How fraud detection is made?

Before Konduto, there were three main ways to detect payment fraud in e-commerce: Check registration data, fingerprint (device ID) or using a rules engine. Our technology combines the very best in the three methods described, but adds a weight difference: the user behavior analysis which navigates in the virtual store site.

From the monitoring of browsing behavior, we can observe the whole process of customer's buying decision, the instant that he accesses the site first time by the time of completion of the transaction and the fill of the data at checkout. Our monitoring can do this analysis also when the Internet user is not identified.

Once the application is finalized, our system analyzes not only the customer's browsing behavior as various other factors and make a real-time recommendation on this purchase: whether it should be approved, denied, or manually reviewed by an agent. The response of Konduto is given in less than 1 second, without halting the operation of e-commerce.

At the same time, our system relies on the technology of machine learning and can learn from the operation itself to e-commerce. In this way, the algorithm adapts uniquely to each shop, learning more and more about the buying behavior of each type of customer - and, over time, reducing more and more the order numbers submitted for review.

The Konduto performs automatic risk analysis, but provides a manual review from a partner with agents to investigate the merits of the transactions flagged as suspicious by our system.