Anomaly detection at Accenture
Applying anomaly detection using artificial intelligence when rules are not enough.
Accenture is an intelligent enterprise that applies artificial intelligence (AI) technologies and embeds analytics into our operations. Our internal IT Enterprise Insight organization collaborates with Accenture business stakeholders to drive new, innovative capabilities and use case-based intelligent products to bring greater insights to Accenture’s businesses.
One capability that Enterprise Insight excels in is anomaly detection, the use of analytics to automate the detection of anomalies and outlying behavior such as noncompliant business expenses, clauses in contracts, and procurement and IT activities. We chose to develop an anomaly detection solution for Accenture’s expense reporting system to augment our existing rule-based analytics. The objective was to help address the high volume of expenses reported that trigger as false positives. A high number of these cases are ultimately compliant, resulting in an inefficient use of audit time by our Compliance Team.
The AI challenge: To reduce the number of false positives in the anomalies being detected to improve audit efficiency and redirect time to more questionable claims.
Accenture processes approximately 25 million expense lines annually. This number will grow as Accenture continues to grow. Every report gets analyzed by a manually designed rules-based system to check for expense compliance. Roughly 10 percent of expenses get flagged for potential noncompliance, which is then audited by our internal Compliance Team.
Traditional rules-based systems—while effective at detecting known and recurring patterns of noncompliance—often return exceptionally high numbers of false positive alerts and mistakenly flag legitimate behaviors as suspicious. Traditional rules-based systems can also be exploited by fraudulent behavior and fail to consider previously identified noncompliant behaviors.
Enterprise Insight in collaboration with our Time and Expense organization took this situation as an opportunity to improve upon Accenture’s current rules-based system and created an AI solution that more accurately identifies noncompliant expenses and reduces false positives. The AI solution is also able to more easily identify noncompliant behaviors with hidden patterns that are difficult for human auditors to identify.
Our Enterprise Insight team developed a deep-learning algorithm that ingests historical expense, time charge, and location data. This development led to an intelligent, unsupervised solution that observes the data itself, rather than just the flags identified by the rules, to detect outliers in the data. This approach enables the AI to detect hidden patterns in the data and co-evolve with the behavior of the employees, rather than the rules, therefore, detecting noncompliance that is going unflagged. Initial results on this have shown a 30 percent hit rate of identifying transactions to be questioned.
The model, which works in parallel with the rules-based flagging, will continue to become more sophisticated and accurate as the AI learns. This “hybrid” model is distinct from traditional anomaly detection models in that it takes advantage of both a rules system and AI models, leading Accenture to apply a patent for it. Enterprise Insight also used explainable artificial intelligence (XAI) to give explanations, which are then used to generate narrative for auditors to review.
Our expense compliance product helps ensure good citizenship around Accenture’s billing and expense reporting compliance. It has reduced the number of false expense entries being detected by 10 percent. It, in turn, improves the employee experience by reducing unwarranted inquiries. This outcome translates into business value gained by reducing the time our Compliance Team auditors spend on expense cases that are actually compliant, reducing the time employees spend responding to questions, and redirecting audit time to more questionable claims. The product adds further value by detecting hidden-pattern noncompliance.
Our Enterprise Insight organization continues to further optimize the anomaly detection model for Expense Compliance and apply new deep-learning models to improve accuracy and business insight. Advanced digital capabilities, especially anomaly detection, hold the potential to be applied in other use cases of high-volume transaction activity generated by human activity. Possibilities include procurement, IT operations, banking, pharmaceuticals, and insurance and health care claims, among others. Human perception and knowledge are not replaceable by models as humans are always leading and proceeding. Human-generated rules and models generated by human and machines can work hand in hand to make work flow more human centric than machine centric.
Enterprise Insight continually drives innovation and applies predictive models, AI and machine learning to products to bring greater insights to the business. Within Enterprise Insight is the Studio, the research and development organization of data scientists, user experience experts and software engineers that experiments and builds advanced analytics solutions. It operates with a culture of creative agility, following emerging technology market trends, prototyping new analytics concepts and working with a fail-fast culture. A thriving analytics ecosphere is promoting winning ideas.
Analytics products are advancing Accenture’s transformation journey to becoming an enterprise that is automated, intelligent and insight-driven. Accenture envisions this future digital-insight culture as one that delivers new value in many ways. Accenture’s reporting landscape will become simplified as more clarity on what to use is gained. There will be broader insights into business performance as all business dimensions will be supported with digital insights—anywhere.