Demonstrating ROI, the Challenge for Healthcare AI

Demonstrating ROI, the Challenge for Healthcare AI

Today, there is a great deal of information available about artificial intelligence (AI), and we find it applied to most all aspects of our life, including healthcare. AI was first used in the 1970s in the biomedical space. Over the years, the use of AI expanded throughout the healthcare industry, resulting in improved patient outcomes, increased efficiency, and cost savings.

The idea that AI can help reduce healthcare spending is not new, a recent research paper estimates the broader adoption of AI could lead to savings between 5% and 10% in healthcare spending, or roughly $200 billion to $360 billion a year. While these numbers are significant, demonstrating that return on investment, or ROI, can be challenging.

We know healthcare resources are becoming increasingly scarce and the utilization of those resources to maximize both patient outcomes and enhance financial performance is paramount to the future success of any healthcare institution. Organizations need to understand how they will deploy AI and the expected return on investment if they are to utilize AI efficiently and effectively in the healthcare setting.

Traditionally when thinking about AI deployment, we think of each individual AI algorithm as a standalone instrument that works to:

  • Enhance the quality of patient care
  • Bring efficiency to workflows
  • Remove the manual investment in repetitive tasks, decreasing the time and resources required

But is that how we should think about implementing healthcare AI? Does the individual algorithm approach work for hospitals? Is it the best way to understand AI ROI?

Challenges in Calculating ROI When Using AI in Healthcare

Evaluating AI ROI based on the resources utilized to identify, evaluate, and deploy AI is at the base of the ROI calculation. ROI can be seen as the net revenue of the investment divided by the cost of the investment.

Sadly, many single-point solution AI vendors fall short in representing the actual costs involved, thereby misrepresenting the ROI. We can assume the vendors don’t intentionally misrepresent the ROI, but the biggest concern is that they give the ROI calculation based on only the cost of the algorithm being deployed, not the true costs based on resources invested by the deploying entity in the identification, evaluation and finally deployment of that AI.

In assessing the value of the algorithm and its deployment, most organizations, when looking at the cost of deployment for an ROI calculation, look only at the cost of the actual deployment of the AI. They don’t look at the bigger picture and the resource investment the organization made to get them to that decision. To demonstrate this situation, let’s use an example of an ROI provided by a vendor.

Healthcare AI Vendor ROI Example

The AI is going to cost $500,000 to deploy, and the algorithm is going to generate $600,000 in total revenue based on the assumptions of the vendor. The ROI calculation, therefore would be the projected net revenue of $100,000 / the cost of deployment $500,000 equaling a 20% ROI. But 20% is not the real return on this investment because the $500,000 upfront cost quoted by the vendor for the AI does not represent the full cost of resources invested by the deploying organization to identify, evaluate, and deploy that AI.

What Does Need to be Included in the AI ROI Calculation

Normally within an ROI calculation, the vendor is only going to look at the cost per scan or the total cost of deploying that AI. The vendor is going to put revenue in the calculation demonstrating hospital inpatient revenue, hospital outpatient revenue, facility fees, chemo/infusion services, radiation therapy services, pharmacy fees that are not chemo related, professional fees, and imaging services all are quite common in that calculation. But what they don’t include are the internal organizational costs of the administrative personnel’s time in identifying, evaluating, and deploying that AI. Whether it’s the CEO, CAO, CFO, CIO, legal, or other administrative support personnel, their time has value, and that is not normally included by the vendor as they are looking at calculating the ROI.

Administrative costs in some organizations can be as high as $8,000 to $10,000. This might not be significant when considering a 500,000-dollar investment, but this is still a cost that has been incurred by the organization when deploying AI. In addition to administration, you have an investment of time by clinicians that are evaluating the AI. These clinicians will experience workflow interruptions at multiple stages of the AI evaluation and deployment process; there are impacts on clinician and department productivity, and there are changes to processes and workflows, which all add up when assessing the overall deployment process. The cost of the clinician’s time in this process can be as much as $100,000 to $125,000 in time, lost productivity, and interruptions to workflow.

You also have IT personnel involved. IT leads, network engineers, interface engineers, and other IT support staff all have time invested. There may be hardware costs as well as hardware maintenance that has to be considered. Often additional software and interface development is needed along with IT and network maintenance for system integration and integrity that must be addressed, and then there are cybersecurity concerns that have become a challenge in today’s environment. All of these personnel and activities have costs, and in many organizations, we see these costs in the $75,000 to $100,000. range.

You also have clinical support staff, personnel, and technologists that are impacted by the AI evaluation and algorithm deployment. These personnel costs can add up as well, whether it’s the APPs, the technicians, and/or the support staff; these costs come in anywhere from $25,000 to $50,000 by the time we account for all the clinical resource investment.

The True ROI Picture

If you stop and consider all of the costs necessary to identify, evaluate and deploy an AI algorithm as have been outlined above, it’s not simply the cost of the algorithm. In the case outlined above, we’re looking at the fact that if the AI algorithm is deployed for $500,000, there can be upwards of $250,000 to $350,000 additional expenses through internal resource utilization that are not accounted for in the AI calculation provided by most vendors.

For the true ROI calculation of a single AI algorithm, we can use the same revenue projections that were provided by the vendor, but we also need to include the internal resources that were utilized by the organization. Now, the total cost of deployment, with the internal resource utilization included, is not the $500,000 outlined by the vendor, but actually closer to $750,000. With a generated return or revenue of only $600,000, we’re looking at $150,000 loss or a negative ROI moving from a +20% ROI to a – 30% once all costs were truly considered. This should be a major a concern for an organization considering deploying AI.

A consistent process for AI identification, evaluation and deployment may result in decreased costs associated with internal resource utilization, however there will be internal resource utilization in all of the areas outlined above and these will not be accounted for in a vendor produced ROI calculation.

Looking Beyond Single Point Solution AI

Ultimately single AI deployment will become one of the most expensive and resource intense ways to deploy healthcare AI. Why? Because the healthcare organization will have to maintain all of the interfaces, all of the interactions, and all of the activities associated with the utilization of the AI algorithm.

There is a different way. If healthcare organizations deploy in a true platform scenario, the entire ROI calculation is altered. In a platform environment, it is not the first AI algorithm that results in a return; it’s the second, third, fourth, and additional AI algorithms deployed where the ROI builds. The cost components of all subsequent AI deployments are impacted, and these costs decrease.

Thinking of IT, there are no longer multiple interfaces to develop and maintain – there is only one. Clinician and support staff time investment is decreased because of the consistency with which the AI is deployed and utilized. Administrative time invested is decreased due to the singularity of contracting. 

With this approach of deploying multiple algorithms via a true platform that is vendor agnostic (positioning and deploying algorithms that are developed by vendors other than the platform vendor), the healthcare organization can enjoy improved patient outcomes by deploying the best algorithm for the organization’s particular practice, patient population, equipment, and disease presence within the population the health facility serves. The ROI changes as well, going from a 3-year calculated ROI loss with a single AI deployment approach to an actual positive net return utilizing the same calculations from above.

If you can look at the calculation above, the ROI of a single AI algorithm over 3 years is minimal.

Under a platform deployment model, you can deploy multiple algorithms at a lower cost and reach a point where your ROI can be in substantial.

These numbers are simply for demonstration purposes, but they do represent the factual possibility of improving the ROI of AI deployment through the use of a true platform vendor.

Healthcare organizations will benefit from developing a solid AI strategy that includes a true platform vendor as an AI partner. Without investing the time necessary to develop an AI strategy, far too many healthcare institutions will find their decision to deploy single-point AI solutions will create financial stress for their organizations.

Picture of Ferrum Health

Ferrum Health

Simplifying Healthcare AI - Ferrum's AI Hubs enable clinicians to focus on their patients and the care they need, not complicated technology deployment.

Contact Us

CASE STUDY

ARA Health Specialists

Use the button below to download your free case study and learn how our approach to validation has improved the number of clinically significant findings in AI software.

CASE STUDY

Sutter Health

Use the button below to download your free case study and learn how our approach to validation has improved the number of clinically significant findings in AI software.