The Digital Transformation of Exception-Based Reports

Over the last 3 decades, many retailers have adopted the use of Exception-Based Reporting (EBR) tools as a method to detect potential fraud or loss occurrences. The software filters through the transactions and looks for anomalies and flags them for further investigation. Many retailers remain unsatisfied with the accuracy levels provided in these reports which leaves them with overworked asset protection agents, no answers into why these anomalies exist and not many other options.

Until now…

The explosion of cutting-edge technology like, computer vision (CV) and machine learning (ML), provides an incredible opportunity to improve exception reports by utilizing Artificial Intelligence (AI) to refine exception reporting accuracy and to provide context behind each line item to ease the strain on asset protection agents. There are many ways to take advantage of this technology, below are just a few.

Exception-based reports are built by compiling data, establishing a baseline, and then assembling a list of transactions that are outside of that baseline for further investigation. Unfortunately, that is typically all the data that goes into building an exception report. This leaves them inaccurate and lacking details sufficient to support an investigation. While there are some technologies that can be layered on top of an exception report, to provide video evidence along with each exception, you are still limited to the results from the original report.

Computer vision at the point-of-sale

Computer Vision can be used to improve the reliability of EBRs and the additional data extracted provides context for asset protection agents to start their investigations on solid ground. In addition, CV can detect suspicious behavior that traditional exception reports cannot.

One thing that CV does extremely well and is a very established practice is the ability to detect people. This can be applied to each point-of-sale (POS) terminal. A camera can be positioned over the POS terminal and can be trained to detect when a person is or is not present on the customer side of the terminal. Then, with a POS integration, we can run that model on interesting transactions. For example, we do not necessarily care if somebody comes in and pays cash for a pack of gum. However, if the clerk processes a cash refund without a customer present, we will want to determine why and can flag that transaction as requiring further investigation.

With the expansion of self-checkout lanes, there is a greater need to ensure patrons are properly scanning each item they intend to purchase. CV models can be trained to detect scan avoidance and other suspicious behavior in these instances.

As for manned checkout lanes, many employees are feeling the strain of being overworked and understaffed at retail locations, and while the pandemic exploited these issues, the fact is that high turnover rates have always existed in retail. Standard POS reports will display the number of items that were a part of a given transaction. Using the same hardware as the above examples, we can deploy additional CV models that have been trained to count the actual number of items in each transaction. Using automation to compare the POS data to the CV data can help to identify additional items that require investigation. These can then be categorized by store, or by clerk and help to identify opportunities for further training.

By reducing the number of transactions that need to be filtered through, companies can have asset protection agents and district managers take on more stores while reducing their workload and improving their ability to complete their tasks more successfully.

Conclusion

EBR's have been assisting loss prevention resources for approximately 30 years, but the technology now exists to improve upon these reports. As enterprises grow more data is generated and EBR outputs can become increasingly difficult for employees to wade through. Utilizing Vision AI ensures that employees are working through lists that truly require investigation.

Taking advantage of Vision AI technology can improve the employee experience, increase the number of successful investigations and provide retailers with a strong return on investment (ROI). Traditional EBR reports rely on batch uploads and processing files sometime after the event has occurred, making investigations very reactive. Vision AI allows for the ability to start investigations much more quickly and to be more proactive in these processes.

See how WWT partner EVERYANGLE approaches this specific Vision AI application, and check out other Computer Vision applications from EVERYANGLE. If you would like to learn more about how WWT can help you implement Vision AI technology at scale in your organization, we're here to help with a complementary 1-hour briefing call.