Fraud alert: Voice authentication platform analyzes 1,380 data points per call

Fraud alert: Voice authentication platform analyzes 1,380 data points per call


Pindrop’s dashboard scores the caller, the device, and the behavior to spot bad actors and authentic customers.

Machine learning platform focused on finance can spot a strip club alias a mile away
AppZen uses computer vision, data augmentation, and natural language processing to spot fraud and compare contract terms with actual bills.

What can you learn from analyzing more than 2 billion phone calls per year? Everything you need to measure to detect fraud and authenticate genuine users. 

Pindrop analyzes 1,380 characteristics of a single phone call to create a risk score. The system doesn’t need to create a profile of every caller to figure out which calls are fraudulent. Because Pindrop collects data about the caller, the device and the carrier, it’s easier to pinpoint suspicious calls.  

“For example, I have no idea who you are but I know you usually call from an AT&T phone from San Francisco, but the audio analysis is telling me it’s a Skype call from Nigeria,” said Vijay Balasubramaniyan, CEO and co-founder of the company. “I don’t know who you are but I do know a Skype call from Nigeria is not you.”

Balasubramaniyan said another sign of fraud is when the system spots the same device with the same 1,380 characteristics making 40 different calls to access 40 different accounts at the same time.

SEE: Special report: A winning strategy for cybersecurity (free PDF)

Preventing call center fraud is the most common use case for the platform, but Balasubramaniyan also sees a growing need for voice authentication, spanning everything from home automation systems to corporate fraud. A recent case of voice fraud was a popular one at RSA 2020. Many people described the story of European executive who transferred $243,000 to a thief because he thought his boss had told him to do so.

“If it’s the CEO’s phone number and it sounds like your CEO’s voice, you’re going to do that or lose your job,” Balasubramaniyan said.

How it works

Amy Reyes was at the company’s RSA 2020 booth to show off the fraud detection technology. On the dashboard, a phone call is scored on three factors: device, voice, and behavior. Each characteristic gets a score which is then combined for an overall risk score. This analysis happens within the first few seconds of a call when the individual is interacting with automated system. Many Pindrop customers are call centers using the platform to identify and isolate fraud.

Reyes said some clients route calls with a high risk score to a voice mail box.

“You don’t even bother call center agents with bad callers,” she said.

The platform analyzes how an individual interacts with a device, including how quickly a person enters a piece of information, such as the last four digits of a Social Security number. 

“The speed is very different when a fraudster is typing from a list of stolen numbers or a robotic process that goes very quickly,” Balasubramaniyan said.

The platform also analyzes the acoustics of each phone to identify the make and model and the line noise which is different for each carrier.  

Balasubramaniyan said that early potential customers would ask Pindrop to identify cases of fraud in a batch of calls from the previous six months. Because the company’s platform isolates particular calls, it’s easy to pinpoint which interactions resulted in the bank losing money. 

“In this case, I know exactly what the person was doing on the other end, so I can attribute a dollar value to the fraud,” he said. “When they go back and investigate the account, they can see that the bank lost the money.”

Deep learning for better personalization

The platform uses deep learning to crunch all 1,380 data points and spit out a risk score.

Balasubramaniyan said the high volume of calls that Pindrop analyzes every year creates a big enough data set to train the model.

Mark Horne, the company’s chief marketing officer, said that some customers want to use Pindrop’s voice analysis capabilities to build better customer experiences. He also sees the technology as a way to make voice communication more secure in devices like Alexa and Siri.

“We see the future of these voice devices being mainstream, but authentication at the voice level is what’s required to make that happen,” he said.

Balasubramaniyan completed a PhD at Georgia Institute of Technology and wrote his dissertation on how to analyze the traits of phone calls. One of his professors connect him with a local entrepreneur and the three men co-founded Pindrop. The company built a phoneprinting fraud detection system for call centers. Balasubramaniyan got his first patent for the technology in 2015.

Also see

pindrop-rsa.jpg

Mark Horn, Amy Reyes, and Vijay Balasubramaniyan showed off the security platform that uses voice analysis to detect fraud.

Image: Veronica Combs



Source link