Claim Adjudication Analytics

About the Customer & Challenges Faced:

Overpayment on healthcare claims can cost millions of dollars. Our client, a payment integrity organization based in Indiana, offers post-payment claim overpayment & recovery solutions. They partner up with their customers to identify, eliminate and recover overpayments. They also qualify specific overpayment errors to determine the root cause, with a focus on situations that can be changed or eliminated.

In the Workers’ Compensation line of business (LOB), our client gets itemized bills for each claim. The dollar value of the claims in this LOB is fairly high. Each claim has many itemized rows with semi-structured textual descriptions known as claim lines. 7-8 nurses manually evaluate claims for which the claimed amount is greater than $50,000. Nurses use their domain expertise to evaluate these claims. It makes the adjudication process manual, labor-intensive, costly and inefficient. They wanted to use analytics to augment the bill review process.

Solution and Approach:

  • Our client wanted to enhance their efficiency of claim adjudication by automating the claim adjudication process. This was enabled by classifying each item description that was part of the claim into an accurate charge type in order to apply business rules of adjudication. ​
  • framework provides out-of-the-box development platforms. framework provides out-of-the-box development platforms. The project was started seamlessly with the relevant environments which were then created automatically. Development images configured based on pre-defined templates were installed on-premises or in a development VM within the infrastructure. This enabled authentication using LDAP, seamless project setup using Bitbucket, Jenkins and Docker (ensuring build and deployment without software compatibility issues).’s MLOps framework allows establishing high-end Alluxio and Presto-based efficient data connectivity and collecting data from diverse sources.​
  • The framework made available by leverages the latest ML and DL tools while preparing models and includes Pachyderm-based data versioning, deployment using a Kubernetes orchestration system, Kubeflow and Spark-based ML and DL build and deployment, Istio-based service mesh enabled microservice architecture, and ELK based monitoring capability; contributing to reduction in latency time.​
  • We used NLP and advanced ML-based algorithms to understand the context of the item description and prepared a continuous learning system to classify each item into a charge type (e.g., laboratory changes, monitoring charges etc.)
  • A classification engine – part of the pre-processing workflow–enabled standardizing each item description via spell correction and abbreviation expansion. Next, concept identification–identification of domain similarity of each item with different charge type – was performed and advanced ML-based algorithms were used to classify items into one-of-a-kind charge classes with a confidence score.
  • The details collected were added as exploratory variables by using libraries and analyzed. The attributes obtained were used for categorization (employing Pachyderm-based data versioning) and then performing univariate, bi-variate and Bag of Words analysis — for both structured and unstructured datasets through xpresso Exploratory Data Analysis (Data and Statistical Analysis).  Different datasets and their different versions were easily controlled and stored into xpresso Data Model (XDM)-enabled data store that enabled easy retrieval and storage of datasets/ files into internal XDM. This was achieved by using two excellent features of ​
  1. Data Connectivity Marketplace libraries​
  2. Data Versioning​
  • The output, including all discrepancies and low confidence items were fed back into the application in order to contribute to the incremental learning. The output and each classified item were further re-evaluated with the help of domain experts. Each classified item was passed through a decision node that sent its basis threshold, either to the business rule engine or for a review from a domain expert. The rule engine that developed incrementally with the help of this classification and contract guidelines, finally decided claim adjudication. The approach resulted in a 40% increase in assessment speed for claim adjudication and reached over 90% accuracy without dependency on manual processes. 


  • By using, one can leverage high-end data connectivity, efficient data versioning, perform exploratory data analysis and generate inferences using an intuitive process and through an industry-standardized manner.
  • The unique, containerized platform-centric approach offered by can be used to employ required infrastructure, deploy rapidly to multiple high-availability environments while aligning with best-in-class DevSecOps practices.
  • also brings in-depth QA-QC testing and logging frameworks, synchronous and asynchronous monitoring, and performance tracking ability.
  • also has SSO (single-sign-on) for various in-built tools and subsystems that make the platform access seamless throughout.​
  • In a nutshell, all the above features in a single plate under the same hood make an unbeatable AI Ops framework.

Have Any Questions?

Need more information about the platform?