The bank, Barclays Wealth and Investment Management, uses 20 to 30 seconds of the conversation with the phone agent and compares the audio WAV file to a sample taken from that customer earlier. If the software thinks it's a match, the agent is silently signaled that the customer's voice has been verified. If the software does not find a match, agents are supposed to use their regular security questions to verify.
As long as voice patterns are a supplemental security mechanism, this could work well and sharply accelerate calls. False positives are much less likely than false negatives, given the large number of elements that could impact the sounds being analyzed, such as if the shopper is sick or very tired. It's not clear whether the analysis is based solely on the sound of the voice or if it also factors in speaking patterns, pauses and whatnot.
The original sample is gathered without the customer's knowledge, by recording and analyzing multiple calls after the security authentications checks have been made, according to information on the site of Nuance FreeSpeech, the vendor that is working with Barclays.
The software also has a list of audio samples from known fraudsters, the company said.
Since its introduction, 84 percent of Barclays' customers have been enrolled in the system, with 95 percent of those customers successfully verified in successive calls.
Matt Smallman, client experience strategy and change, Barclays Wealth and Investment Management says that customer feedback has improved since the technology was introduced five months ago, with 93 percent of customers rating the bank at 9 of 10 for speed, ease of use and security.