Artificial intelligence (AI) is undeniably the hottest topic in innovative technology today. Many healthcare vendors are opting to capitalize on this buzz by positioning themselves as “AI companies” with the promise of “effortlessly” transforming health systems’ operations, patient care, and the bottom line.
The allure of streamlined workflows, improved patient outcomes, and data-driven insights generate significant interest among the industry’s leaders and decision-makers, all tasked with doing more for less without compromising care. The cautionary tale is that too often, behind the flashy, state-of-the-art demos lies a complex and costly reality — implementing AI to utilize its full potential and provide a complete solution involves so much more than “throwing tech at the problem.”
Hospitals, health systems, payers, and other healthcare organizations are best served by building sustainable, scalable clinical data management solutions to problems that fit their evolving challenges and budgets. Organizations looking to identify a tailored solution that will evolve with their operations, care delivery, and budget would benefit from understanding the hidden costs associated with purchasing and optimizing an AI platform. Simply purchasing and installing a software solution without a deeper look can be a short-term and costly remedy leading to a dwindling return on investment.
An Engine Isn’t Enough, You Need a Car
AI can and does mean many things to many people. One type of specialized AI that has risen to the top of the industry’s focus is large language models (LLMs). LLMs are a type of AI specialized in processing and generating language. They use deep learning-based NLP models trained on extensive text data to understand and generate language. Too often, organizations focus on simply purchasing an LLM, thinking it will solve their operations, care, and quality challenges.
However, there are hidden costs and additional solutions that should be considered. It may be helpful to consider an LLM as an engine. Having an engine alone doesn’t mean you can take a drive; you need a vehicle to utilize that engine fully. To reach your destination safely and efficiently, it’s essential to choose a car that best suits your journey and your budget. Each Interaction with an LLM costs money. For example, if you use a consumer version of ChatGPT, you’re paying just a few dollars a month to use it as much as you want. However, if an organization uses a commercial version, as a healthcare system would, the system is charged for the number of characters input and output. These characters represent significant data, ideally feeding valuable insights into hospitals, clinics, and payers.
One of the biggest challenges a healthcare organization encounters is utilizing the sheer volume of data that is collected and shared within its ecosystem. Some patient records can reach hundreds of megabytes, with hospitals potentially paying around one cent for every 10KB of data. So, to truly get the biggest bang for their buck, health systems must be able to ingest, understand, and produce actionable insights from this data to enhance care, cost efficiency, and quality outcomes.
While it’s true AI is increasingly an invaluable tool for healthcare organizations, the reality is that simply “throwing” data at an LLM will not get usable, informed, and sustainable results. LLMs aren’t built to read and instantly process terabytes of information in a way that then provides better-informed clinical decisions that improve patient, cost, and quality outcomes. Instead, LLMs have the capability to analyze extensive medical literature, clinical guidelines, and patient data to provide real-time, evidence-based recommendations.
However, a significant portion of the cost associated with AI arises from the need for software platforms that enable LLMs’ to provide clinical data management by filtering, parsing, and categorizing data. These platforms offer the essential infrastructure necessary to derive value from LLMs. Just like having an engine alone is not enough to reach your destination—you need a car.
To extend the car analogy, you would still need to design the vehicle’s powertrain, transmission, braking, fuel systems, cabin, and various features after acquiring an engine. Additionally, you would need to purchase and install wheels, a chassis, and other components, build and conduct multiple crash tests on identical models to ensure safety and register the vehicle once it is approved for road use. That takes time and money, and no performance guarantees exist. There’s also no support; if something malfunctions in your bespoke vehicle, you’ll have to spend the time and money to fix it. Ideally, it won’t break down on the highway in rush-hour traffic.
But what if you bought a complete car? You get the engine and all the necessary parts assembled and evaluated for functionality. Consider that it also comes with a warranty and a service agreement, so if the air conditioning or power seats stop working, you can take your car back to the dealership to be quickly repaired under warranty for free – by people who are trained, experienced, and know what they’re doing.
Scale and Scalability Matter
Finally, let us return to the beginning of our car-building journey. Whose engine do you want your car to rely on? The one from your neighbor, who builds five engines a year out of his garage as a hobby? Or would you be better off buying an engine from an established company that makes 10 million engines a year and spends billions on R&D?
Some AI companies targeting healthcare customers argue that LLMs trained on healthcare-specific data are inherently better than those trained on exponentially more non-healthcare-specific data. Interestingly, research demonstrates the opposite: LLM models built on greater volumes of data perform better.
Further, companies building healthcare-specific LLMs lack the funds and resources to train these models thoroughly. While they could put on a crowd-pleasing demo at a trade show, integrating an LLM into a healthcare IT infrastructure to work at scale in the real world is another matter entirely.
An AI platform for healthcare organizations requires supporting infrastructure that filters and processes user data. In other words, it takes more than an engine to get somewhere; it also takes a car. Do you have the time, money, and experience to build a car from the ground up?
Conclusion
While AI has great potential to transform healthcare, it is crucial to recognize and address the hidden costs associated with its implementation to ensure responsible and effective use. The excitement around AI may pressure healthcare organizations to act quickly in adopting these technologies. However, instead of solely focusing on acquiring a specific technology—regardless of how promising it may seem—healthcare organizations should thoroughly evaluate the hidden costs of that AI solution and its long-term benefits and return on investment (ROI). A more effective approach is to partner with an experienced technology provider that offers a comprehensive end-to-end clinical data management solution tailored to the organization’s unique needs, ensuring successful implementation without leaving you stranded.
About Chris Mazzanti
Chris Mazzanti is the Chief Operating Officer at Carta Healthcare. He is responsible for product strategy, software product development, software development life cycle management, and systems architecture. Mazzanti has nearly 25 years of experience with extensive startup and small company leadership experience in the healthcare and regulated technology space.
