Dialogue with an online brain

Seamless interaction between human and computers may be the dream of an inventor. But Mark Lucente has to think about bread and butter issue. And that means putting his talent behind the e-commerce arena.

Technology supergenius Mark Lucente wants to give your Web site a brain so it can have a nice, profitable conversation with customers. Any questions?

"Give me the world," Mark Lucente called out to a wall-size screen projected in front of him. He spread his long arms to define a circle, and a large blue globe materialized, spinning on the screen. The trade-show crowd drew closer - a guy controlling a computer with gestures. "Make it this big," the lanky inventor said to the wall, moving his hands closer together, and the world got smaller. The crowd gasped.

The public got its first hard look at Mark Lucente - and the things that preoccupy him - three years ago in the IBM booth at computer trade show Comdex. At the time, the 30-something MIT-trained researcher was exploring new ways for people to interact with computers, and many still talk about the demonstration he gave - part David Copperfield, part high-tech soothsayer - in the crowded Las Vegas Convention Center.

Lucente, now 36, has spent his career inventing technology like this - natural interfaces to computer data that might someday be the way we all control machines. His work at IBM making computers responsive to body movements was a milestone on the road to seamless human-machine relations. Someday, indeed, the break throughs Lucente has pioneered might help seismologists model oil fields without touching a keyboard or online shoppers inspect merchandise that isn't really there. Nothing would make him happier.

But for an ambitious young inventor, the promise of someday is a carrot. After a while, the meat and potatoes of now start to look tasty too.

Now means e-commerce, which Lucente hopes to revolutionize with user-friendly tools. So in a Silicon Alley loft space, as chief technology officer of startup Soliloquy, Lucente is building tools that let Web merchants bring their products to shoppers on more human terms. We're not talking about video holograms and gesture-based interaction - not yet - but we're talking about tools that can bolster companies' bottom lines today.

The first of Lucente's tools is something called Notebook Expert, which lets visitors at a Web site type a phrase like "Find me the lightest laptop that plays DVDs," and receive a list, plucked from a product database, of lightweight, movie-playing computers. It sounds like a simple natural language database query, but it's much more. Lucente's computer expert has a dialogue with the user, assembling what resembles a human understanding of what the shopper really wants. It might be just an explanation of how much RAM you need for playing games. Or it might be a very accurate and insightful product recommendation, and an easy way to buy the notebook you want.

Online Brain: What do you want to use your notebook for?

Shopper: I play Tomb Raider.

Brain: The latest computer games require a fast computer with a big screen, a large hard drive, and lots of memory. There are 60 items that satisfy your requirements; here are some examples. How much money can you spend?

Shopper: I need something cheap.

Online Brain: There are 25 items that satisfy your requirements ...

In the course of that short dialogue, the brain has pulled data from various places: from its internal knowledge base to interpret what "play" and "cheap" mean in the context of notebook computers, from the seller's product database, and from the profile it is building of the shopper.

Does it work? In a five-month trial on Acer America's Shopacer.com site, nearly one-third of shoppers who used Notebook Expert clicked through the "buy" page, and Acer's online notebook sales rose by 21 percent. To e-tailers, that's magic - even more amazing than waving at a screen to make a blue ball appear.

And it's just the beginning. Soliloquy is working on more experts - to help shoppers learn about and choose mutual funds, home mortgages, and cars.

Lucente brings his technological genius to e-commerce at a critical time. BizRate.com, an e-shopping rating and comparison site, estimates that shoppers abandon online shopping carts 78 percent of the time. Boston Consulting Group estimates that e-tailers missed out on $15 billion thanks to aborted purchase attempts in 2000, on total sales of $61 billion. This sky-high rate of e-commerce interruptus frequently happens when shipping charges are revealed or shoppers are reluctant to fill in forms, but it's also often a result of disorientation or inability to find the desired information.

And that's why Lucente hopes his work connecting customers with information could give the e-tailing industry a boost. It turns out that being there to hold shoppers' hands when they need it - to help them find what they need, even to literally close the sale - is another one of those fundamental good-business practices that never really went away.

There's more potential than reducing customer-support costs. The aim is to turn these valuable moments of customer contact into profit.

Soliloquy, of course, is not alone in recognizing the potential of providing consumer guidance. Software companies have swarmed in to help Web site operators solve the problem of being there.

Most of these companies operate as application service providers: For a set fee, they host the software that provides answers when a Web site visitor asks a question (for more details, see "Next Question, Please!" page 108). New York-based AskIt.com matches customer queries to a stored bank of responses to frequently asked questions. It routes unmatched questions to experts for response by e-mail. ALife-RoboShop, from Boston-based Artificial Life, offers a 3D animated character that appears on a Web site, blinking and smiling, and acts as a virtual storekeeper. LivePerson, Human Click, and others let customers chat via text with human customer service reps. iPhrase Technologies of Cambridge, Massachusetts, offers its own version of a brain that makes it easier for shoppers to extract Web site content by typing plain English questions (for more on iPhrase, see "Meet the Other MIT Grads with Online Brains," right).

The common goal of these services is to reduce customer-support and call-center costs. Forrester Research estimates that it costs $33 for a company to handle a phone call, but just under $10 per session for interactive live chat with a human assistant. Soliloquy says its experts can lower a company's cost to just 7 cents per customer session.

But there's more potential to this technology than reducing customer-support costs. The aim is to turn these valuable moments of customer contact into profit. Boosting conversion rates -- the number of shoppers who click through and buy instead of ditching their shopping carts -- is part of that proposition. And if you can really know what a shopper wants and make him or her happy with the experience, you create a special bond between e-tailer and customer -- and new business opportunities. For example:

Online Brain: You mentioned that you want a computer to set up an accounting system. We have negotiated a special arrangement for our customers with Acme Accounting Software . . . click here to learn more.

With some people fueling his visions and others trying to reign them in, Lucente has had a colorful history as an inventor.

Lucente's ambition began early. He chose MIT, he says, because it was a "party school," and ended up meeting many of the friends who were to become his colleagues at Soliloquy. For his doctoral research, he gravitated toward the university's famed Media Lab and chose one of its most far-out fields: taking holograms, like those etched on credit cards, and making them come to life as a video that could be altered in real time.

That had never been done, says Stephen Benton, head of the lab's Spatial Imaging Group, where Lucente did his research. "Holographic video is one of the most long-term research projects at the Media Lab. . . . When Mark was here, . . . nobody believed it could be done at all. There were 20 years of received wisdom that it was foolish to try."

With a series of lenses and a supercomputer, Lucente and his colleagues modulated light waves that projected into space a golf-ball size 3D image of a Volkswagen Beetle. An observer turning a knob could rotate the image -- the world's first interactive video hologram.

Even today, though, that technology is light years away from consumers. "Ten years from now will it be hard? No. You'll be buying novelty items, the equivalent of mood rings," Lucente says. He was more eager than that to bring his technology vision to real people.

His next stop was IBM, where in 1995 he ran his own research team at the prestigious Thomas J. Watson Research Center in Yorktown, New York. There, Lucente seized upon IBM's ViaVoice speech recognition software as a way to capture input from people "on their own terms." He then added gesture recognition for his memorable human computing demonstration at Comdex, which he called DreamSpace. Still, his inventions remained far from use by the general public. "I had some people telling me to focus on the farther out stuff, and then I'd have other people say, 'You know, it really helps loosen up the wallets if you can say and this can easily be part of IBM's product line in the next couple of years.' "

IBM had brought Lucente closer to real users than the Media Lab had, but it still wasn't close enough.

Meanwhile, in downtown Manhattan in mid-1997, Catherine Winchester, a Silicon Alley programmer-turned-entrepreneur, had started a company with a vision in sync with Lucente's. She showed investors a prototype "interactive expert" system that spoke aloud in a dialogue with users.

"The vision was that you'd be able to talk to the Web and hold conversations with it as if you were talking to real people," she says. The demo was an expert on Beatles music. Using a headset connected to a PC, a user could ask, for example, which song contained the lyrics "rocking horse people eat marshmallow pies." It would respond to the verbal query with a synthesized voice, saying "Lucy in the Sky with Diamonds," and would then provide an onscreen list of albums containing the song -- and let users hear it. "It was ahead of its time," she says. "People said, just give us a typing interface in natural language."

When Lucente came across Winchester's Soliloquy Web site in mid-1998, on which she explained her vision and her prototype, he knew he'd found his next home. He became Winchester's first full-time hire and CTO. Winchester and Lucente agreed that the only plausible beginning would be with a specific field of useful expertise. So while she was out raising capital from sources like the Genesys Angelbridge Fund and Gabelli Group Capital Partners, Lucente put together a team of brainiacs, including many of his MIT buddies, and reworked her prototype as a text-based tool focusing on notebook computers.

There are keyword tools such as Ask Jeeves and live chat such as LivePerson. But Lucente hopes his online brain will be the no-brainer choice.

The best known question-answering tool online is Ask Jeeves, which under the name Ask Jeeves Relevant Answers is helping consumers navigate corporate sites, including Radioshack.com and Officedepot.com. Ask Jeeves also offers a tool called Ask Jeeves Advisor that lets shoppers respond to multiple-choice questions to narrow a product search. It serves as the Product Recommender at Nike.com, for example. But Soliloquy's brain doesn't just match keywords to preconceived questions and precooked answers. It has a dialogue through which it learns what you want.

Notebook Expert may be able to interpret dozens of terms that in some way define weight (pounds, heavy, light, and so on). The concept of weight -- and how much of it you want in a laptop -- is stored in its own object, a smart chunk of computer code that combines with other objects so the expert brain knows your goals when it asks or answers its next question. The shopper isn't traveling down a flow-charted path. He or she is having a free-form dialogue with an entity that learns what the shopper wants and cleverly leads the way toward the cash register.

Online Brain: You mentioned that you travel. A lot of travelers like to carry an extra battery . . .

The shopper's goals are, of course, matched as accurately as possible to the goals of the Web site -- which is to sell products. When working with potential clients, "We say, 'Just tell us what you want it to sell, what you want it to look like. Should it do a lot of cross-selling? Should it do a lot of upselling? Should it be laid back and chatty or more direct?' " Lucente says. "We meet with them for an hour, then we go off for two to three weeks and make visual changes and link it into their product database."

Behind the scenes, the software does what it calls Dialogue Mining, generating reports on the captured customer conversations that help merchants understand what people want. It can be simple enough: The sales bump that Acer got during its test of Notebook Expert came largely because the company learned customers wanted two laptop models that weren't on its site. The company quickly added them. The brain can get deeper into shopper psychographics: If 80 percent of customers buying a particular laptop asked first about weight and then about speed but never about price, the seller might figure that buyers of that model aren't highly price sensitive, and it might fetch a higher price.

Following up on Notebook Expert will be Desktop Expert, and Soliloquy is working on brains for other specialized areas. Each takes about two months to build, and each will borrow chunks of knowledge from the master brain Soliloquy is building. For example, the computer- buying expert's understanding of colors -- that tangerine is a kind of orange and teal is a shade between blue and green -- can be patched easily into a car-shopping expert.

Eventually the result will be a huge online brain that knows, well, a lot.

Then what? Voice and animated characters that eliminate typing are on the drawing board. A speaking computerized expert could vary its intonation to help customers along, Lucente says: "Should it sound happy? Should it sound insistent, confused? If there's an animated character on the screen, it might look at the person at a certain time or look away," depending on how it wants to interact.

The real question is whether the Soliloquy brain works better than flesh and blood. Online chat with live customer service representatives is relatively easy to set up. People come with brains and conversation skills of their own, and they're ready to work tomorrow.

"It reminds me of the original mentality of the Internet," says Robert Lo Cascio, CEO of LivePerson, whose software allows online chat with human helpers. "The Internet created the ultimate way not to touch a customer. Just to create more technology to do that, I don't think it's the answer." For kicks, LoCascio goes to the Notebook Expert demo screen that Soliloquy has on its site. He types "I want a computer with a broken screen," and the expert says it has found 711 matching laptops. "I'm just spoofing it," he says. "But, you know, obviously that's not something a real person would come back with."

On top of that, Soliloquy's ambitious technology is a tough sell. Though Lucente has been working on it for nearly two years, Soliloquy just signed its first paying customer last fall, a version of Notebook Expert for CNet's shopping site. By contrast, LivePerson, founded in 1998, claims to be used on 800 Web sites.

Still, Soliloquy's approach may ultimately be much less expensive -- and is clearly more scalable than a human staff. Soliloquy's solution will cost a company six figures to implement, Winchester says, declining to be more specific. Live Person's fees are lower -- $350 a month per seat for its chat software -- but then you need to train and pay a staff of human reps.

And, of course, Soliloquy's brains will get harder to spoof. Betting on Lucente to make that happen is a no-brainer.

iPhrase turns plain english into data.

Who knew it would be so hard to get someone to take your money? Toward the end of 1998, Noam Ben-Ozer (pictured), then a management consultant at Bain & Co. in Boston, had carefully keyed in the configuration for a Dell computer at that company’s Web site and was just about ready to buy it. But he had one quick question: Was the machine powerful enough to run multimedia software? After fruitlessly searching for an answer on the site, he abandoned his online shopping cart. Ben-Ozer then called Dell’s toll-free number and talked to a salesperson who assured him that yes, the computer he wanted was powerful enough for the job. He made the purchase by phone.

“So I went through a lot of pain when I should have been able to solve the problem on my own,” Ben-Ozer says. “And Dell’s margin on the sale was reduced by the cost of the call and the salesperson’s commission.”

Shortly after that e-shopping episode, Ben-Ozer quit his job, pulled together a tech team, and started a company to help others get around the same problems. Founded in November 1999, iPhrase was launched to solve the Web’s “information access crisis”—the difficulties people have finding data that is, in many cases, already posted on a site. Ben-Ozer’s cofounders were three MIT-trained computer scientists with natural-language expertise. They included Raymond Lau, his chief technology officer, who in 1987 invented StuffIt, the widely used file-compression utility for the Mac OS.

iPhrase’s first product is an impressive financial-data query tool. Type “Five-year chart for Intel” into a box on a Web site and hit Enter, and a chart showing five years of Intel stock performance instantly appears on the screen. Type “How about CSCO” and a five-year chart for Cisco Systems immediately takes its place. Type “upgrades for Ericsson” and you get a list, going back several years, of all stock analyst upgrades or downgrades of the Swedish telecommunications company’s stock. iPhrase can pull up pages, like those charts, that may be buried six or eight levels deep in a Web site, or assemble new pages on the fly. “Compare EPS, PE, and market cap for Dell, Compaq, and Gateway” produces the table you’d expect it to, drawing from multiple data sources. “Biotech companies with market caps over $1 billion and their locations” sorts biotechnology firms by valuation—and from a separate database it retrieves an address for each.

Like Soliloquy’s subject-area experts, iPhrase’s technology isn’t cheap—incorporating it into a site costs from the mid six figures to more than $1 million. The tool is slated to go live sometime this quarter for its first customer, Schwab.com. There it’ll be customized to give customers easy English-language access not only to market data, but also to their own accounts. Getting answers will be a breeze. Though some questions—like “How much did I lose on NASDAQ stocks last quarter?”—may never be easy to ask.