Fighting fraud isn’t easy. It’s time-consuming, it’s challenging, and it’s expensive. Taking an artificial intelligence (AI) approach seems like a better strategy.
After all, AI can quickly analyze extensive amounts of customer data to identify emerging fraud patterns. By incorporating this insight into fraud risk scoring algorithms in near real-time, businesses can lower their risk of falling victim to fraud.
Where AI excels is its ability to quickly review incoming transactions. The algorithms can calculate fraud risk scores far faster than manual reviewers ever could, which means for businesses that must deal with a high volume of incoming orders, AI can be the difference between speedy approvals and disgruntled customers.
There’s a cost aspect to this as well. Because this initial review of incoming orders can be automated, businesses don’t need to spend valuable (read: expensive) man-hours reviewing each individual order. This can tremendously cut down operating costs, and it enables organizations to focus their analysts’ attention only on those specific orders that require further investigation.
And finally, AI can be very helpful at spotting obscure patterns in fraud that might not be readily apparent to the average reviewer. This is particularly true if businesses have a good amount of high-quality, validated information on past orders.
However, for all the benefits AI-based fraud solutions can offer, the technology does have some limitations when it comes to evaluating credit card transactions.
That said, there are several circumstances where AI alone will not be sufficient to manage ecommerce fraud and may even prevent businesses from maximizing sales.
1. Auto-Declining Orders
AI does a great job of auto-approving good orders and flagging potentially fraudulent orders. But AI should never be trusted on its own to auto-decline orders. Statistics show false declines – that is, accidentally declining orders that are actually legitimate – is a very high, very real risk for ecommerce merchants.
By all means, AI should be used to flag orders that might be fraudulent; but they always need an additional review before declining them. False declines cost businesses 13 times more than credit card fraud. Organziaitons need to validate every decline decision, to be confident they’re not inadvertently ruining a relationship with a perfectly good customer simply because they’re trying to ship a gift to a different address (which is a common reason for false declines).
2. Lack of High-Quality Data
AI is only as good as the data it receives. If product is brand new or highly unique, AI-based fraud solution may not have the data it needs to make accurate approval decisions. In these cases, an AI solution may end up declining orders that are good – and the AI algorithm may inadvertently become more conservative as time goes one and decline more and more future orders.
For these types of products, flexibility will be key to ensuring merchants are maximizing sales without also increasing chargebacks. This may require turning off (or loosening the standards) for the AI solution while they closely monitor actual sales and fraud trends. Businesses should study their potential risk, and modify their fraud strategy based on their learnings.
3. Troubles with Fraud Filters
At the core of many AI solutions lies the ubiquitous fraud filter. Fraud filters use rules to evaluate incoming transactions; if a transaction meets certain criteria, the transaction will be either flagged for review or auto-declined.
The problem, therefore, is whether the fraud filters are set up properly. It’s tempting to think that if one fraud filter catches some fraud, more fraud filters will catch even more fraud. Unfortunately, this isn’t always the case. Layering filters incorrectly can result in some rules canceling others out, leaving you as vulnerable as if you had no fraud protection.
Moreover, smart fraudsters know how to “play” the fraud filter game. For example, it’s not terribly difficult for a fraudster to test a system with a series of orders and eventually learn that orders under $1,000 are typically approved, whereas orders over $1,000 are typically reviewed. Once they learn this, they’ll flood the system with batches of orders of $999 that enable them to fly under the fraud filters’ radar, completely undetected.
A Winning Approach = AI + Human
Perhaps the smartest approach is to combine the best of both worlds: Businesses should implement a comprehensive fraud management solution that combines AI technology with expert fraud analysis.
This multilayered approach enables organziations to harness all the efficiency benefits of AI, so they don’t have to slow down the order approval process – and it also safeguards businesses against the risk of accidentally auto-declining orders from their best customers. In this type of approach, AI might be used to auto-approve good orders and flag orders that are suspicious, while a human analyst that understands the nuances of fraud can manually review and either validate the order or confirm the decline decision.