Reducing manual workload with AI for RCM company

About client

Our client is a leading revenue cycle management provider with healthcare clients across the US. They offer RCM services, patient access, clinical services, and patient financial services to healthcare organizations of all sizes. Having more than a decade of experience in RCM, they help healthcare providers streamline operations, optimize financial performance, and introduce innovation to address the complexities of modern healthcare companies.

Challenges

Our client’s major operations include running an AR team whose responsibilities include following up with account holders of various insurance policies. How did this process become cumbersome, extending their hours of work, efforts, and wages spent on shift-based workers? Explained below.

Every claim is different, so is its followup procedure: Some of the claims they process require contacting the user on phone and get more inputs, whereas others don’t. As they couldn’t work around this, they manually reached out to everyone, spending more time than necessary. It demanded handling a large AR team and still got less work done by the end of the day. 

More wages and unsteady workloads: Voice calling process made more team members burn the midnight oil, inviting additional wages for the night shift. The work nature was so dynamic that some of them were swamped with heavier processing workflows while the rest with not so much to do, creating an unfair balance among them.

Millions of data: They had 3+ years of data with millions of rows and 120+ columns, something they cannot provide valuable insights on its own.

The solution

We suggested optimizing the process of policy validation and claims processing using a machine learning classifier system. The system would segment the accounts in two categories - voice and non-voice. 

The work started as our data scientists collaborated with the client’s operations team to understand their landscape and domain nature. This had a significant impact in clever feature selection and data transformation, dropping a whopping 80+ columns out of their data.

Given their machine and compute capabilities, we chose the Cat boost classifier model, which is proven to work well on categorical datasets. 

After training on historical datasets, the model exhibited 85% accuracy during the testing phase in segregating accounts for voice and non-voice reach-outs.

The impact of the model includes:

Reduced burden on workforce: The AI-based solution led to a more evenly distributed workload and alleviated the burden on night shift employees. They could choose the appropriate method to process the claim, depending on what the situation warrants. They were able to save 10+ hours of work and reduce the number of workers involved too. 

Low operational costs: By operating fewer people on night shifts and overall, they could reduce operational costs during the period we tested the model with live data.

More decision accuracy: The model’s adeptness to pick intrinsic patterns from denial codes and other features of historical data led to more accurate choices, thereby leading to accurate decisions.  

Quick turnaround and more satisfied customers: The classifier algorithm enabled faster decision-making by determining whether a call to the account holder was necessary. This resulted in quicker turnaround times for policyholders, enhancing customer satisfaction.

Future-proofing for a more streamlined workflow: This paved the way for the client to adapt AI-based solutions in other areas of their business as well.

Conclusion

The classifier model has performed well during the concurrent testing phases. However, our team didn’t stop there and worked on hyperparameter training to improve the accuracy rate. We built multiple use cases around, performed cross validations, did feature merging and transformations, attempted iterative training, and other algorithm tuning techniques. These efforts very well reflected in the final outcome, increasing accuracy to a near 90%. Much beyond these values, the AI-powered application smashed their bottlenecks and laid a firm foundation for automated and fast insurance processing and better customer satisfaction.