Ernestine, Meet Julie

Natural language speech recognition is markedly improving voice-activated self-service.
Karen BannanJanuary 4, 2005

Try to book a train ticket on Amtrak’s home page and you might just swear off train travel. For sure, you’ll swear. A recent attempt to book a journey from New York to Boston, for example, required toggling between windows, looking up obscure station codes, and waiting for slow page repaints. When the order was finally submitted, an error message came up noting that, due to technical difficulties, the request could not be processed.

If only Amtrak’s Web designers were as attentive as the makers of the railroad’s telephone self-service system. That system, which features the digitized voice of an operator named Julie, is a primer on good customer service. Rather than requiring Amtrak’s 20 million or so yearly callers to punch in numbers, the system allows them to voice responses to questions like “What city are you departing from?” And unlike many Web-based self-service setups, Amtrak’s voice-activated operator does most of the legwork for the customer.

Expect to bump into more Julies out there. A new technology, called natural language speech recognition, is markedly improving voice-activated self-service. Powered by artificial intelligence, these speech-recognition systems are altering consumer perceptions about phone self-service, as calls for help no longer elicit calls for help. That, in turn, is spurring renewed corporate interest in the concept of phone self-service. In 2004, sales of voice self-service systems topped $1.2 billion. “We’ve seen voice systems move from emerging technology to applied technology over the last few years,” says Steve Cramoysan, principal analyst at Stamford, Connecticut-based research firm Gartner. “It’s still fairly immature. But it’s proven and moving toward the mainstream.”

4 Powerful Communication Strategies for Your Next Board Meeting

4 Powerful Communication Strategies for Your Next Board Meeting

This whitepaper outlines four powerful strategies to amplify board meeting conversations during a time of economic volatility. 

We Call It Maze

Users of first-generation telephone self-service systems probably would have bet against such a development. Those early programs required customers to punch in an endless series of numbers or symbols when responding to questions. The advent of cell phones, and their Lilliputian keyboards, only amplified the ire of customers who used Touch-Tone customer-service systems. First-generation voice-activated self-service systems weren’t much better. Customers who responded to prompts with complete sentences, rather than monosyllabic grunts, often found themselves backtracking through a labyrinth of menus. “Things still weren’t quicker, because of the number of [menu] trees the person had to get through,” says Maureen Govern, chief technology officer with call-center technology provider Convergys Corp., based in Cincinnati.

Software based on natural language speech recognition, however, has eliminated many of the hang-ups of phone self-service. Instead of matching user responses to predefined lists of isolated words, the software uses a statistical model to connect those words and better understand their intent. The result? “Customers are frustrated less often,” says Govern.

Help Wanted
The average estimated cost of customer service, per customer, per interaction.
Web Self-Service $0.65
Phone Self-Service $1.85
E-Mail $2.50
Phone Chat $4.50
Web Chat $7.50

Same thing for those who sell to customers. While voice-activated systems are more expensive than Web self-service, they’re a whole lot cheaper than operator-assisted customer service (see “Help Wanted”). Amtrak, for example, recouped its $4 million investment in Julie in 18 months, says Matt Hardison, chief of sales distribution and customer service at the railroad. In fact, voice-activated self-service is so cheap that it may eventually cut into the call-center outsourcing business in India. According to research firm Datamonitor, an inquiry handled by a call-center agent in India (rather than domestically) saves U.S. companies about 30 percent per transaction. By comparison, a call handled by a speech-activated system costs about 80 percent less than one handled by a live person.

Moreover, voice-recognition systems enable companies to consolidate 1-800 numbers. Reportedly, Bank of America consolidated more than 4,000 such numbers into one number simply by switching to voice-activated customer service. Says Bill Meisel, president of TMA Associates, a speech-technology consultancy in Tarzana, California: “That would never have been possible with Touch-Tone menus.”

Further, many voice-activated customer-service packages include simplified menu trees and plenty of exit points for users. Thus, customers tend to need less time on the phone. In fact, Lynda Smith, vice president and chief marketing officer at technology provider Nuance Communications Inc., based in Menlo Park, California, claims the average length of a voice-activated self-service call is about 20 seconds. By comparison, customer calls to customer reps typically last nearly a minute and a half.

This is not to say that speech-recognition systems are perfect. It will take some time before the search engines driving these systems can actually mimic real conversations. And the duration of agent-based calls actually goes up once a voice-automated service is put in place.

Still, with 90 percent of consumers saying they prefer speech-recognition systems to Touch-Tone systems, it won’t be long before the Julies of the world become familiar voices in the 1-800 world.

Karen J. Bannan is a freelance writer based in Commack, New York.