At ClearSale, we get that as an online merchant, your goal isn’t first and foremost to stop fraud.
Yes, card-not-present fraud can take a big piece out of your revenue, and therefore, you want to minimize it. But your true objectives are to grow your sales, to turn satisfied customers into loyal shoppers. Fraud prevention is simply a means to an end, a tool to avoid tarnishing your reputation and incurring costly chargebacks.
You might envision a fraud protection solution as an overzealous hall monitor. You remember the type from middle school, scrutinizing bathroom passes down to the smallest detail, looking for the slightest justification to deny passage.
But at ClearSale, we don’t see ourselves that way. We look for ways to approve orders, not deny them, because we know sales drive your business.
It’s right there in our name. Our mission is to “clear the sale.”
We’re proud to have the highest approval rates and lowest false-positive rates in the industry. That means we’re doing what we set out to do: help e-commerce merchants maximize their revenue.
How do we stop genuine cases of fraud without turning into the mean hall monitor? Simply put, we combine the best of two powerful processes: proprietary AI technology and targeted expert fraud analyst reviews. In today’s article, we look under the hood to learn how it all works.
A Brief History of Fraud Prevention
ClearSale has spent the last decade-and-a-half innovating better ways to approve more sales, via a unique, multilayered, people-first approach. To understand how we got where we are, it might be useful to follow how fraud prevention techniques have evolved since the early days of e-commerce.
The first systems, which came on the scene about 20 years ago, were rules-based. In this context, a “rule” is a statement, translated into code, that determines whether fraud protection software will approve or deny a transaction.
For example, a very simple rule might say, “Every order under $50 should be approved.” A more complex rule would say, “Every order under $50, from the same email address, delivered to the same shipping address, should be approved.”
Rules-based filters can be very flexible; you can adapt to new trends quickly by adding rules. The downside is, rules pile up on top of each other. It doesn’t take long before you have 100 or even 200 rules.
Using our example from above, one rule might say, “Every order under $50 should be approved.” Another might say, “Every order shipping to a certain city should be reviewed manually.” It’s possible for an order to fall into both categories. Which rule gets priority?
To solve the problem of rules upon rules, companies began calculating fraud scores about ten years ago. Fraud scores weigh variables and calculate the likelihood of fraud based on the total weight of the variables associated with a specific order.
In the example above, the low price-tag might carry a weight of one point, while the risky delivery address might carry a weight of three points. A flagged IP address might add several more points to the total.
In a score-based system, the total score required for denial or manual review can be adjusted depending on a company’s tolerance for risk, market, and other factors.
Beating the System
Neither rules- nor score-based fraud filters are foolproof. Savvy criminals have learned their ins and outs and know how to exploit their weaknesses to mimic legitimate customers.
Here are some ways fraudsters get around some of the most common fraud filters:
- Address verification systems (AVS). Fraudsters match billing and shipping addresses on the first line of their orders but use the second line to provide their actual shipping address.
- Card verification values (CVV). Fraudsters steal physical credit cards, use keyloggers to capture information from consumers illicitly, or run small-scale tests on different values until they find the right one.
- Age and quality of email addresses. Fraudsters buy email addresses or use account takeover methods to hijack older, paid domain addresses.
- Velocity filters. Fraudsters place single large orders (instead of rapid-fire smaller ones) or cycle through e-commerce merchants.
- Purchase amount filters. Fraudsters use trial-and-error to discover the maximum amount they can buy without triggering filters.
- Blacklists of email or IP addresses. Fraudsters simply try another email address, IP address, or stolen card number.
While criminals have learned to circumvent most common automatic fraud filters, genuine customers have not. After all, why should they have to? But when a legitimate buyer’s order raises a red flag – for whatever reason – to an automated system, it can cause a false decline.
With all the hype about fraud, many e-commerce merchants don’t realize that false declines can cost them up to 13 times as much. False declines drive customers away and inspire poor online reviews and social media mentions. You might need to complete 12 or more transactions to compensate for the financial loss of even one false decline.
Rise of the Machines
The next, and latest, step in automated fraud prevention is machine learning.
You may have heard about machine learning in descriptions of how Google perfects its search algorithms or how self-driving cars navigate crowded intersections. Machine learning refers to systems that improve at tasks as they gain and analyze new data.
You can see how machine learning would be useful in the credit card fraud arms race. As cybercriminals adapt their strategies for committing fraud, artificial intelligence systems use machine learning to pick up on these new patterns and integrate them into existing data.
Machine learning can draw on a wide range of data, such as information on a merchant’s segments and geographic markets, the fraud patterns that exist in those spaces, and unique customer and credit card data. Machine learning systems can evaluate and process a higher volume of transactions than human analysts can.
Machines Can’t Do It All
At ClearSale, a cutting-edge artificial intelligence (AI) is one of the two pillars of our industry-leading fraud protection solution. Machine learning ensures our AI always stays one step ahead of the fraudsters.
Like any student, an AI solution is only as good as its teachers. That, in part, is the role of our analytics team. Their job is to feed the system the data it needs to identify trends.
For example, a software algorithm might recognize that transactions made on weekends are less likely to be fraudulent than those made on weekdays. But the system wouldn’t have been able to make that correlation if it hadn’t occurred to someone to provide it with date data.
Our expert analytics team is made up of data scientists, statisticians, mathematicians, and development professionals dedicated to solving the fraud prevention problem for e-commerce businesses. One of their most critical roles is to customize the fraud-spotting algorithm for our enterprise users.
Every business is unique; customers exhibit different behaviors in different markets. Therefore, an off-the-shelf fraud score won’t work for every merchant. For example, a $1,000 purchase that may raise red flags for a fashion seller is an everyday occurrence in the jewelry market.
Our 50-plus member analytics team allows us to train our AI to look for trends that vary by vertical. The analytics team is just half of the human element in the ClearSale fraud protection solution. The other half is our manual review team.
Why Is Manual Review Necessary?
Computers have come a long way in their ability to replicate and (in some, very narrow ways) exceed human intelligence – and we’re sure they’re going to go quite a bit farther in the coming years. However, when it comes to identifying fraud, we believe that algorithms still lack the flexibility, creativity, and intuition human experts can provide.
Remember, our goal is maximizing your revenue. That means not just identifying and stopping fraud (which computers are very good at) but spotting legitimate orders that only appear to be fraud (which specially trained humans can be very good at).
Algorithms flag fraud based on factors such as location, delivery address, and shipping speed. But genuine customers may place orders that, on the surface, fit the fraud profile. This is especially true of wealthy customers who shop online while they travel for work or pleasure.
Human fraud analysts – when appropriately trained and enabled with the right data – can use their knowledge and “sixth sense” to identify these high-value customers and approve their orders.
The ClearSale System
Here’s a broad overview of what happens when a customer places an order with an e-commerce merchant that uses the ClearSale fraud protection solution (keeping in mind our solution is highly customizable):
The customer places the order
The ClearSale system works in any card-not-present scenario: web orders, email orders, telephone orders, and mail orders. As soon as a customer arrives on your website, our system starts collecting useful data about them, such as where they came from and their browsing activity.
AI technology scans the order
Our proven statistical algorithm is built on a powerful, proprietary machine-learning platform, plus a series of fraud rules customized for your business. The algorithm processes an extensive amount of data, including (but not limited to):
- Information contained in the order
- Device information
- External data sources
- Behavioral data
- Historical data
After reviewing the data, the algorithm determines a fraud score for the order. If the score lands below a pre-determined threshold, the order is tagged as not fraudulent, and it is immediately approved. If the score exceeds the threshold, our system takes it as a sign the order might be fraudulent. The order is immediately sent to our expert review team for further analysis.
This is where our system diverges from most other fraud prevention tools on the market, which simply auto-decline suspicious orders. We know that more than 90 percent of auto-declined orders are legitimate, and we want to make sure you get the revenue you deserve.
Expert fraud analysts review the order manually
Our team of fraud analysts picks up where our AI technology leaves off. Thanks to their experience, expertise, and training, our human agents can see beyond the algorithm.
If the first human reviewer determines an order is legitimate, the order is immediately approved. If the first reviewer determines the evidence supports the fraud score, the order is passed on to a second reviewer. The second fraud analyst makes further efforts to verify the legitimacy of the order.
If the second fraud analyst determines no fraud has occurred, the order is immediately approved. If the analyst still can’t verify the order as legitimate, then, and only then, will it be declined.
In-Depth: The ClearSale Manual Review Team
We use the term “manual” to describe our human review process, but we do this mainly to differentiate it from the hands-off, AI-driven, algorithmic phase. Technology – specifically, the technology we have pioneered and perfected at ClearSale – has made manual review more precise and efficient than ever before.
It starts with how we assign reviewers to cases. We employ several hundred fraud analysts at ClearSale, each of which has different levels of experience and areas of expertise.
Our system – again, informed by machine learning and optimized by a crack analytical team – assigns reviewers to cases that best fit their abilities and styles.
One of our fraud analysts might, for example, have extensive experience adjudicating orders that originate or ship to Mexico. If a suspicious order comes from south of the border, the Mexico specialist, or someone with similar expertise, will get the assignment.
Another way our system assigns cases is by risk tolerance. Some of our reviewers are more aggressive about approving orders, while others favor a conservative approach. In the interest of maximizing speed and efficiency, the system will assign riskier cases to the conservative analysts and lower-risk cases to the aggressive analysts.
What Do Fraud Analysts Consider When Manually Reviewing Orders?
To determine the difference between fraud and legitimate transactions, ClearSale’s fraud analysts rely partly on experience, partly on intuition, and largely on expertise. Our reviewers keep up on the latest cybercrime trends, so they know what new strategies to look for.
Factors our fraud analysts consider can include:
- Customer history. Longstanding customers with no history of fraudulent behavior are typically more trustworthy than new customers. Our analysts also look for existing customers who make unusual purchases – a common sign of identity theft.
- Recently created email accounts. Fraudsters will often create new email accounts for a single transaction.
- Proxy IP addresses. Criminals frequently use proxies to hide their real locations.
- Virtual phone numbers. Virtual numbers are virtually untraceable, unlike landlines or mobile numbers.
- First-time customers that quickly make multiple purchases. Legitimate customers tend to develop relationships over time with merchants.
Of course, a customer can raise some or all these red flags without attempting to commit fraud. People switch email addresses for any number of reasons. Some people use proxies because they’re concerned about their privacy online. A customer might make an unusually large purchase because they’ve been saving up for a big-ticket item.
This is why we use human fraud analysts who can piece together a buyer’s profile to find the real story behind the fraud indicators.
What Does the Second Fraud Analyst Do?
Our system will not deny a transaction until two reviewers have signed off on it. The second fraud analyst’s role is to confirm the findings of the first analyst and to dig deeper. This may include contacting the customer by phone.
Our team of reviewers always uses call scripts customized to your business and the expectations of your customers. A script for a fashion brand will not be the same as the script for an automotive parts supplier.
How Long Does the Manual Review Process Take?
A well-trained fraud analyst can take between two and 24 hours to gather information and decide on a transaction. Purely automated systems are faster – nearly instantaneous – but the extra time for manual review is worth it. Human analysts will slash your chargeback rates and false positives down close to zero.
At ClearSale, we’re not comfortable making your customers wait any longer than necessary. We’re always looking for ways to speed up the manual review process. Our methods include group analysis, which lets us batch orders by specific criteria, enabling a single fraud analyst to review multiple orders simultaneously.
Another approach we use is double (or triple) validation. With this technique, the same order may be analyzed by two or three reviewers in parallel. We compare their results to make a final decision.
For certain kinds of purchases, time is of the essence. Customers buying movie tickets have showtimes to make, for example. In these cases, we may follow one of two strategies:
- Damage control, which allows your customer service team to add customers to a temporary whitelist if they challenge a declined order.
- Control groups review, in which reviewers analyze random batches of automatically declined orders to identify false positives, define trends, and refine fraud-detection algorithms.
Taking the Next Step
We hope this behind-the-scenes look at how ClearSale achieves the highest approval rates and lowest false-positive rates in the fraud prevention industry has been illuminating.
You’ve learned how fraud prevention has evolved from simple rules-based systems to systems that combine the strengths of machine learning with human expertise, creativity, and intuition. You’ve also seen how ClearSale fraud analysts make their decisions, and how everyone at ClearSale is committed to helping your business get the most revenue through online sales it possibly can.
Is a fraud managed services solution such as ClearSale’s right for your business? Get help deciding with our free e-book.