7 Ways AI Reduces Mobile Fraud Just In Time For The Holidays
- There has been a 680% increase in global fraud transactions from mobile apps from October 2015 to December 2018, according to RSA.
- 70% of fraudulent transactions originated in the mobile channel in 2018.
- RSA’s Anti-Fraud Command Center saw phishing attacks increase 178% after leading banks in Spain launched instant transfer services.
- Rogue mobile apps are proliferating with, 20% of all reported cyberattacks originating from mobile apps in 2018 alone.
On average, there are 82 new rogue applications submitted per day to any given AppExchange or application platform, all designed to defraud consumers. Mobile and digital commerce are cybercriminals’ favorite attack surfaces because they are succeeding with a broad base of strategies for defrauding people and businesses.
Phishing, malware, smishing, or the use of SMS texts rather than email to launch phishing attempts are succeeding in gaining access to victims’ account credentials, credit card numbers, and personal information to launch identity theft breaches. The RSA is seeing an arms race between cybercriminals and mobile OS providers with criminals improving their malware to stay at parity or leapfrog new versions and security patches of mobile operating systems.
Improving Mobile Fraud Prevention With AI And Machine Learning
Creating a series of rogue applications and successfully uploading them into an AppExchange or application store gives cybercriminals immediate access to global markets. Hacking mobile apps and devices is one of the fastest-growing cybercriminal markets, one with 6.8B mobile users worldwide this year, projected to increase to 7.3B in 2023, according to The Radicati Group. The total number of mobile devices, including both phones and tablets, will be over 13B by the end of 2019, according to the research firm. And a small percentage of mobile fraud transactions get reported, with mobile fraud losses reported totaling just over $40M across 14,392 breaches according to the U.S. Federal Trade Commission. Mobile fraud is an epidemic that needs to be fought with state-of-the-art approaches based on AI and machine learning’s innate strengths.
Traditional approaches to thwarting digital fraud rely on rules engines that thrive on detecting and taking action based on established, known patterns, and are often hard-coded into a merchant’s system. Fraud analyst teams further customize rules engines to reflect the unique requirements of the merchants’ selling strategies across each channel. Fine-tuning rules engines makes them effective at recognizing and taking action on known threat patterns. The challenge for every merchant relying on a fraud rules engine is that they often don’t catch the latest patterns in cybercriminal activity. Where rules-based approaches to digital fraud don’t scale, AI, and machine learning do.
Exploring The 7 Ways AI Is Reducing Mobile Fraud
Where rules engines are best suited for spotting existing trends in fraud activity, machine learning excels at classifying observations (called supervised machine learning) and finding anomalies in data by finding entirely new patterns and associations (called unsupervised machine learning). Combining supervised and unsupervised machine learning algorithms are proving to be very effective at reducing mobile fraud. The following are the seven ways AI and machine learning are reducing mobile fraud today:
- AI and machine learning reduce false positives by interpreting the nuances of specific behaviors and accurately predicting if a transaction is fraudulent or not. Merchants are relying on AI and machine learning to reduce false positives, saving their customers from having to re-authenticate who they are and their payment method. A false positive at that first interaction with a customer is going to reduce the amount of money that they spend with a merchant, so it’s very important to interpret each transaction accurately.
- Identifying and thwarting merchant fraud based on anomalous activity from a compromised mobile device. Cybercriminals are relying on SIM swapping to gain control of mobile devices and commit fraud, as the recent hack of Twitter’s founder Jack Dorsey illustrates. Hackers were able to transfer his telephone number using SIM swapping and by talking Dorsey’s mobile service provider to bypass the account passcode. Fortunately, only his Twitter account was hacked. Any app or account accessible on his phone could have been breached, leading to fraudulent bank transfers or purchases. The attack could have been thwarted if Jack Dorsey’s mobile service provider was using AI-based risk scoring to detect and act on anomalous activity.
- AI and machine learning-based techniques scale across a wider breadth of merchants than any rules-based approach to mobile fraud prevention can. Machine learning-based models scale and learn across different industries in real-time, accumulating valuable data that improves payment fraud prediction accuracy. Kount’s Universal Data Network is noteworthy, as it includes billions of transactions over 12 years, 6,500 customers, 180+ countries and territories, and multiple payment networks. That rich data feeds Kount’s machine learning models to detect anomalies more accurately and reduce false positives and chargebacks.
- Combining supervised and unsupervised machine learning algorithms translates into a formidable speed advantage, with fraudulent transactions identified on average in 250 milliseconds. Merchants’ digital business models’ scale and speed are increasing, and with the holidays coming up, there’s a high probability many will set mobile commerce sales records. The merchants who will gain the most sales are focusing on how security and customer experience can complement each other. Being able to approve or reject a transaction within a second or less is the cornerstone of an excellent customer buying experience.
- Knowing when to use two-factor authentication via SMS or Voice PIN to reduce false negatives or not, preserving customer relationships in the process. Rules engines will often take a brute-force approach to authentication if any of the factors they’re tracking show a given transaction is potentially fraudulent. Requesting customers authenticate themselves after they’re logged into a merchant’s site when they attempt to buy an item is a sure way to lose a customer for life. By being able to spot anomalies quickly, fewer customers are forced to re-authenticate themselves, and customer relationships are preserved. And when transactions are indeed fraudulent, losses have been averted in less than a second.
- Provide a real-time transaction risk score that combines the strengths of supervised and unsupervised machine learning into a single fraud prevention payment score. Merchants need a real-time transaction risk score that applies to every channel they sell, though. Fraud rules engines had to be tailored to each specific selling channel with specific rules for each type of transaction. That’s no longer the case due to machine learnings’ ability to scale across all channels and provide a transaction risk score in milliseconds. Leaders in this area include Kount’s Omniscore, the actionable transaction safety rating that is a result of their AI, which combines patented, proprietary supervised and unsupervised machine learning algorithms and technologies.
- Combining insights from supervised and unsupervised machine learning with contextual intelligence of transactions frees up fraud analysts to do more investigations and fewer transaction reviews. AI and machine learning-based fraud prevention systems’ first contribution is often reducing the time fraud analysts take for manual reviews. Digitally-based businesses I’ve talked with say having supervised machine learning categorize and then predict fraudulent attempts is invaluable from a time-saving standpoint alone. Merchants are finding AI, and machine learning-based approaches enable to score to approve more orders automatically, reject more orders automatically, and focus on those gray area orders, freeing up fraud analysts to do more strategic, rewarding work. They’re able to find more sophisticated, nuanced abuse attacks like refer a friend abuse or a promotion abuse or seller collusion in a marketplace. Letting the model do the work of true payment fraud prevention frees up those fraud analysts to do other worth that add value.
With the holiday season rapidly approaching, it’s time for merchants to look at how they can protect mobile transactions at scale across all selling channels. AI and machine learning are proving themselves as viable replacements to traditional rules engines that rely on predictable, known fraud patterns. With 70% of fraudulent transactions originating in the mobile channel in 2018 and the influx of orders coming in the next three months, now would be a good time for merchants to increase their ability to thwart mobile fraud while reducing false positives that alienate customers.
RSA 2019 Current State of Cybercrime Report (11 pp., PDF, opt-in)
The Radicati Group, Mobile Statistics Report, 2019 – 2023 (3 pp., PDF, no opt-in)
U.S. Federal Trade Commission, Consumer Sentinel Network, Data Book 2018 (90 pp., PDF, no opt-in)