logo-blue
How much is fraud costing your business?

Measure your true cost of fraud using our free, interactive calculator.

Get my results

Approve More Orders by Modeling Good Customer Behavior

 

Most retailers can agree that investing in a robust fraud system is the key to safeguarding their business. However, not all fraud systems are created equal. Some fraud prevention tools rely on outdated methods such as rules-based models or static decline lists to power their decisioning. 

Without access to real-time behavioral data, these models often take a more conservative approach to fraud detection — focusing heavily on rejecting any orders that trigger any of the predefined rules (yes, even legitimate ones). So while a business may sign with a new fraud solution and see their fraud rates decline, they often fail to spot the growing false decline issue that can prove just as detrimental to their business. 

 In this article, we’ll cover the limitations of rule-based models, review some real-world examples of falsely declined orders, and show you how adopting an approval focused approach can provide more accurate fraud decisioning for your business. 

Are you leaving money on the table? Understanding the limitations of rule-based models and the true cost of false declines.

What is a rule-based fraud detection model? 

Rule-based decisioning was one of the earliest methods retailers could utilize to curb the number of fraudulent transactions on their site. Rules-based fraud detection systems use pre-defined rules to identify high-profile fraud patterns and subsequently categorize orders as authentic or fraudulent.

Why rule-based systems aren’t as effective as you may think

  1. They require constant upkeep.
    Fraud patterns evolve constantly. For your rule-based models to be effective, your risk team needs to review and update these rules periodically to make sure you’re not leaving your business susceptible to the latest threats. This is a stark contrast from machine learning-based models that can continuously train on data sets to make more accurate predictions over time. 
  2. They are inherently conservative. 

Most rule-based systems have to work with limited information and are forced to be extra conservation in their decisioning to avoid letting fraud through. Without the ability to look at payment and checkout data to provide additional context, these systems tend to overcorrect and reject any orders that trigger any of the rules. The rampant false positive problem means that in addition to catching fraudsters, these retailers are also rejecting good orders and severely limiting their revenue potential. For hyper-growth retailers, this misalignment of prioritizes (catching fraud instead of growing sales) can significantly set back your growth. 

How much harm can false declines have on your business?
Turning away good customers results in more than just missed revenue opportunities, it also frustrates customers and lowers retention. According to research by PYMNTS.com, 32% of customers said they would not return to a merchant where they had been falsely declined. Instead of delighting and capturing, false declines can cause your most loyal customers to feel alienated from your brand.

Case study: Real-world examples of false declines 

To underscore just how easy it is for rule-based models to get it wrong, let’s explore these three real-life examples of falsely declined transactions. On paper, many of these orders have been categorized as risky, but upon closer inspection, these rulings were overturned and the orders were manually approved as legitimate. Some names and identifying details have been changed to protect the privacy of individuals. 

Scenario #1: High ticket health product 

Alice has been contemplating this purchase for a while now: an expensive health product price at about $2,700. After searching for hours online for reviews and coupon codes, she finally decides to pull the trigger on the purchase. Despite completing the checkout it seems something went wrong and despite re-trying the order twice, her purchase still doesn’t go through. Alice’s credit card issuer denied the transaction on the grounds of suspected fraud. Frustrated, she steps away unable to make the purchase she spent time researching. 

One month later Alice comes back to the site and is able to get the transaction approved by her bank. Bolt is able to see this order came through on the same credit card that was denied one month prior. In addition to this, her IP address was found to be associated with a hospital network where she was confirmed as an employee by her LinkedIn page. Despite the frustration and seemingly negative signs, Bolt was able to approve the transaction.

Scenario #2: 

Lauren browses for earrings on a jewelry site known for their collaborations with popular TV shows and hip-hop artists. A merchant that came to Bolt with an existing fraud problem, they regularly see attempts for unauthorized transactions. This order is shipping to an address where the cardholder is not listed and is different than the one registered to the credit card. On a merchant known for their high loss rate, these signs alone might be enough for many fraud detection systems to reject. However, Bolt’s unique position allows us to see that she was browsing the page by looking at the least expensive items first. After applying a coupon and checking out, we were able to confidently approve her order despite being a seemingly risky transaction. By understanding how consumers browse and make purchases across a wide range of retailers, Bolt is able to safely approve more orders.

Scenario #3: 

Tyler is shopping for a hot food table, a piece of commercial kitchen equipment typically sold to restaurants. This one coming in at $2,300 is no small purchase. The order is being sent to a residential address not registered to the credit card, in addition the name on the card being used is Stephanie but the email belongs to Tyler. How can it be proved that a transaction is authorized when we know the cardholder is not actually the one making the purchase? By scouring social media we were able to tangibly connect the cardholder and purchaser. Both can be traced to a ranch who’s website notes that they offer hunting and lodging packages. It also specifies that they have a large dining room with a serving table for guests to gather after a long day out in the blinds. These additional insights allow us to assuredly make sense of an order with multiple high-risk signals. By connecting the two people to a business and making sense of the purchase for said business, we were able to approve another seemingly risky transaction.

Accept more good orders using fraud indemnification 

Fraud detection using online customer behavior indicators

Fraud indemnification represents a huge departure from the way retailers typically combat fraudulent transactions. Instead of focusing on identifying and rejecting fraudulent orders, fraud indemnification makes maximizing revenue a priority. To do this, fraud indemnification solutions focus extensively on finding ways to qualify and approve more orders (usually by pulling behavioral data from the customer’s own shopping journey with machine learning).

With an indemnification model, the incentives of both vendors and retailers are aligned. Since retailers are only charged for approved transactions, these vendors are encouraged to approve as many good orders as possible. However, because these vendors are responsible for indemnifying or covering any fraud-related chargebacks, it’s also in their best interest to only approve legitimate orders or risk incurring a lot of extra fees. 

How modeling good behavior helps improve order approval
Fraud decisioning shouldn’t be made in a silo. As evident from the scenarios above, a lot of normal shopping behaviors can cause legitimate transactions to incorrectly appear fraudulent. Combating false positives is as simple as adopting a more approval focused mindset and making sure you’re taking a holistic approach to order review. 

Why it pays to partner with a comprehensive fraud solution 

When used correctly, rules-based anti-fraud solutions can be a powerful first line of defense against fraud. However, depending solely on rules-based systems to completely safeguard your business could potentially result in a higher-than-normal false decline rate, causing long term harm to your profits and reputation. 

With fraud trends evolving at an exponential rate, retailers will constantly find themselves turning to more sophisticated approaches just to keep up. Luckily Bolt is here to help. Our platform mitigates risk for retailers by using behavioral data collected throughout the checkout experience to inform our fraud decisioning. By combining proprietary behavioral data points and machine learning algorithms, Bolt helps retailers stay protected against fraud while approving more good orders.

Want to learn more about Bolt’s fraud solution? Click here to speak with an ecommerce expert. 

If you enjoyed this post, share it.

Want to see a 5-minute demo?