Natural-Language Search Engines

The Challenge: Weeding through thousands of pages of Web-site documents to locate the answer to a question can be a laborious process and a challenge outside the scope of many search engines. Now, financial institutions are turning to natural-language search engines to make them more efficient and improve the self-help capabilities of their Web sites. Since opening its call center in 1999, LPL Financial Services of San Diego, Calif. has seen its call volume surge to more than 580,000 a year - almost 50,000 a month - as its 4,350 registered representatives turn to the head office for answers to their questions.

However, on closer inspection, Rochelle Putnam, senior vice president of the service center, found that 40 to 45 percent of the inbound calls "were procedural inquiries as opposed to specific account inquiries." They involved questions like: Which form should be used for a specific event?

She knew that many of the answers were buried in documents deep in the bowels of the company's Web site. The problem was sifting through the various documents in search of the answer.

"We estimated we could deflect a chunk of those calls if the representatives had a tool to search properly," says Putnam.

So she set out to find a search engine that would allow the reps to query not by a keyword but by asking question or entering phrases. That's when she learned about natural-language search engines, which hunt down a query by analyzing the language patterns, grammar and concepts used in the search terms. Proponents say it produces a more accurate search.

Matthew Berk, a senior analyst at Jupiter Media in Darien, Conn., says, "There's essentially three ways to log in a search."

The first is a keyword search, which most traditional search engines rely upon and is often used on the World Wide Web. The problem, he says, is that employees don't always know what keywords to enter or how to spell them. Another downside is that they might not spider through all the possible data that's available, especially when the search is confined to a single Web or Intranet site.

Another search is a parametric search. An individual can set the parameters of what they want or don't want, similar to a stock filter search where people ask the engine to show them companies with price-earning ratios between 10 and 20, but only for companies that pay a dividend of 6 percent or more. Berk says these tend to be "very difficult searches" and are labor intensive. Employees have to know the specifics of what they are looking for.

The final is natural language, which assess the concept of the information being sought and provides answers based on questions, phrases or keywords.

According to Jupiter, although businesses are spending money on their Web sites, scant attention is being paid to the search capabilities and that could be costing companies in lost productivity and missed revenues.

It surveyed Web-site users in 2002 and found that surfers see a site search as a chance to short circuit the laborious task of navigating Web sites. However, surfers say the effectiveness of site searches is waning and they are becoming frustrated with trying to find information online.

According to the survey, only 8 percent say a site search is very effective, down from 19 percent in 2001. As well, surfers have beefs with the typical site search:

In Putnam's case, she says LPL's representative Web site is menu driven, meaning that a number of clicks are required to access content. That meant reps might only get a partial answer in one part of the site and then would be required to visit another section to cobble together the information. Rather than do that, they were simply contacting the call center.

So the firm used Forrester Research to help them identify natural-language search vendors and they narrowed the hunt to InQuira, Inc., of San Bruno, Calif. and iPhrase Technologies, Inc. of Cambridge, Mass.

Both have been winning clients in the financial-services space. InQuira boasts Fidelity Investments and Bank of America as clients. iPhrase has Charles Schwab, TD Waterhouse and drives the search on the Yahoo! Finance portal, where investors can ask simple questions to get investing information. For example, investors can ask for a list of companies announcing earnings that day or request the top software companies, according to market capitalization, and the information is presented in a tabulated, chart form.

Tony Frazier, senior vice president of development at iPhrase, says, "Brokerage firms have been very aggressive adopters of this (technology)." He says with traditional search engines, users get success rates on their queries only 40 to 60 percent of the time, which means they abandon that channel and go somewhere else, either to customer support or a competitor. "That's a lot of money left on the table." Natural language can boost query success to 85 percent or more, he says.

In TD's case, Frazier says, the firm "wanted to lower its customer-support costs." So it analyzed the high-volume topics that were coming in to the call center, such as, "How do I change my password?" Or, "What's the difference between a limit and market order?" Then the parties made sure the search engine could respond to those questions. The result is that TD is reducing its customer-support costs.

Reducing support costs was also the goal for Schwab, which says it saves $125,000 per month in customer-service time by using the iPhrase system.

InQuira also boasts support savings. Chuck Williams, president of the company, says Fidelity uses it for its call-center reps so they can quickly find information. "It makes self service much more efficient." Moreover, he says, online customers tend to be more loyal, so anything financial firms can do to make the online visit more palatable means solidifying their relations with clients, an important goal in this economic climate.

However, he says, "I'm not so sure that brokerage firms really understand what the benefits are to (natural-language search engines) or how bad (traditional) search engines are."

He says one unidentified Wall Street firm conducted a return-on-investment analysis and found that it had $3 million in cost savings and an additional $2 million in revenue generation by deploying a natural-language search engine.

LPL had both firms carry out a proof of concept, where they reproduced the Web site and added their search engine. Putnam says she was "blown away by the relevancy of the answers" generated by the natural-language search engines.

She says the competition was "nip and tuck" and, in the end, LPL went with iPhrase. "Out of the box, it seemed to work better with our particular site," she says, but, "It could have gone either way."

The system went live a month after the deal and within eight weeks, 65 percent of the reps were using it to access information from the more than 20,000 documents on the firm's Intranet. Although calls to the call center continue, she attributes that to the fact it's tax time and expects them to decline. "We may roll it to the public site next," she says.

Putnam says firms thinking about deploying a natural-language search engine "really need to be cognizant of a few things." First, executives need to have a good understanding of the terminology specific to the financial-services industry and their firm so that information can be properly indexed for the searches. Second, they need a commitment from the various business units to develop a glossary.

Third, expect to do some Web site maintenance. "You need a little bit of organization and clean up." And will likely have to retag some documents so that the "search engine can more easily find them."

Fourth, it's critical to make sure a firm can integrate all its databases to its Web site and search engine so that employees can tap into the corporate knowledge.

Fifth, consider the reporting capabilities of the tool. "We ended up building some of our own reports. Now what I can do is go online and look at questions that the brokers asked yesterday. Sometimes you develop content in a vacuum," says Putnam. By knowing what people are looking for, firms can make sure their content is relevant.

That's important, notes Williams, because the "answer can be no better than the content."

Jupiter's Berk warns that too often firms sweat the vendor-selection process at the expense of implementation. "Most companies spend 80 percent of the labor and anxiety on which vendor to pick and only 20 percent of the focus on implementation." That's the part firms need to keep their eyes on, he stresses.

---

44 percent say they don't know what to type in

21 percent lament that they can't use full sentences in their search

39 percent say that misspellings are handled poorly

38 percent say that results are typically irrelevant

39 percent complain that searches work differently across sites

This article originally appeared in Wall Street & Technology magazine. Wall Street & Technology is the leading information portal for the financial technology marketplace, allowing traders, investment advisors, analysts and IT professionals to track the latest developments in IT that impact the capital marketsa weekly e-mail newsletter. Visit www.wallstreetandtech.com