Taking a tour through anomaly detection using AI
Anomaly detection is a way to capture suspicious events which differ significantly from the majority of the data. Detecting anomalies can help automate auditing transactions for fraud, support reclaims efforts, and enable preventive actions going forward.
This technology is becoming more and more popular as fraud remains a serious issue across industries and traditional methods cannot adapt to the changes over time.
Why AI for Anomaly Detection
Humanly Not Possible-
- Data Volume and Complex Patterns – Due to data flood and the complexity of anomaly patterns, humans provoke excessive false positives and overlook false negatives. Contextual and Collective anomalies patterns are also very difficult to notice with the human eye. If even 10% of 10 million annual expenses are marked problematic, that means a million expense line items would be needed to be investigated by the compliance team annually.
- Time-Consuming – Investigating all skeptical transactions itself is already time-consuming, not to mention checking signatures and reviewing complex profiles one by one with historical data manually. AI, on the other hand, outperforms humans in accuracy, speed, and productivity in identifying all kinds of anomalies invalidation.
- Adapt to New Changes- As the transactions get more and more complicated, so does the fraud. A proactive approach is needed to adapt to new fraud patterns as the overall data structures change and shift. While humans usually look for known patterns, AI machine learning can adapt to new changes over time, search for unknown patterns, and keep up with anomalies.
Human in the Loop
We are not suggesting replacing humans with AI in all aspects of this matter. Rather than replace the human role, our solution brings employee and AI efforts together throughout the solution lifecycle as technology is utilized to realize the business benefit.
After AI models are built, anomalies have been monitored, and benefits have been quantified, employees can easily change the underlying assumptions/parameters based on different business scenarios and needs. Finally, the interpretable real-time reports generated by AI will guide users through the investigation process, empowering employees with previously unrevealed insights and taking further actions. What we provide is not simply identifying skeptical events at transaction levels, but leveraging AI to enhance the quality of the process and improve decision making.
AI Anomaly Detection at Kavi Global
Kavi Global provides a wizard-driven, no-code software solution that applies AI to identify invisible anomaly signatures and flag the collusion of multiple fraudulent actors, in addition to identifying anomalous patterns for an actor or transaction level. We capture fraud at multiple dimensions to ensure point, global and contextual levels of detection.
Use Cases Across Industries
Invoices – Our AI solution is able to identify and monitor recurring fraudulent transactions and flag actors and even the conspiracy of fraudulent actors, such as repair billing fraud, inventory return abuse, fake shops/vendors collusion, duplicate charges, efficiently and automatically.
Pharma & Healthcare Billing Fraud, Waste & Abuse – Medical fraud detection is another popular field. In fiscal 2019, the Department of Justice recovered more than US$2.6 billion claims relating to the healthcare industry out of US$3 billion from civil cases involving fraud and false claims against the government. AI highlights anomalous systemic fraud patterns at a batch transaction level and flags the improper payment generated between actors i.e. doctors and pharmacists, and doctors and patients collusion.
Anti-Money Laundering (AML) – Banks in the U.S. spend more than US$25 billion annually on anti-money laundering compliance on average according to Forbes. As the volumes, complexity, availability, and regulations change, AI machine learning could adapt to new changes over time and flag recurring fraudulent actors categories to augment the traditional rule-based monitor systems.
Insurance Fraud – Insurance Research Council (IRC) also estimated that up to US$7.7 billion of auto injury claims were excessive payments in 2012, accounting for 13%-17% of the total payments of 5 main private passenger auto injury coverages. Traditional anomaly detection technology in the insurance industry recognized anomalies by fitting them into a preprogrammed template. As transactions get more and more complicated, so does the fraud. A proactive approach is needed to adapt to new fraud patterns and perform real-time dynamic analysis.
Supply Chain Sourcing, Production, & Quality Defects – The more components and processes in the production line, the greater the remake time and costs. Revealing the root cause of defects in each manufacturing process to enhance the yield is one of the most inexpensive ways to guarantee the quality, shorten the lead-time, and stabilize the supply chain. AI can easily detect whether the outliers derive from the nature of the process, human factors, machine factors, time factors, or environmental factors, enabling stressors to be eliminated or controlled prior to any significant deterioration, and help you to identify improvement opportunities to achieve Six Sigma (3.4 defects per million, 99.99966%).
< About the Authors >
Priyansh is a data scientist with six years of professional and academic experience in time-series analysis, machine learning, and data visualization. He has expertise in fraud detection and product analysis from previous roles. Priyansh holds a Master of Science in Business Analytics from Oklahoma State University.
Andy (Yen-Dah) Kuo
Andy is a data-driven solution provider with 5+ years of experience in finance and 2+ years in supply chain. He has a Master’s Degree in Supply Chain Management from University of Michigan. Expertise in developing logistic models, optimizing global supply chains, and building inventory models.