Frauds have always been present within the telecom industry. Fraudsters have invented numerous ways to use services for free, or on somebody else’s behalf. Their techniques vary from abusive service use, stealing IDs, credentials or hardware, to breaching physical core network security or copying devices. Each scenario has different anomaly patterns that need to be detected and updated as fraudsters tweak their methods.
This is where AI/ML techniques come to the rescue, detecting fraudulent activities and patterns quickly and autonomously.
Telco Fraud Management targets suspicious behavior and detects anomalies to prevent an array of fraud techniques to keep both you and your customers more secure.
Detection of service usage anomalies
One of the most popular cases of excessive usage activity is so called SimBox fraud. It consist in creating gateways which, on one hand, accepts the VOIP traffic, and on the other – uses several SIM cards to terminate the call locally, but outside of a CSP network.
SimBox is typically a “magic box” which can be pretty sneaky to avoid being detected by CSP as it causes excessive traffic on CSP’s SIM cards. Numerous techniques can be applied to prevent the fraud, starting with the simplest ones, like limiting number of unique called numbers in a month or limiting maximum connection time. But fraudsters are quickly adapting SimBox algorithms, so telcos must keep pace with these changes and use AI/ML techniques accordingly.
AI/ML is used to analyze service usage patterns to create different usage profiles so as to detect anomalies and trigger configurable actions. It uses historical data but at the same time continuously trains its models in order to adapt to changing conditions and modify alarm thresholds.
Fraud detection based on movement analytics (Asset Tracking)
By analyzing mobile subscriptions, it’s possible to detect numerous frauds, including the use of duplicated SIM cards and abusive roaming cases. This is extremely useful in the IoT space where large assets need to be supervised. The AI/ML engine continuously checks if the place and nature of the service usage don’t raise an eyebrow.
The engine is supported by a graphic presentation of the device movement paths right on a map. Potential frauds become even better visible (“why are my devices running in Australia?”)
By analyzing invoice content and amounts, combined with service usage patterns and historical data, it’s possible to detect potential frauds before the invoice is sent to the customer. All suspicious invoices will be detected, prompting billing teams to run checks and determine if there’s a fraud. All before the customer receives a wrong invoice.
Uncommon prepaid usage
Accumulation of credit on a prepaid account, recharging using stolen codes or “guessed” secret codes, fund redemption, numerous credit transfers, or manual balance updates are just a few examples of possible frauds using prepaid subscriptions. There are countless types of such frauds, and transaction monitoring addresses just a few of them. To achieve a significant advantage over fraudsters, one needs to compare data from numerous sources and use AI/ML to detect and analyze suspicious customer behavior.