Over the last few years, the global eCommerce market has been growing by leaps and bounds, accounting for 16% of total retail sales today — a 6% growth since 2017. Despite the coronavirus disruption sweeping through all industries, the online shopping sector is holding strong, recording an average transaction volume growth of 23.5% year-to-year.
However, as retail and eCommerce are changing, so is the risk landscape. Increasing sales volumes make fertile ground for fraud and abuse. In the new reality dominated by data, traditional, manual anomaly and fraud detection have become inadequate. To get a grip on modern threats and recover lost margins, retailers — both online and traditional — must tighten security with AI-driven tools.
Retail eCommerce sales worldwide by regions, 2020, in billions
What is fraud in retail and eCommerce?
Following the convergence of on-site and online shopping, fraudsters have successfully exploited the resultant vulnerabilities, finding new ways of attacking security at every stage of retailers’ interaction with buyers. Online stores are especially at risk, facing an overwhelming workload of transactions, orders, and deliveries as a result of an intensifying consumer shift away from in-store purchases towards online buying.
From “traditional” methods like unauthorized card usage and charge-backs, to mobile hacking and affiliate marketing abuse, retail businesses need to be primed and ready to face even the most elusive of threats.
Traditional retailers are mainly exposed to inventory shrink, false cash returns, and credit card fraud. In eCommerce, as a recent industry report reveals, most losses are incurred as a result of unauthorized card transactions and account takeovers. Still, the array of retail threats is constantly expanding, and fraud patterns are becoming increasingly elaborate.
eCommerce fraud in numbers
Types of eCommerce and retail fraud
Identity theft and account takeover
Identity theft is one of the most common fraudulent transactions, both in traditional and online retail. It occurs when a criminal takes over a person’s ID, credit card, or bank account, and makes illegal purchases. This type of crime often targets millennials, who readily use social media logins to validate their identity. Consumers who use convenient, weak passwords that are easy to guess are another frequent victim of identity theft. Currently, most online thieves use bots to gain access to their victims’ details, instead of trying to compromise that information on their own.
A charge-back occurs when a customer is trying to have money returned from a legitimate purchase, when they should be applying for a refund instead. Although many charge-backs are considered ‘friendly fraud’ and aren’t a result of a malicious attempt to extort money, the consequences for retailers are equally damaging irrespective of the buyer’s intentions. Each case of charge-back cuts into the merchant’s profit margin, and may lead to lost merchandise and high penalty and processing fees. The cost of friendly fraud is estimated to reach $50 billion by the end of 2020.
The cost of friendly fraud is estimated to reach $50 billion by the end of 2020.
Affiliate fraud is related to affiliate marketing programs. These programs work by providing a website owner or an influencer with a commission for endorsing a brand and encouraging followers to visit the brand’s site. Affiliate fraudsters game the system to fraudulently collect the commission — for example by using clicking farms to generate fake traffic and simulate buyer interest.
Also known as rewards fraud, loyalty fraud refers to the misuse of a company’s rewards program to steal unearned points. On the surface, this activity seems harmless, as it doesn’t directly involve monetary transactions. In reality, about $1 billion gets stolen each year as a result of illegally redeemed gifts and vouchers. 72% of program managers have experienced issues with fraud.
72% of program managers have experienced issues with fraud.
Discount abuse bears some resemblance to loyalty fraud. Here, violators abuse a company discount program — for example, by repeatedly using the discount code or taking advantage of someone else’s discounts. Notably, this type of fraud is popular among company employees. They exploit the system by buying massive amounts of discounted merchandise and reselling them with a profit. Another form of discount abuse is when employees share their coupons with friends and family who aren’t authorized to use them.
Inventory stock discrepancies in stores have many root causes, such as shoplifting, ticket switching, or clerical errors. Unfortunately, quite often, these are caused by employee and supplier actions. Dishonest employees recognize how difficult it is to keep an accurate count of inventory stock by hand, and they take full advantage. While manual stock-taking creates plenty of opportunities for pilfering, inexpensive deep learning analytics systems make it extremely easy to detect and prevent stock shortages in real-time and reduce POS inventory loss. Discover how fashion retailers overcome inventory management challenges with data analytics software
Discover how fashion retailers overcome inventory management challenges with data analytics software
As the volume of mobile transactions rises, fraud attempts become inevitable. Mobile fraud remains a blind spot for many eCommerce sellers. Fraudulent activities over mobile devices take many shapes and forms, including account takeovers, phishing, and charge-back. All of them impact merchants directly or indirectly, through declined transactions, increased abandonment rates, and tarnished reputations. A new mobile phishing page launches every 20 seconds, which means that more than 4,000 phishing sites appear each day.
A new mobile phishing page launches every 20 seconds, which means that more than 4,000 phishing sites appear each day.
How AI and ML help detect and prevent retail anomaly and fraud
Malicious and fraudulent activities can ruin businesses within a few weeks. Therefore, retail companies are on the constant lookout for methods that allow them to isolate patterns of behavior that indicate risks. 36% of eCommerce businesses identify fraud as the greatest risk to their prosperity.
36% of eCommerce businesses identify fraud as the greatest risk to their prosperity.
However, the “old” methods of blocking payments or transactions are grossly inefficient. One of the major problems encountered by retailers who keep relying on obsolete fraud prevention tools is the generation of a massive number of false positives, or identifying an honest customer as a fraudster. As a result, valid transactions are declined, preventing the buyer from making a legitimate purchase. False positives erode customer trust, leading to lost sales and lost clients.
Another issue with legacy fraud detection solutions, even those utilizing big data, is that they mostly provide retailers with retrospective analytics to detect already known, systematic patterns. In contrast, modern AI solutions for retail apply predictive algorithms to proactively assess vulnerabilities and raise the red flag on suspicious activities before they even occur. These systems help retailers not only by identifying the existing weak spots in security but also by anticipating issues and preventing new types of attacks.
This is how AI-based anomaly analytics solutions work. You feed them data, specify fraud signals, train algorithms to make predictions and recognize fraud, and create a model specific to your retail business and use cases.
AI fraud and anomaly detection
AI fraud detection systems for retail work by analyzing immense volumes of historical and current transaction data to understand buyer motivations and identify anomalies (suspicious activities that deviate from the norm).
By applying deep learning and AI algorithms, eCommerce fraud solutions can extract thousands of attributes out of each transaction and analyze millions of data points to keep your sales secure. This way, they divulge hidden patterns that a human would have found impossible to work out. Depending on the established protocols, when an anomaly is detected, AI-driven anomaly analytics platforms can either block a user or a transaction, or send a notification to an employee who will investigate the activity further.
Sophisticated deep-learning anomaly-detection engines are capable of self-learning, continuously educating themselves, and optimizing algorithms based on past transactions. These retail fraud prevention systems are becoming more accurate and powerful with every single transaction they examine.
Supervised and unsupervised retail fraud detection
There are two types of anomaly analytics that AI and ML engines can handle:
- Supervised anomaly detection – Supervised anomaly detection trains an ML model based on labeled historical fraud data. To detect anomalies, the deep learning engine is first presented with huge sets of data that represent normal and abnormal transactions. Then, based on this previously identified information, the system learns how to classify and recognize fraud in the future, with great accuracy.
- Unsupervised anomaly detection – Unsupervised anomaly detection, on the other hand, works with “unlabeled” data. It independently looks for data set instances that seem abnormal based on a multitude of parameters. Unsupervised machine learning systems are less accurate when it comes to the identification of well-known types of fraud. Yet, they excel at discovering new and unknown fraud patterns.
What types of fraud can eCommerce fraud solutions deal with?
Highly adaptable ML and AI anomaly detection systems can serve virtually all types of use cases. Thanks to a comprehensive analysis of mass data in various contexts, they can identify non-standard patterns that are invisible for legacy fraud prevention systems, helping retailers find and prevent fraud in physical stores and online. Here are a few examples:
Fake IDs and credit cards
In the traditional retail setting, fraud prevention systems use artificial intelligence and its subsets of machine learning to examine hundreds of parameters of an ID or a credit card. These include paper and ink validation, facial recognition and biometrics, and security threads. In the eCommerce context, a wide array of methods is used by these AI solutions for detecting anomalies, such as comprehensive customer profiling, for example.
Promo code and loyalty program abuse
Retail fraud detection software helps retailers spot the fraudsters trying to trick the business by redeeming a coupon or loyalty points multiple times. AI-driven fraud detection for retail also works in other situations where violators set up several accounts or use proxy servers to make illegal purchases and extort money.
Machine learning and AI anomaly detection systems automate the transaction validation process, providing the fastest and most accurate risk assessment. By analyzing various risk identifiers and reviewing buyer interactions based on customized policies, they prevent fraudulent charge-back transactions and approve valid orders.
Retail inventory loss
There are numerous ways in which AI-based retail solutions can help alleviate the financial impact of inventory loss. One example is when an intelligent fraud prevention system uses computer vision to capture videos of items that haven’t been scanned or have an incorrect barcode, and raises an alarm at the checkout to verify the transaction.
Deep learning algorithms are an essential ingredient of a multi-layered defense against mobile eCommerce fraud. They provide a real-time transaction risk score that makes it possible to identify and prevent fraudulent attempts in a split second. ML tools analyze and interpret the nuances of specific behavior, scanning every user and transaction and triggering alerts when anomalies are detected. Maintaining customer trust is an essential ingredient of a successful retail business. Find out more ways in which retailers use machine learning to keep valuable clients
Maintaining customer trust is an essential ingredient of a successful retail business. Find out more ways in which retailers use machine learning to keep valuable clients
Arm your retail business with AI anomaly detection
With its unique ability to prevent fraudulent activities in real-time, artificial intelligence has become an essential weapon in the fight with retail fraud. AI-based anomaly analytics and detection solve many issues faced by modern retail and eCommerce businesses, replacing inefficient, rule-based manual reviews with advanced fraud detection algorithms that spot anomalies in a fraction of a second.
Yet despite AI’s proven track record in retail and eCommerce fraud prevention, many merchants are still slow in the adoption of the technology, as they are intimidated by its complexity. A team of experienced retail software development consultants will help them deploy AI fraud detection solutions one step at a time, starting from the most urgent use cases and slowly progressing to more advanced issues.
Would you like to tackle retail fraud before it strikes? Contact our team to learn which fraud-detecting AI application is the best fit for your eCommerce or retail business.