Speech Is Right for Directory Assistance

According to the spring 2009 "National Directory Assistance Performance Index," an independent analysis published semi-annually by The Paisley Group, automated speech-based systems for directory assistance are performing at 99 percent accuracy. This score measures how often the speech recognition system has a high degree of confidence that it has located the caller's request and gives the number without forwarding the caller to an operator.

In the report, automated systems performed better than operators, who delivered 98.7 percent accuracy, and databases, at 96 percent accuracy. Across all call types, 93.7 percent of all calls are being handled accurately, the research found.

“Companies have really taken speech recognition to the next level,” says Andrew Matson, vice president of The Paisley Group. “Gone are the days of TTS systems that sound robotic or deliver garbled responses where they said something unrecognizable.”

According to Matson, speech recognition systems for directory assistance “have really come a long way in the past four or five years,” and have benefited greatly from adaptive technologies. “Directory assistance is such a difficult area for speech recognition because there are so many utterances,” he says, “But companies have gathered so many utterances that the systems can learn from as they move forward. If the system is not able to find a listing and the call goes to an operator, the speech system watches the operator to learn from her and adapt.”

Speech recognition systems also benefit from a lack of human error that often occurs with directory assistance, and also delivers the information at greater speeds. “As good as directory assistance is, a lot of times the calls are much shorter when they’re done with full automation,” Watson states.

In compiling its data, The Paisley Group used a "mystery caller" approach that tracked:

  • Customer fulfillment, measuring speech recognition, operator, and database accuracy to determine how often DA callers receive a correct listing report;
  • Customer care, the success operators demonstrate in balancing customer advocacy, practice adherence, and call-handling efficiency; and
  • Passed calls, combining customer fulfillment and customer care to measure how often directory assistance users receive a correct report and are treated in a competent, courteous, and professional manner.

The spring 2009 Paisley Index measured AT&T, Qwest and Verizon in the incumbent local exchange carrier (ILEC) segment. This segment averaged 91.3 percent in customer fulfillment, 89.6 percent in customer care (up from 87.3 percent in the fall of 2008), and 83.7 percent in passed calls (up from 82 percent in the fall of 2008).

The third-party segment included iTouchPoint, kgb_USA, TELUS, and Verizon LiveSource. Third-party providers averaged 95 percent in customer fulfillment, 92.7 percent in customer care, and 88.3 percent in passed calls.

The wireless segment included AT&T, Sprint, T-Mobile, and Verizon Wireless. Customer fulfillment scores were clustered in the 93 percent to 96.3 percent range; customer care averaged 96 percent, and the average for passed calls was 91.3 percent, an industry best.

The Free DA Segment included AT&T, Jingle, Microsoft, and Verizon. Both AT&T and Verizon offered operator backup, while Jingle and Microsoft were fully automated with no operator backup. Among them, customer fulfillment averaged 81.6 percent, customer care averaged 75.9 percent, and passed calls averaged 61.3 percent (up from 55.7 percent in the fall of 2008). 

The Free DA segment also included requests by name and by category. The average customer fulfillment for name searches was 81.5 percent and for category searches was 82.1 percent.

The automated DA services segment included MCI, Qwest, Sprint Long Distance, and Verizon. The average for customer fulfillment was 93.3 percent, for customer care was 92 percent (up from 90 percent in the fall of 2008), and for passed calls was 86.7 percent.

SpeechTek Covers
for qualified subscribers
Subscribe Now Current Issue Past Issues