Case study

Harness the power of NLP Analytics

How we supported a revenue cycle management suite for the healthcare industry improve the success rate of appeals for payment completion and thus increase their overall revenue using Natural Language Processing (NLP) Analytics.

IMPROVED ACCEPTANCE RATE WITH 25%

~65% UPLIFT IN OVERALL DOLLAR RATIO

IMPROVED OVERALL REVENUE

The Business Challenge

With the previous appeals made by the hospitals to the insurance company for outstanding payment completion, about 70% resulted in failure.​

​Our client wanted to understand the reasons leading to these appeals resulting in success or getting retired. They also wanted to identify the combination of words that could improve the chances of appeals to be successful. The success of the appeal was defined by a payment made after it, which can be quantified by calculating the actual payment-to-payment request ratio, considered as the dollar ratio.

Harness the power of NLP Analytics 1
Harness the power of NLP Analytics 2

The Solution

The appeal letters were of a wide variety. These letters were studied thoroughly to understand the content and the way details were mentioned in them. Based on the exploratory data analysis, the letters were segregated into different groups. Amplify created a deep learning-based model to classify these appeal letters in each group as successful or likely to be rejected.​

​To further identify the likelihood words and sequence of words used in these appeal letters likely result in success or rejection, an Explainable AI model was created. Based on the explainability of the words, a new sequence of words was suggested to further improve the verbiage of these letters, which could in turn result in improving the overall acceptance rate of the appeal letters sent by the hospitals and health systems, maintaining the correct semantics and syntax as per English Grammar rules.

Our Approach

The methods used

Machine learning models were used to achieve the right lead scoring. ​
Decay scores were created for some relevant engagement features: Email sent, Email open, Email click & Page visit. The model doesn’t only count the number of important engagement features but is assigning some scores to the time at which the actions were done. If somebody visits a page today, he/she would be given higher importance than somebody who visited the same page a month back. These decay scores are calculated using an exponential equation.
Decay rates are different for the diverse engagement features: the decay rate for “Email open” is slower than the decay rate for “Email sent”.

ML Model deployment

This machine learning model was deployed successfully on Azure Cloud with no bugs and is being used by the client’s sales team. ​
Amplify pulled data from the client’s marketing applications such as SFDC and Pardot and used the Databricks platform, to train multiple models and MLflow, part of Databricks, to track experiments. ​
The fresh scores are then pushed back to the Salesforce application in an automated fashion.

Business impact

The case study showcases how Amplify Analytix provided a solution to the client using Text Classification and NLP Analytics. This helped in revising the content and format for the letters sent by healthcare service providers to appeal for payment to insurance companies. The new and improved appeal letters resulted in a better acceptance rate, and payment completion by insurance companies further helped in improving overall revenue.​

Related content

Want to know more?

NLP Analytics

NLP Analytics offers significant benefits to businesses by unlocking valuable insights from unstructured textual data.

READ MORE

Book a free consultation

We are happy to answer any questions regarding our services, your data, and how we could work together.

Sitemap

Useful Links

Newsletter Sign up

This project has indirectly received funding from the European Union’s Horizon 2020 research and innovation programme under project REACH Incubator (Grant Agreement no. 951981).

© Amplify Analytix 2023. All rights reserved.