In addition to the bump in B2C online shopping, B2B ecommerce accelerated, as did the importance and availability of marketplaces and B2B2C opportunities. According to the U.S. International Trade Commission, the share of online retail sales rocketed from 13.6% in 2019 to 18% in 2020. That trajectory is expected to continue - it’s projected that ecommerce will command nearly 22% of global retail sales by 2024.
Despite the massive growth in ecommerce sales, retail has always had a dark shadow following it. Fraud follows the money, and as consumers spend more online, bad actors see greater opportunities to target retailers’ vulnerabilities. Just as ecommerce sales are expected to rise, global losses to fraud are expected to exceed $48 billion by the end of 2023.
Criminals are making good use of technology advancements to discover and accelerate complicated fraud schemes. Digital tools make it possible to test thousands of accounts in seconds, making Account Take Over (ATO) attacks easier than ever before. Fraud tactics are changing and evolving at an unprecedented rate - often too fast for simple filters or manual recognition and review to keep up with.
Keeping up - or keeping ahead - of fraudsters requires being able to process huge amounts of data quickly, pull in information from disparate sources, and recognize patterns at scale. As AI moves from the realm of science fiction to day-to-day business operations, this new technology is the perfect solution to help identify, detect, and prevent fraud for ecommerce enterprises.
Why traditional fraud methods no longer work
Ecommerce fraud isn’t new. What is new is the ways bad actors are discovering vulnerabilities in online shopping experiences, using digital toolsets and even AI, to help them exploit fraud opportunities at scale.
In an attempt to keep up with the onslaught of fraud attempts, companies are spending millions, and still falling short. Worse, the customer impact of poorly applied prevention methods can drive customers away. Instead of boosting revenue and enhancing the customer experience these strategies can end up hurting a business’s bottom line.
How traditional fraud prevention works
Historically, online fraud prevention and detection went something like this:
- A fraud team would determine the kinds of activities that typically indicated fraud.
- These activities would be scored and programmed into a filter.
- That filter would pass through transactions that scored low enough to be safe, and flag for review or decline transactions that scored high enough to be of concern.
- Transactions flagged for review would be passed to a fraud team member, where a determination would be made regarding the validity of the transaction, and either be allowed or rejected.
In truth, this basic pattern is similar to how transactions are handled today. But the old way of doing this presents a lot of challenges as the volume of transactions grow and the complexity of fraud methods increase. Here’s why traditional fraud methods need an innovative boost with AI:
Traditional fraud methods rely on manual management
The seismic shift to online shopping has been a boon for companies. At the same time, it’s created a dilemma for fraud teams. Even for a small business, there are too many transactions to reasonably review manually. For an enterprise with potentially hundreds or thousands of transactions an hour, it’s simply unrealistic.
The filters used to sort transactions into “good” and “bad” can be rigid and non-adaptive. Plus, spotting new trends in fraud attempts may require the review of massive amounts of data. A fraud scheme could be operating on a business for months before a pattern is discovered.
In-house fraud teams can be expensive
Fraud experts and analysts are able to spot bad transactions and trends faster than someone with no training or experience. It’s not feasible, though, to scale up a fraud team to identify and manage the ever-increasing number of fraud attempts, for a number of reasons.
First, there are simply more fraudsters in the world than the number of fraud analysts on your team, even in a large company. Second, talented and experienced fraud analysts are hard to find. An organization’s HR team could spend a significant amount of time finding just one analyst, drawing HR and recruiting members away from other hiring needs.
And even if you could find enough analysts to keep up with fraud attempts, would your OpEx budget allow you to do so? According to talent.com, an entry-level fraud analyst's salary starts at around $50,000 a year, whereas an experienced analyst earns an average of close to $125,000 annually. A team of just ten analysts, on average, would then cost more than half a million dollars in salaries a year. It’s a model that simply doesn’t scale.
Rules-based fraud prevention can do more harm than good
Certainly, rules-based filters can help cut down on the number of transactions that must be manually reviewed. Unfortunately, those rules must be created and scored by a human, who may not see an emerging trend or underlying threats.
Of course, a conservative strategy could be to err on the side of rejecting suspicious purchases. That can be a costly mistake as research shows that for every $1 lost to fraud, companies lose $13 to false declines. And further data shows that nearly 90% of declined purchases are actually legitimate transactions.
How AI works for fraud detection and prevention
What today’s ecommerce companies need are fraud solutions that have the ability to evaluate millions of transactions quickly, pulling in data from a variety of sources, while evolving risk models quickly to keep up with criminals. It’s a tall order - but one that fraud solutions that incorporate AI can solve.
The Two Sides of Artificial Intelligence
To grasp how AI works to detect, prevent, and spot evolving trends, it’s important to first understand that what we call Artificial Intelligence is composed of two parts - machine learning and AI. Both are important in ecommerce fraud protection.
Machine learning (ML): Machine learning is the learning part of what we collectively call AI. With ML, data is ingested, evaluated, segmented, and categorized to understand patterns and behaviors. This can happen in a supervised manner, where data is labeled for the model by an analyst or data scientist, or unsupervised, where the model is developed without labeled data, and the learning process can be ongoing without intervention. In fraud detection, it’s ideal to use both when possible.
Artificial Intelligence (AI): The artificial intelligence portion of AI is the application of the model that is developed with ML. Basically, ML defines the rules for transaction evaluation, and AI is the part that looks at the incoming transactions and decides if they are valid, invalid, or need further investigation.
The advantage that AI brings to the detection and prevention of ecommerce fraud is speed and scale. Even an experienced analyst will take time to review and evaluate an individual transaction. Analysis of fraud data is a heavier lift, with new patterns only emerging to human experts over days, weeks, or even months and years. Time restrictions also mean that the data points that can be explored by a human are limited.
AI, however, can evaluate thousands of transactions in seconds, and, most importantly, continue to learn on its own so that new threats can be identified and added to the evaluation model in near real time. Plus, the data used to feed the machine learning model can be expansive - in many cases, the more data, the better.
Even non-financial data - things like a recent move, recent return patterns, and password reset requests - can all be factored into the AI’s decision-making as it examines transactions.
The benefits of AI for fraud prevention and detection
Clearly, speed, scale, and the ability to continuously evaluate and learn from real transactions are the major benefits of using AI for fraud detection. However, the practical benefits of AI usage also present a compelling case for AI in ecommerce fraud management.
Reduces costs
AI can help enterprises reduce costs, and not just from loss prevention. Chargebacks have become a major concern for ecommerce companies. For every dollar in chargebacks, companies lose $2.50 or more on time handling disputes, lost physical merchandise, processor fees, and shipping costs. It’s expected merchants will pay more than $100 million in chargebacks in 2023 alone.
With enough chargebacks, companies may get listed as high risk with their processors, meaning even regular credit card transactions will carry a hefty processing fee. AI makes it easier and faster to recognize the kinds of transactions that lead to chargebacks.
Spots “friendly-fraud” faster
Friendly fraud occurs when a customer - perhaps even a good customer - makes a purchase, receives an item, but then reports to their credit card company that it never arrived, arrived damaged, or didn’t match the store’s description. Why do they do this? Maybe they don't want to go through the hassle of returning the item, or they get a case of buyer's remorse, or a million other reasons–the fact of the matter is that credit card companies make the chargeback process so easy for consumers that it has become a commonplace occurance.
This isn’t a small problem, either. In a recent study, 23% of consumers admitted to committing friendly fraud - and those that do so are likely to try it again in the months following a successful chargeback.
AI can quickly pull in and evaluate data related to the current transaction, and even include non-financial data. That means friendly fraud can be detected more reliably and nipped in the bud before it becomes a problem.
Maximizes sales and minimizes losses
One of the great misconceptions in ecommerce is that erring on the side of conservative fraud prevention is an effective risk mitigation method. In essence, some believe that declining the occasional good transaction is better than letting a bad one through.
Unfortunately, the data doesn’t support that. False declines lead to lost sales, and not just the immediately declined ones. Thirty-three percent of consumers reported that they would stop shopping with a retailer after a false decline of their method of payment.
AI can quickly learn from shifting conditions and take other information into account, too. For instance, at the start of the pandemic, many consumers shopped online for the first time - a traditional red flag for fraud filters. However, these were legitimate purchases, and AI can adapt quickly to changes in first-time e-commerce shopper behavior, where fraud filters remain rigid.
Improves the customer experience
The less friction there is between the cart and the sale in an online shopping experience, the simpler it is for a consumer to make a purchase. Asking for tons of information, holding up purchases for investigation, or declining a legitimate purchase are all detrimental to conversions.
With AI monitoring sales for fraud, data can be gathered and processed from multiple sources in seconds, and nuanced decisions can be made without asking the consumer to enter more information or risk upsetting a good customer by turning away a legitimate sale. This increases safe approvals of transactions and shortens the distance between “I want” and “I bought”.
AI fraud prevention and detection is the future of e-commerce
There was a time when AI was inaccessible to even large enterprises for tasks like e-commerce prevention. As computing power has grown, data storage costs have reduced, and advancements in AI have accelerated, advanced digital intelligence has become a realistic tool to fight online fraud.
The benefits of leveraging this technology - from improved customer experiences to the ability to rapidly scale and adapt - make the adoption of AI an easy choice. When evaluating fraud prevention and detection solutions, look at the range of tools the platform employs to protect your customers and your business, including AI, advanced analytics, and human expertise. As fraudsters continue to evolve their techniques, a combination of prevention and detection tools will help minimize losses while improving revenue.
###
Rafael Lourenco is Executive Vice President and Partner at ClearSale and holds more than two decades of experience providing e-commerce fraud detection and prevention services in major international markets. Rafael combines ClearSale’s innovation-driven culture and emphasis on communication with a deep understanding of the statistical tools that underpin excellent fraud protection. Follow ClearSale on LinkedIn, Facebook, Instagram Twitter @ClearSaleUS, or visit https://www.clear.sale.