the subtleties of how humans think and how we might apply that to the Web using ever more sophisticated techniques. Several companies are trying new approaches, hoping to develop the breakthrough technology leading second-generation search.

For example, Google broke new ground last year, making it practically a venerated innovator in Internet time. Google's technological premise was that the more links point to a site, the more valuable that site is likely to be. So rather than just searching on keywords, Google factors in the number of sites linking to a particular site, and uses that information to figure out whether something is likely to be relevant to the user. Multiple words in a search were also ranked higher when they appeared closer together. Google searches were faster, produced more relevant results, and resulted far fewer "dumb hits." After a slew of awards for its breakthrough technology, what some people consider the ultimate validation came when Yahoo adopted Google search in June 2000.

Another interesting new approach is Soliloquy's automated "expert," which engages users in an online question-and-answer session using artificial intelligence. This technique is the same style as the technology responsible for the popularity of Ask Jeeves, but extended to a two-way conversation. Soliloquy experts are currently at work helping sell notebook and PCs, and are being developed to handle printers, cars, mutual funds and houses. Among its clientele are Acer, CNET Data Services, Egghead.com, HardwareStreet.com, Hewlett-Packard and Net Perceptions.

What's impressive about this approach is its non-technical nature, which the company hopes is likely to attract average Web shoppers and help them shop comfortably. The irony here is that this would-be simplicity comes at a company with a dozen PhDs on staff, led by Chief Technology Officer Mark Lucente.

The Notebook Expert went live on Acer's site in February 2000. It aims to make online shopping a conversational experience, guiding users through the user's key purchase criteria to direct them to the right product, just as a conversation with a knowledgeable sales person might achieve the same result. Acer reports that after five months with the "expert" on the site, notebook sales are up 21 percent, and 29 percent of the shoppers who use it click through to the buy page.

Here's how the dialogue begins on ShopAcer.com:

Greetings and welcome to Pick-a-Notebook(TM)! My name is Prospero.

I can help you buy a notebook and answer your notebook-related questions. I'm focused and single-minded, so please don't ask me to tell you a joke or to discuss the meaning of life. Just tell me, in your own words, what you are looking for in a notebook.

I want a low-cost notebook.

There are 10 notebooks that meet your requirements.

If you tell me how much you can spend, I'll find the notebooks that are within your budget. You can also ask me about other notebook features such as processor speed, weight, or memory.

Asked where the idea came from for this dialogue with computers, CEO Catherine Winchester unhesitatingly answers Star Trek. "The future of dealing with computers is being able to hold conversations."

A key byproduct of the dialogues with customers is dialogue mining, which Winchester said is yielding unprecedented market intelligence. "For the first time, shoppers are interacting with sites in their own words. You can't find out from clicks why a buyer abandoned a shopping cart. You can't find out from clicks what information the customer couldn't find."

For example, "When people talk about price early on in the conversation, they're much more likely to make a purchase. We see correlations between what people want to use the computer for and which computer they buy. Road warriors tend to buy Toshiba or Sony. Gamers are much more likely to buy Dells."

Winchester said voice versions of Soliloquy experts are on the way, but the market isn't ready for that step just yet. Meanwhile, the company is planning to roll out these experts into electronics financial service markets.

There are but a few of the technologies looking for a foothold in the search marketplace. Traffick.com's Goodman says he sees innovations in "what I see to be several macro-trends about which the jury is still out: pay-for-placement, metasearch, natural language search, popularity/reputation analysis, meaning and context recognition, and expert guide sites such as infomediaries, consumer reviews, etc. The human element is also vital in the major directories."

Another interesting search-technology company is Mohomine in San Diego, a company with data mining software that uses machine learning to mimic the judgment of human directory editors. It hopes to capitalize on the latest fashion in Web site search technology: "verticals," which are portals that focus on one particular type of information. By their very nature, they limit the universe of a search and thus exclude irrelevant hits.

Robin Nobles, author of the new book Maximize Web Site Traffic from Adams Media, says, "The beauty of verticals is that you can submit your site to, and search on, very area-specific engines. For example, FindLaw is a vertical engine that focuses on the law. Its crawler indexes pages just from legal sites, so the topic area is extremely focused." The result, says Nobles, is that "for those who are searching for a particular legal issue, they may be able to find what they're looking for easier and in less time than going to one of the major engines with literally millions of pages and literally millions of irrelevant searches."

The mohoPlatform was built by four college computer science engineering students looking for a way to sniff out existing source code and categorize it for re-use, which is now sourcebank.com. Together, the students founded Mohomine (named for the a deep layer that lies miles beneath the earth's crust) to market their data mining tools for e-commerce sites to hunt down specific products, from vertical portals, auctions, exchanges and government sites to come up with a buried bit of data.

Neil Senturia is Mohomine's chairman and CEO, although his business card reads simply "Chief Moho" of the company launched in January 2000. A onetime television comedy writer, Senturia is a free spirit who was founder and CEO of Atcom/Info, a computer kiosk and hotel-room connection that he sold for more than $100 million.

"We're like the guy who comes in and straightens out your closet," Senturia says. "California Closets, that's our business."

Advanced machine learning techniques in the mohoPlatform brings context to content. It can distinguish between Apples such as the iMac and the red delicious variety in the produce department. And it will pick those apples whether they're buried in text documents, databases, or Web pages.

On the surface, a Mohomine search looks like any other search. It's on the back end, in the process of the search, that Mohomine makes sure results are relevant with topic-specific crawling, automated classification, changing categories on the fly and a search that learns from its experience. Because the "closet" full of data is organized, the front-end of the search can cough up better results.

Next, Senutria compares the Internet to the old days of the U.S. Post Office. "Guys stood there, they looked at the address, and they put it in boxes. Today mail in America is sorted automatically. But it's not done that way on the Internet. The way it's done today is that editors, sitting in rooms, put letters -- content -- into boxes so that you can find your mail. The post office is automated, but the Internet isn't."

The answer, says Senutria, is to automate the classification and make it customizable. "We just change the boxes, resort, reclassify, re-extract," he says. "That's what I do for a vertical portal."

Senturia hops impishly from one example to another. "Cars.com has the need to find custom content that's relevant to its topic. They need to sort it, classify it, extract it and pull out the relevant properties so you can get just fuel-injected engines. They're heterogeneous because some documents are in white papers, some are in sales documents, some are in data sheets, some are articles in Road and Track. They're all different, but we can tell that they're all about cars and engines. Take all the disparate, heterogeneous stuff like little rows of red spaghetti and line 'em up."

A completely different approach uses not human editors, but what they call a community-powered directory that introduces a new version of its software this month after the first release in February. Quiver in San Francisco recruits a mass of hard-core Web surfers to use a downloadable tool called a Qbar, which helps the users manage their bookmarks more effectively. The Qbar tracks user movements, builds a database of the aggregated movements, and uses algorithms to analyze the user preferences from their actual behavior.

"Search is the number two activity on the Web behind e-mail," said Scott Potter, Quiver's CEO. "What we've created is a product in which we track the behavior of a given group of users, then categorize it into a Yahoo-like directory," said. "The Web has become so unwieldy that we've seen an evolution to focus on verticals," which begins narrowing the focus to a specific area.

"We're all about being the Inktomi of vertical search, providing people with a set of tools. We see the market as a series of specific portals: international, product, lifestyle," Potter said. "We want to extract the knowledge of hard-core users," whose collective opinions are better editors than the editors themselves, and gradually build an entire directory in that vertical market based on the expert users' behavior.

The vertical markets Quiver already addresses are the communities that populate gay.com, gorp.com (Great Outdoor Recreation Pages) and the baby boomers' myprimetime.com.

The Quiver plan is that as more users use the Qbar, it becomes more representative and therefore a better index of how people are using the Web. But like many Web sites, it depends on developing a critical mass of users to be truly effective.

One advocate for Quiver's approach is Andrew Goodman, editor of Traffick.com, a site on portals and searches. "Quiver has taken the human element and 'verticalized' it," Goodman said. He says the sees the need to popularize the Qbar as a critical factor, adding that "If Quiver can tap the power of the relevant community to build vertically-focused search solutions, I think they'll do well." Goodman also makes clear that Quiver and his Traffick.com are on the verge of a partnership over a forthcoming PortalHub.com site.

OneChannel.net is the only e-business intelligence service that utilizes actual transaction data enabling businesses to optimize on-line sales performance in Internet time. By aggregating daily SKU-level transactions from e-tailers, subscribers are able to manage their company's e-commerceperformance in a dynamic, easy-to-use, and customizable report on the Web....anytime

onechannel.net