How to Leverage Machine Learning for Manual Reviews
The biggest expense in fraud mitigation for nearly all online businesses involves manual review agents and the time they spend conducting manual reviews. Employing automated rules and machine learning can dramatically minimize the number of ordered needed for review. Once the orders are held for review, there are key strategies that can be used to help any online retailer’s manual review team become rock stars.
Let’s look at how artificial intelligence (AI), machine learning and agent automation can catch more fraud, increase efficiency, and reduce false positives.
Minimize manual reviews. AI and machine learning, when coupled with the other critical fraud prevention capabilities found in enterprise-class anti-fraud solutions, make it more likely that borderline transactions can be resolved without escalation to manual review. AI and machine learning have unlimited computing and storage capacity to assess in real time the hundreds of data points analyzed by fraud screening technologies like Device ID, Geolocation, Proxy Piercing, Order Linking, and more, and provide a score that quantitatively measures the risk involved in a particular transaction. Setting up rules that automatically approve or decline transactions—based on their quantitative score—that once may have been escalated for manual review can help minimize how many transactions are reviewed.
Agent automation also helps eliminate manual reviews by automating checks that may have once been part of the manual process. For example, a mismatch between billing and shipping addresses may always prompt a manual review to check a third-party source like Whitepages Pro to confirm both addresses. With Agent Automation, you can automate this query and eliminate the intervention of a human agent. This saves time and speeds resolution of orders.
Conduct more efficient manual reviews. AI and machine learning can help manual reviewers avoid repetitive tasks and speed up resolution. For example, they can capitalize on their access to Big Data, unlimited processing power, and memory capabilities to automatically normalize addresses when fraudsters introduce slight variations—for example, substituting “St.” for “Street” in an address line–in an attempt to disguise multiple orders and make them appear to be different customers.
Similarly, agent automation can automate repetitive tasks to boost productivity and improve outcomes. For example, it can auto-route cases to the most qualified or most appropriate agents. Thus, Spanish language orders can be automatically routed to Spanish-speaking agents. It can also improve order flow by automatically prioritizing the review queue. For instance, overnight or expedited orders can be always promoted to the top of the queue without intervention by human agents, ensuring those high-priority orders get reviewed and processed first.
More than just day-to-day operational benefits. Another way AI and machine learning can improve productivity is by helping make improvements in the manual review process at a strategic level. Data from the past can be used to generate suggested pro forma rules for future orders. These pro forma rules project the costs and gains of changing the parameters that trigger a manual review. Fraud Analysts and managers can see quantified ROI and decide whether or not to implement the proposed change. The impact can be powerful: one Kount customer used this capability to refine rules and processes in order to reduce their manual review agents from 8 to 4 to 2 to eventually 0 (these personnel were redistributed to revenue-generating activities, further boosting the contribution to increased profits).
Want to know how you can take advantage of these best practices so your team can review two times as many orders and improve accuracy by as much as 75%? Download the eBook “7 Best Practices to Become a Manual Review Rock Star.”