\tLIVE FROM WEMEET EUROPE 2019, CASCAIS, PORTUGAL: The potential for machine learning and AI to tackle one of the telecoms industry\u2019s prevailing problems, subscription fraud, was highlighted by operators and other stakeholders.<\/p>\n
In a panel session, Rui Horta, head of the fraud and security department for Angolan operator Unitel (pictured, centre), was clear on why subscription fraud is, and will continue to be, a key challenge for the industry. \u201cSubscription fraud will always exist, because it\u2019s easy to do. And because normally it\u2019s the gateway for other things.\u201d<\/p>\n
A range of factors are playing a role in enabling subscription fraud, including availability of the personal data of genuine customers (whether current or previous) following well-publicised security breaches at companies across a range of sectors. This can cause problems for detection systems.<\/p>\n
Josie Harty, fraud and revenue assurance system optimisation manager for Virgin Media UK (pictured, far left), said: \u201cWith machine learning, if you had that data in the system, it could mark that as safe, because historically it would have been safe. So it\u2019s all about the data you feed into the machine learning and making sure it\u2019s up to date constantly.\u201d<\/p>\n
Parthiban Balasubramanian, telemetry product owner at Canada-based Telus (pictured, second from left), noted the actions of customers themselves can play into the hands of fraudsters: \u201cOn social media, it\u2019s easy to get information\u201d, he said.<\/p>\n
Vulnerabilities
\nOther challenges in addressing subscription fraud come from internal sources, with Harty noting weak points such as outsourced customer centres, where staff turnover is high; and retail, where a large number of staff are employed, including many on a more casual basis.<\/p>\n
Balasubramanian said bad actors \u201cknow all the internal secrets: once inside the system, how to get around and what to do\u201d.<\/p>\n
Increased digitalisation of services also means it is becoming more common, and easier, to buy and activate services online without human involvement. While this offers real benefits for genuine customers, it also creates another point of attack for fraudsters.<\/p>\n
Harty explained: \u201cWith that comes a shortening window of opportunity for us to actually screen the orders, so that\u2019s where machine learning comes in, to screen more data.\u201d<\/p>\n
Tech challenges
\nWhile machine learning and AI provide essential tools to address such challenges, they also come with issues of their own.<\/p>\n
Carlos Martins, RAID cloud product owner at WeDo Technologies (pictured, far right), noted operators’ ability to share data on fraud is being limited by regulation such as GDPR, which places restraints on data use.<\/p>\n
With fraudsters often using the same techniques across operators once they have proven to be successful, data sharing could be useful in enabling timely detection.<\/p>\n
\u201cSharing the information would be very beneficial to all of the operators, but at the same time hiding everything that is sensitive information would probably remove the critical data which we need for the machine learning data,\u201d Martins said.<\/p>\n
But even without sharing, operators do have access to large amounts of data, which is critical for AI and machine learning. \u201cOperators have historical data for ten to 15 years or more, which the fraudsters don\u2019t,\u201d he said. On the flip side, \u201cwe have seen evidence that fraudsters are using machine learning models.\u201d<\/p>\n
Balasubramanian summed up why machine learning and AI are critical for operators in the subscription fraud battle: \u201cWe are already way behind what the fraudsters are doing, this is the only way for us to go faster, to get up to the speed.\u201d<\/p>\n","protected":false},"excerpt":{"rendered":"
LIVE FROM WEMEET EUROPE 2019, CASCAIS, PORTUGAL: The potential for machine learning and AI to …<\/p>\n
The post AI takes key role in tackling fraud<\/a> appeared first on Mobile World Live<\/a>.<\/p>\n