A very large enterprise software company posted a video on finding fraud, waste, and abuse using a new query tool. The tool mainly compares names and addresses across multiple “watch” lists, which is a common approach used by several software companies and for numerous in-house projects. At first blush, this is a reasonable approach, and it provides some measure of risk mitigation. However, reality has a way of impinging on theory, making these tools less useful than expected. In practice, simple comparison analytics either miss matches when faced with even the most basic concealment techniques, or they yield an intolerable number of false positives. The latter is particularly insidious, since if the system produces too much noise, it will be ignored. This provides the impetus for a multi-analytical approach. In addition to comparing names and addresses, an analysis engine must take into consideration other supporting evidence, which likely goes beyond the single master data list. For example, is the activity of the suspect typical of this kind of entity (employee, vendor, etc.)? How about for the partner entities involved at the other end of the transaction? Has there been a sudden change in activity? Has the transaction analysis system uncovered other suspicious activity related to the entities? This approach, called evidential reasoning, takes a lesson from law enforcement and forensic analysis, which uses as much heterogeneous information as possible. It’s an approach also used by the autonomous vehicles, when many possibly conflicting inputs and sensor data are available, yet the system must still select an action or risk crashing. To use a simple name matching system would be like driving a car forward by sight alone while ignoring peripheral vision, and external clues such as the sounds of sirens, the smell of burning rubber, or the feel of the steering wheel and brake pedal vibration. The Oversight platform was built from the ground up using this principle, so it can operate semi-autonomously, thereby minimizing the demands placed on the human user. The efficacy of this approach has been demonstrated repeatedly, with successful implementations where competing systems have failed. Check us out for those war stories, and stay tuned for more on the technology of semi-autonomous analysis.