Sponsored Content

Maximizing returns from your AI investments 

Written by Kaushik Roy, Vice President of AI Product Management at Change Healthcare


Healthcare organizations are looking for solutions to help them create efficiencies and increase their return on investments. Learn how Change Healthcare approaches AI development initiatives as a way to help improve existing business processes and improve AI investment’s success rate.

Full Article

AI has broad application across the board in the financial healthcare ecosystem. From preregistration to care, billing and coding to claims processing, AI can be used to:

  • Improve preregistration for providers. 
  • Simplify coding and claims processing workflows for providers and payers. 
  • Assist case managers and doctors with clinical workflows. 
  • Reduce fraud, waste, and abuse for payers. 
  • Act as a virtual assistant for patients to interact with payers or providers. 

However, not all AI investments pay dividends.

First, let’s address the key issue for executives and management in terms of AI investment’s success rate. According to Harvard Business Review, 76% of the organizations surveyed barely broke even on the AI investment. According to two recent Gartner reports, 85% of AI and machine learning projects fail to deliver; and only 53% of projects make it from prototypes to production. Capgemini’s study found that only 27% of data projects are regarded as successful. However, it does not have to be that way. Success rate depends on the approach for AI investments. If it is treated as science, then the probability of failure is high.

One of the main reasons identified for AI failure is that businesses fail to make the program scalable. Change Healthcare has approached solving the investment thesis problem by focusing on healthcare AI programs that can deliver tangible returns in a short period of time (within a year). These AI programs specifically focus on productivity gains such as augmenting staff productivity, reducing the time it takes to do their current task, whether it is reviewing medical charts or processing claim denials. 

Change Healthcare’s approach is focused on inspiring a better healthcare system, and we work with a value-driven mindset for healthcare organizations. We start by looking at AI and machine learning as a way to improve existing business processes rather than as a new business opportunity. We analyze the business processes and determine how improvement in productivity would impact the margins. We start small, focusing on initiatives that can give short- to medium-term returns and then build on those successes. These successes, in turn, can then be expanded into bigger initiatives. Finally, as we build on these discrete individual opportunities—connecting the dots—we then demonstrate how AI can transform complex workflows into more efficient systems. 

Case in point is chart retrieval and review workflows. We start by AI-enabling discrete pieces, such as pre-processing of charts, to verify if the task belongs to one member—separating charts based on service dates, classifying charts to determine if they contain HCC codes (hierarchical conditions categories as defined by the Centers for Medicare & Medicaid Services). Once we achieve success, we then stitch all the pieces together to transform the entire workflow into AI-first. 

A key component toward successful AI investment and scaling is to experiment and fail fast. AI is still a science where data scientists determine if they can glean patterns from the data, so it is a probabilistic approach to problem-solving, unlike a rules-based deterministic approach as seen with traditional application software. Hence, success is not guaranteed. If there is no precedence with the problem being solved, it’s important to experiment to see if there are enough signals in the data and draw proper inferences, which can be mapped to business ROI. Understanding the business success criteria is critical so that one can determine if an experiment meets the established success criteria. 

The other major factor to address early on is all things pertaining to data:

  • Do you have the access rights? 
  • Do you have the usage rights? 
  • Is the data de-identified? 
  • Are all the data sets in a centralized data storage? 
  • Are all the data records correlated? 

You must have the data and the infrastructure to manage large volumes of data. You need legal frameworks to ensure you have the usage rights and that you are in compliance with contracts and industry regulations. De-identification of data is a prerequisite for training AI systems. It is ideal when the organization has a setup to manage the process centrally and has gone through all legal and privacy checks. 

Another key prerequisite is having subject matter experts (SMEs) with the right knowledge and expertise. For example, if the task at hand is to extract diagnosis codes from medical records, you need ample coders in-house to assist data scientists with labeling the data. If it pertains to denials prediction, denials specialists need to share their domain knowledge with data scientists and engineers, where industry rules can be overlayed on top of patterns learned from data to augment the models’ performance.

Finally. There is the matter of choosing the right use cases. This varies from company to company based on opportunities at hand. With 80% of healthcare data being in an unstructured format, natural language processing (NLP) finds its application in many use cases—entity extraction and medical comprehension for AI coding, drawing inferences from clinical notes and images for prior authorization or determining medical necessity for case managers, document processing and filing, and more. Where there are large volumes of claims and remittance data available, that can be leveraged to improve claims processing such as predicting denials, identifying missing charges, prioritizing work queues based on likelihood of payment, predicting days to reimbursement, and more. 

Consumerization of healthcare and the pressures to reduce agent costs in call centers has led providers to roll out voicebots and chatbots for faster, better, and cheaper service. Change Healthcare has been rolling out voicebots on behalf of providers to authenticate patients, thereby saving 20% of average call handle time. Once basic productivity gains are established, businesses can extend the capabilities to perform end-to-end customer functions such as scheduling an appointment, making changes to an appointment, and more without involving an agent. 

To benefit all participants in the industry, once a business has achieved success at scale, we can roll out these features as APIs via Marketplace. For those businesses who may not have a large AI development team in-house or for those channel partners or service providers who may not have access to domain-specific data, they may want to take advantage of these APIs to augment their solutions with AI-based features. Availability of such APIs allows businesses to retain their investments in existing workflows while taking advantage of new AI-based features. In summary, businesses can maximize their return on investments in AI initiatives by establishing success criteria prior to considering any AI development and ensuring that the relevant data needed is available and can be used.