TRENDS IN NATURAL LANGUAGE PROCESSING IN HEALTHCARE

Introduction

EXL Health is working with several top healthcare payers and providers in the US who have started using Natural Language Processing (NLP) in their operations to bring efficiency in their Human-Computer Systems. EXL Health, with its deep operations knowledge and advanced analytics prowess, is at the forefront of this technology change. Through this whitepaper, we will describe the emerging technical trends in NLP and share a few use cases where EXL Health is leading digital transformation through NLP.

What is a Human-Computer System?

The term “Human-Computer System” was featured in a MIT Sloan School of Management Executive Education session, A unique perspective of AI: Creating humancomputer systems, over 2 years ago.

Per MIT’s point of view, organizations should focus on attaining the maximum efficiency of human-computer systems, not just the computer systems.

To understand human-computer systems, we need to understand what kind of roles a computer system can play:

  • Computer system acting as a “Tool”: In this role, a computer system acts as a medium to achieve certain objectives. For example, word processor, excel spreadsheets, etc. are very common tools.

  • Computer system acting as an “Assistant”: In this role, a computer system utilizes vast processing power and advances in Artificial Intelligence (AI) to mimic the human assistant. For example: Google search service, IBM Watson or Google Dialogflow (Chabot service), etc

  • Computer system acting as a “Peer”: In this role, a computer system can perform tasks that a human being can. In most scenarios, neither computer systems nor humans are perfect in completing the task. Yet, a combination of humans and computer systems can achieve the best results. For example:

    1. Automatic insurance claim adjudication by a machine, requiring human intervention only when certain parameters are out of bounds.
    2. Humans and computers are both participating in a task to predict something.

  • Computer system acting as a “Manager”: In this role, computer systems act as managers, providing real-time feedback or monitoring the activities of humans. For example:

    1. An AI program that monitors the conversation between CSRs (customer service representatives) and customers, and gives feedback to the CSRs in real time.
    2. Mechanical Turk online labor market coordinating the effort of the humans and creating a final solution.

As we can see in the list above, computer systems do not work in isolation, rather in combination with humans. Therefore, an organization’s goal should be to optimize human-computer systems, not just computer systems.

Use Case: Computer System as an “Assistant“

Client Situation

Our client was looking for ways to reduce operations cost due to inbound calls. Unfortunately, the data sources were unreliable and transcripts of the conversations unavailable.

EXL Health Solution Approach / Methodology

Our team was tasked with finding out the reasons for inbound calls and proposing a solution to reduce the overall costs.

Using natural language processing (NLP) techniques, such as Tri-gram analyses on customer service representative (CSR) call notes, we identified patterns in the calls. These patterns helped us in understanding the intent of the conversation. An illustrative snapshot can be seen below.

Top 10 Frequently Co-occuring Words (Tri-Gram)

 

Furthermore, our team has also used techniques like NER (Named Entity Recognition), which are helpful to identify if the user mentioned anything specific in the conversation. For example, let’s have a look at a question posed by a user:

 

Question: What is the status of my claim for my visit to provider “X” on May 31, 2020?

In the above question, the name provider (“X”) and May 31, 2020 are the two specifically named entities mentioned by the user, whereas, “status of claim” is the intent of the user, as an output provided by the technique used.

Now that we understood the intent of the conversation and recognized the entities in the conversation, we proposed to build a Google Dialogflow-based Chatbot for the client.

Anticipated Results

Early estimates from a single use case related to a technical issue associated with client’s member portal is a savings of over $500,000 based on 50% remediation rate.

Use Case: Computer System as a “Manager”

Client Situation

Our client was looking for ways to improve the performance of customer support representatives (CSRs).

EXL Health Solution Approach / Methodology

Our team was tasked with studying the performance of CSRs on calls and generate insights that can help improve their performance in a quality of conversation metric.

In this case, we received transcripts of the calls between the CSRs and the members. Using various NLP techniques, we measured the sentiment of the caller several times during the conversation. Along with the sentiment score, we looked at the outcome of the calls to measure the quality of the conversation.

Additionally, we identified the classes of activities that a CSR has to perform on the call. Mapping the activities performed by an average CSR against the top performers gave us the insights into their strengths and weaknesses.

Results

Based on the identified mistake patterns, we created a self-service tool that recommends the creation of a specific training plan for each individual CSR, with the goal of improving the quality of conversation by 20%. Additionally, we created a tool that helped in allocation of top performing CSRs to high-priority cases.

Results

fmovies

Contact US