TNS uses big data, machine learning to foil robocalls

Transaction Network Services (TNS) has been around for decades, and as one of the largest independent providers of inter-carrier call signaling and routing, it’s an established player in telecom. But it’s that long-time positioning that’s helping it compete in the wild, wild west of the robocall detection business.

This week, the company announced that its Neighbor Spoofing feature is enabling wireless carriers to protect their subscribers from the robocall tactic that uses local area codes or other means to make the consumer think the call is originating in their local area. The thinking is, if a call matches or closely matches their area code, they’re more likely to trust the call is real and pick up.

Carriers will use messages like “Potential spam” or “Likely spam” to let their customers know when a call is coming from a bad number so they don’t pick up. But these calls are getting so prevalent that some people won’t even pick up their phone—which isn’t good if the caller happens to be the local pharmacist or a doctor’s office. And it isn’t good for business if people increasingly don’t accept phone calls.

“Neighbor spoofing is more than a nuisance,” said Bill Versen, chief product officer at TNS, in a release. “If consumers stop answering calls from local numbers, they could miss an important call from a hospital, pharmacy, school, or loved one. With neighbor spoofing detection, carriers can offer even greater protection to subscribers.”

RELATED: AT&T leads in robocalls with 15.1 calls per customer in March

While over-the-top (OTT) robocall blocking apps are plentiful, they don’t have the same history or placement that TNS has established—embedded with carriers. It’s close ties with operators means it also has to live up to higher standards of reliability and accountability. In the U.S., TNS customers include Verizon Wireless, US Cellular, Sprint and Verizon’s wireline business.

“We have to be very, very careful,” Deirdre Menard, director of product management at TNS, told FierceWirelessTech. “We can’t cast a wide net and say, ‘well if we catch a lot of false positives, that’s no big deal because people are opting into the service.’ It’s not like an over-the-top service in that respect.”

The company has been working on the problem since it noticed a spike in neighbor spoofing more than a year ago. It made tweaks along the way, and in a recent study, Mind Commerce found the TNS Call Guardian correctly identified spoofed numbers 98% of the time during tests compared to 64% for the nearest competitor.

They’re not giving away their special sauce, but acknowledge a lot of time went into perfecting it—and likely more in the future as well. It’s like a game of whack-a-mole, said Jim Tyrrell, senior director of Product Marketing at TNS. The bad actors constantly change their behavior as detection systems improve.

“It’s constantly evolving and our methods and procedures are getting better,” he told FierceWirelessTech. “We’ve gotten to the point where we’ve got it down pretty good. Obviously there’s still opportunity for improvement.”

Robocalls of the illegitimate variety have been a big problem for a while. The FTC received 4.5 million robocall complaints in 2017, up over the prior year’s 3.4 million. The FCC recently put out a public notice asking for input for a report it's compiling; the deadline for reply comments was Aug. 20.

"We’ve got a lot of big data and it helps us figure these things out,” Tyrrell said.

The company analyzes 1 billion call events per day across more than 500 telecom operators to evaluate whether calls are legitimate. Because some robocalls are fine—Walgreen’s will use an auto-dialer to send out prescription refill reminders or schools will make calls if it’s a snow day—distinguishing between the two requires a lot of machine learning.

If a signaling event originates from a network that doesn’t own that number, it gets flagged. If it’s a deactivated number or unassigned number, that’s also flagged.

TNS is using a machine-learning based algorithm, but it also employs a team of about a dozen humans—data scientists who look at the data to find trends and feed that information back into the machine.

“We want to put some human intelligence into it too,” Tyrrell said.