Voice authentication and clone detection at a life insurer

A large Indian life insurer with 3.1 crore policies in force — approximately 1,200 seats across three servicing centres handling around 9.4 lakh inbound calls a month on NICE CXone, with an IVR front end and on-premises recording and retention.

The challenge

Knowledge-based verification was slow, failed too many genuine customers, and could not see a cloned voice.

• 

  Six high-fund-value surrender requests in four weeks passed knowledge-based verification, all six in the same voice, matching none of the policyholders; the answers came from a leaked database and the voice was cloned from public social-media audio.

• 

  Verification took 47 seconds per call across 9.4 lakh calls a month, and scored lowest of every item on the post-call survey.

• 

  12% of callers — ~112,800 a month, mostly legitimate — failed the knowledge script and were routed to a branch visit or callback.

• 

  Quality review covered just 2% of calls, so scorecards and complaint categories were built on sample bias.