By John Pyhtila, PhD, Chief Product and Strategy Officer and Gokul Varadharaj, Chief Revenue Officer & Co-Founder

Healthcare organizations have high hopes for generative artificial intelligence, with roughly half of payers and providers planning to invest between $1 million and $10 million in GenAI and about 70% doing so within the next two years, according to a 2024 HFS Research survey of 350 payer and provider CXOs.

With this spending, only 20% of payers and providers have a vision for GenAI that is influencing their investment decisions, according to HFS Research, raising the question of how to prioritize GenAI investments for a strong ROI.

As payers consider investments and use cases, four requirements should form the foundation for GenAI exploration and development. These requirements are critical for building successful payer-customized models that leverage open-source GenAI solutions with company-specific use cases and proprietary data to deliver enterprise impact. Payer-customized models, described in the first blog of this series, that are based on these requirements will move payers beyond the GenAI pilot stage and into long-term adoption, enabling both important early wins and sustained enterprise success.

The four requirements for GenAI development are Trust, Security, Scalability and Data Usability, and they apply to the entire tech stack from data infrastructure and models to applications as well as to both LLM (large language models) and LMM (large multi-modal models).

Four critical GenAI development requirements for payers

1. Trust

Without trust, organizations will never gain traction with their workforce to maximize GenAI impact. As part of the broader AI strategic plan, payers need a formal framework that ensures trustworthy GenAI solutions:

• Will work and add business value.
• Won’t do harm.
• Understand and minimize bias in the solution.
• Respect and manage workforce impacts.

As with any new technology, GenAI will not be 100% accurate, and therefore a cornerstone of initial use cases is keeping humans in the loop, standing up an effective governance framework, and not focusing initially on business-critical functions.

For effective governance, payers should establish a cross-functional governing body made up of both technology and business executives who have clear decision-making authority for both development and deployment of GenAI solutions.

In this early phase of the GenAI revolution, the governance committee should focus on high-value use cases that at the same time don’t have the potential to negatively impact critical business functions. An example is creating a payer-customized GenAI model to automate a highly manual process, but if an error occurs, it doesn’t negatively affect member care or financial performance.

With the initial use cases and models, payers will be institutionalizing significant learnings. As they become more comfortable with GenAI and can better manage risks, payers can then start to deploy solutions for more business-critical functions.

To ensure harm isn’t done, payers will need appropriate subject matter experts reviewing and validating output. If an error is found, they will need a feedback loop so the error is addressed and doesn’t happen again.

The governing body needs to keep up with the rapidly changing world of GenAI to inform company direction. It also must monitor the impact of GenAI to measure business value while maintaining trust throughout the organization. In the process, governance must address enterprise change management so employees can adapt to a world where using GenAI solutions is the norm.

2. Security

Strategic GenAI solutions will involve highly sensitive information including company confidential material, personal health information (PHI) and personally identifiable information (PII). Payers will need this data to train/tune models and for queries to the models.

As payers develop customized LLMs and LMMs and begin to use those solutions, they must work in environments that are HITRUST/SOC2 certified to ensure the data and user queries are kept safe and private. Developing within these parameters and working with data and technology partners that also adhere to these strict standards will go a long way to protecting PHI/PII and other proprietary data.

3. Scalability

No single GenAI model can effectively meet all needs. As more use cases are added, a single model will become less accurate for any specific use case and become much harder to manage and scale into new use cases.

Instead, payers should focus on developing targeted models to solve specific problems or support specific tasks. As these models are developed, weaving them together in a model mesh, with a master LLM that routes queries to the appropriate LLM, will provide scalability.

To achieve enterprise reach and impact, LLMs/LMMs must be engineered to leverage bigger amounts of varied healthcare data. GenAI is highly effective at generating valuable new insights based on unstructured data including text (e.g., clinical notes), images, and SQL/code. To maximize the value generated from GenAI, payers also need to add their core structured data, such as claims, membership, and labs, which is a bigger lift.

Cost is also a major consideration when training, tuning and running LLMs. For any GenAI effort, payers must ensure that the approach is efficient and optimizes performance vs. cost curve. Additionally, payers must put in place capabilities to actively and automatically monitor and manage costs.

4. Data Usability

“The big challenge with GenAI is to clean up enterprises’ messy data so they can benefit from the tools. Otherwise, GenAI becomes lipstick looking for a pig.”

— Phil Fersht, CEO and Chief Analyst, HFS Research

Data is the critical foundation for GenAI to be successful. A full 72% of leading organizations say managing data is one of the top challenges preventing them from scaling AI use cases, according to McKinsey.

More than 95% of executives believe their companies would be more competitive, more innovative, and able to make faster decisions if their data quality was two times better, according to 2023 HFS Research. However, just 8% said their data was at least 91% to 100% usable compared to  53% who said just 60% or less of their data is consumable and can be operationalized. For 26% of respondents, the percentage of usable data is below 40%.

Not surprisingly, then, nearly 40% of a variety of insurance companies say investing in cloud and data as foundational technologies to realize AI promise is the top priority, according to a 2023 Gradient AI survey.

In all cases, the key is making data usable, which means it is accurate, complete, timely, relevant, versatile and use case and application agnostic. To transform data so it’s usable requires investments in core data infrastructure, architecture, integrity and the associated data governance framework.

GenAI development requirements drive success

Payer CIOs are wildly enthusiastic about GenAI, picking it as the top game-changing technology in the next three years, according to the 2024 Gartner Healthcare Payer CIO and Executive Survey. And with good reason: GenAI may rival the transformational impact of other technological breakthroughs, from the printing press to the Internet.

To date, most technological breakthroughs have largely impacted manual labor by driving automation. In contrast, GenAI will significantly transform white-collar knowledge workers by reshaping how their work gets done and driving vast efficiencies not previously thought possible.

Even as generative AI continues to rapidly evolve and improve, these four requirements will keep payers on track to develop secure, trusted payer-customized solutions with a feedback loop designed to deliver increasing value to their organizations. With a development program built using the guardrails of trust, security, scalability, and data usability, payers will lay the foundation for future transformation and accelerate and magnify their benefits from GenAI.

Contact us at info@abacusinsights.com if you’d like to learn more about getting the usable data you need to power GenAI and strategic initiatives.

This blog is second in a series about the use of GenAI by Payers. Look for the third blog, about the criteria for determining high-value use cases for payer-customized GenAI.