FCCU Finds The Good Word In Speech Recognition

New members are experiencing easier-and faster-phone-based banking, thanks to speech-recognition technology launched in January at Financial Center CU.

Speech recognition is "not such a chore to learn for new members," as compared with Financial Center's previous touchtone phone-based banking, explained Susan Ward, senior vice president of the retail division at the $315-million CU.

Members direct the system to perform transactions with common commands such as "What is my checking account balance?" or "I want to transfer $250 from my savings account to my checking account."

Dubbed "Moneyline" by the 49,000-member Financial Center, the solution recognizes both members' vocabulary and accents in order to direct a call. The system is OFX-based and linked to the CU's core processing system, according to Greg Schmeisser, director of Management Information Services.

With the touch-tone system, members had to request a menu of transaction codes, and then choose from about 25 options. "That was very difficult for our new members who had not yet memorized the codes," Ward said.

"We had to publish a service flyer listing the frequently-used codes," Schmeisser added. Financial Center thus was able to "free the service flyer from being a users-manual, so that it could promote the speech recognition service."

The CU's former phone-based system "used a DOS-based machine with boards no longer available on the market," he said.

The speech recognition technology, provided by Wixom, Mich.-based Maxxar Corp., adjusts more precisely to members' voices as use continues, learning regional accents and other nuances. "Our members live all over the country and even overseas," Ward said. "One of our biggest concerns was that, with so many accents and variations, our members would run into a huge block because the system wouldn't recognize them. But the system self-corrects, and we get fewer and fewer complaints as time goes by."

Whereas the speech recognition system has performed as expected by making phone-based banking easier, Schmeisser was surprised to learn that the average call time increased while the transaction volume per call increased.

Schmeisser explained that, with touch-tone, "we had a population of members using speed-dial to use the system. Given that, we concluded that "talking" to the system would take slightly longer than speed-dialing the system. We expected perhaps 15-45 seconds of increased average call time."

With speech recognition, however, callers conduct 2.5 transactions in 69 seconds, as compared with 1.9 transactions in 53 seconds with the touch-tone system in April 2002, he said. Instead of increasing call times, speech recognition has decreased call times by enabling members to breeze through the system, Schmeisser suggested.

In addition, one transaction type is no longer applicable in the speech recognition program- specifically, the transaction of asking for the transaction code menu. About 13% of transactions performed in April 2002 were of members requesting the transaction code menu, he said.

Call times have also been reduced because Telephone Service Center agents at the CU spend less time on three-way calls guiding members through the touch-tone transaction menu, Ward said.

Natural language speech recognition technology was chosen over directed language technology because "we thought our members would be more pleased to be able to talk to the system and have it do what they ask," Schmeisser said.

The directed language approach requires callers to respond to a series of automated prompts, becoming more system-directed than self-directed.