Deus Ex Machina: AI for CON Applications?

CON/CHNA/Regulatory

Similar to other use cases, AI tools represent a supplemental resource when performing CON tasks, but are (for now) no substitute for personal experience and human intelligence

Stories about generative AI technology have proliferated in newsfeeds and publications since the November 2022 release of ChatGPT, the latest chatbot from San Francisco-based technology company OpenAI. Users have applied the tool’s ability to produce textual responses for functions that range from songwriting to law school exams, with mixed results.  Within the healthcare industry, AI tools are viewed as one solution to acute workforce shortage and productivity challenges affecting healthcare providers.

At Ascendient, we believe AI will play a critical role in transforming healthcare delivery. Astute organizations will recognize the benefits of AI investment and plan for rapid integration of AI tools to position themselves for success as the industry evolves.  For providers, the use cases for AI investments will likely progress from back-office functions to care transformation models as the clinical workforce gap widens.

With that in mind, we wanted to test generative AI’s capabilities on what we believe is the “readiest” area of our work – CON application development.  The results point to an approach where AI is employed in a supplemental role, while humans continue to provide superior creative and critical thinking abilities.

As a healthcare consultant who has produced successful CON applications spanning six different states, I reviewed ChatGPT’s performance on several types of CON projects to evaluate its potential. After a moderately rigorous ‘test drive’ of the technology, my impression is that the publicly available version of the transformer may be appropriate for initial and nonspecific purposes, but not for creating content based upon specific market knowledge or precedent from earlier CON applications and findings. Thus, choosing highly persuasive arguments will continue to be a human-led process.

I can see five reasons why a team of human specialists should remain your preferred choice for this type of work:

1. ChatGPT is unable to write with a high level of specificity. While CON applications may include general questions that do not require depth of knowledge, ChatGPT’s responses tended to be more cursory outlines and less fully developed arguments. An experienced human writer will be able to recognize when a topic needs further explanation or discussion.

The AI tool did prove to be more helpful in an editorial role. The chatbot suggested answers to CON application questions, but without the requisite details for crafting a successful response. The AI suggestions may help create a preliminary framework for ordering arguments. ChatGPT can also serve as a proofreading resource and ensure sections of text are grammatically correct. Ultimately, the creative writing element must be owned by the human team assigned to the project.

2. Generative AI does not know your market as well as you. As a generalized transformer, ChatGPT is best used for initial themes that do not depend on specific state regulations or market conditions. The tool can offer direction on information and resources for the human team to follow up on, such as demographics, statistical reports, and regulatory guidelines. The team will likely need to identify appropriate examples that support the general arguments ChatGPT provides.

As the project becomes more focused on describing the unique scope of the proposal, the less useful a tool like ChatGPT becomes. ChatGPT comes up short in recognizing the dynamics of a particular area. It is not familiar with the availability of resources, specific service providers, or the level of competitiveness within a market. More robust web scraping capabilities and refreshing the underlying data will have a marginal effect on correcting this blind spot.

3. ChatGPT can’t replace team diversity. Just like an individual human, ChatGPT provides a singular viewpoint for approaching CON responses. A team of CON specialists is likely to possess a broader range of perspectives that in turn produce more solutions. A team of specialists will understand the question’s intent and be able to respond more comprehensively than an algorithm-driven chatbot. Further, ChatGPT cannot offer experience-based strategies such as incorporating content and techniques from earlier applications. Generative AI does not have this sense of recall that would bring to mind successful examples from prior work.

4. AI resources fall short on critical thinking. The responses from ChatGPT were unconvincing for prompts where a high degree of differentiation was needed. A human team will produce far better responses to questions such as “Why was this project selected over possible alternatives?” or “How will this project improve upon existing care resources for this service?”

The most conspicuous shortcoming for generative AI is the post-application review and public comment phase. AI tools do not possess the critical thinking skills and real-time responsiveness required for these deliverables. Experienced CON planners can evaluate competitors’ applications and respond to comments by affected parties in a manner that these tools cannot.

5. There’s no substitute for real-world experience. It is always a sound strategy to collaborate with professionals from diverse backgrounds and functional areas to generate a full set of ideas about how best to respond to CON questions.  The experience of your project team remains the best intelligence source for composing a CON application.

The goal of the CON planner should continue to be assembling teams of subject matter experts with relevant operational, regulatory, and strategic experience. Some fortunate healthcare providers may possess this scope of knowledge in-house. For those requiring assistance, consulting experts can facilitate the CON application process and locate relevant informational resources. This is the most fitting way to ensure the appropriate knowledge gets applied to the specific challenges of your project.

Conclusion

A successful CON application requires a complicated mix of ingredients. These include data analysis and forecasting, specific market knowledge, familiarity with jurisprudence, eloquence in laying out narratives, and a measure of salesmanship. Synthesizing all these elements requires an uncommon level of expertise and seems like an insurmountable challenge for AI tools ­– for now.

Keep one eye on the horizon for what lies ahead. The initial promise of AI signifies a long timeline of discovering new ways it will benefit healthcare providers. Although ChatGPT does not meet CON planners’ needs right now, specialized versions of generative AI tools may come closer to establishing a role in the highly complex, highly regulated healthcare industry. The swift advances underway and the substantial investment in this sector point toward a short timeframe before tools like ChatGPT become more suitable to CON work.