Can Airtel really curb spam/scam calls using AI?

The new feature is automatically enabled for both prepaid and postpaid Airtel subscribers at no additional cost. Smartphones aside, the telco has said that it is still working out how to enable the AI-powered spam detection capability for feature phones.

Airtel’s latest move follows a series of not-so-successful attempts by telecom operators, tech companies, and regulatory authorities to curb the menace of pesky calls and SMSes through various techniques, including setting up a Do-Not-Disturb (DND) registry.

“We have spent the last twelve months to solve this comprehensively. Today marks a milestone as we launch the country’s first AI-powered spam free network that will shield our customers from the continuous onslaught of intrusive and unwanted communications,” Gopal Vittal, managing director and CEO of Bharti Airtel, said at a press briefing in Delhi on Wednesday.

But, what is Airtel doing that hasn’t been tried before? Is AI the solution for Indians to finally overcome the nuisance of unwanted calls and texts?

How Airtel will detect spam/scam calls

Essentially, Airtel is relying on subscribers to not pick up scam calls or click on dangerous links in SMSes marked as Suspected Spam. As more and more subscribers avoid falling for such tactics, it hopes that the frequency of such fraudulent communication will drop.

“The user will see Suspected Spam, and the moment they see that, in my view, they are fully protected,” Vittal said.

The AI spam filters cover incoming calls and SMSes from any Airtel subscriber as well as any other telecom operator. (Image: Karan Mahadik/The Indian Express)

In order to detect them, Airtel said every call and SMS will pass through two spam filters: one at the network layer and the second at the IT systems layer. The company claimed its proprietary AI algorithm will analyse 250 different parameters related to calls in around 2 milliseconds.

Some of these indicators that will be studied by AI include velocity of calls, frequency of device change, distinct locations to which calls are made, numbers of calls not answered, DND complaints, incoming and outgoing call trends, volume of calls made, frequency of change of SIM cards, location of purchase, number of very-short duration calls, robocalling devices, low age on the network, number of SIM cards registered on the same KYC, inputs from cyber cells, and more.

“By cross-referencing this information against known spam patterns, the system flags suspected spam calls and SMSes accurately,” Airtel said.

To alert customers about malicious links received via SMS, the telco said that its AI model will cross-check links against a centralised database of blacklisted URLs while detecting frequent changes of IMEI, a 15-digit number that is unique to every device. In response to privacy concerns, Vittal clarified that the AI algorithm would not need to read the content of SMSes and mainly look for markers of spam.

Based on its testing, Airtel said that the AI model currently has a 97 per cent accuracy rate in spam calls and 99.5 per cent for spam SMSes. “In the next four to six weeks, we will see the AI model bridge the 3 per cent gap in accuracy further as it is a learning model,” Vittal said.

How it is different from apps like Truecaller

Vittal was confident that Airtel’s AI spam filters were a better solution than Truecaller as such crowdsourcing apps rely on user feedback to detect spam and need internet connectivity to work. “Even if you’re on Truecaller, the calls are not being identified as Suspected Scam so you don’t even know. In our case, it will be detected in real-time,” he said. Vittal said that Calling Name Presentation or CNAP proposed by the Telecom Regulatory Authority of India (TRAI) will not work either as many subscribers may not have the SIMs registered in their names.

However, Vittal also acknowledged that the AI spam filters come with certain limitations. For starters, the telco will not be currently blocking the spam/scam calls because three per cent of calls may come from businesses such as Swiggy that could show up as spam even though it is genuine.

A few days ago, the Airtel CEO had penned a letter to the heads of rival telecom companies such as Jio, Vi, and BSNL about the need to share data on B2B subscribers in order to distinguish between spam and legitimate calls from businesses.

Since the AI spam filters are only applied to the Indian network, the proposed measure is not likely to deter the rising number of international spam/scam calls. Recent data shows that over 85,000 scams targeting Indians originated from Southeast Asian countries this year so far. Furthermore, OTT communication will not be covered. “We can’t do anything about spam on OTT apps. For that, they have to be brought into some kind of regulation,” Vittal said.

Can AI help squash spam/scam calls and SMSes?

In June this year, TRAI had directed all telecom service providers to enable AI/ML spam detection systems in a bid to curb unsolicited communication from unregistered telemarketers (UTM). It specifically instructed telcos to devise systems capable of evolving constantly to deal with new signatures, new patterns, and new techniques used by UTMs, after a similar directive did not yield expected results.

Experts have also previously pointed out that AI spam detection mechanisms may fall short without serious consequences for repeat offenders. Recently, TRAI floated a discussion paper proposing differential tariffs for subscribers making over 50 calls or sending out 50 SMSes a day.

When asked if spammers identified through Airtel’s AI model will be flagged to regulators as well, Vittal said, “We are open to doing it. We are open to sharing data where there is legitimate scam or spam. This solution tries to stop scams by making sure that the calls don’t happen because every scam begins with that call.”