Use of AI to identify fraudulent claims leads to valuable data on opioid misuse

We were approached by a large health insurer to provide a solution that would identify fraudulent health insurance claims.



Improve predictive modelling, network analysis and reporting using data-driven methodology that identifies normal patterns from real data and detects deviations from the norm for opioid use or misuse.

Applied Intelligence

Through our product, Artificial Intelligence Enterprise Server (AIES) we identified normal patterns from real data and detected deviations from the norm in insurance records, ICD-9 disease classification data, National Drug Code data and Physician Code data.

Our solution included pattern recognition Machine-learning, deep learning, image processing and speech recognition techniques to detect approximate classes, clusters, or patterns of suspicious behaviour either automatically (unsupervised) or to match given inputs (i.e. automated screening that focuses on detecting diverse forms of anomalies). In addition, probabilistic reasoning techniques were used to build graphical networks that can learn suspicious patterns from samples which can be used later to detect them.

Outlier / suspicious components are detected on the basis of shared characteristics identified and these defined set of indicators was then developed for customers / government to autonomously track for outliers in specific areas of space and time. Outlier / suspicious entities were further detected by performing simulations, temporal analysis, geo-spatial analysis, anomalous temporal changes combine with graph analysis to support a visual representation of the findings. Importantly, this outlier engine combined with other AIES components enables data to “stay in place”, avoiding the need for data to leave the confines of owner / collector of data.

Cluster of Opioid Misuse / Abuse (Characteristics) and Spatial Location

Cluster of Opioid Misuse / Abuse (Characteristics) and Spatial Location


Our solution provided details on prescribing patterns across 50 million records in a U.S population, with an easy-to-use interface to examine predictive features and clusters of opioid prescription use and misuse. Results were presented using interactive and virtual reality graphics.

Most fraud solutions on the market today are rules-based, designed to identify suspicious transactions by assessing if any rules have been violated making the system easy to manipulate and get around the rules.

Other Use Cases