Determine if AI should be integrated

Why does this matter?

Until this point in the process, the integration evaluation of an AI tool involves multiple decision points, including performance assessments and feedback from usability testing, as well as evaluating benefits and risks during risk assessment activities. Refer to the guide locally validate prior to integration for details. Once everything related to establishing the success of the product is complete, the organization must consider more than just the product’s performance and safety before making a final decision on integration.

Need to consider ROI. Determine if the product will lead to a return on investment (ROI), which involves a holistic consideration of the resources necessary to deploy an AI product successfully. Calculate the costs associated with implementation, maintenance, and any potential training required to use the tool. Assess if the ROI justifies the resources necessary to implement the tool. Remember the benefit may be non-financial, such as improved patient safety or quality of care.

To avoid sunk cost fallacy. Do not fall into the trap of the “sunk cost fallacy,” which involves choosing to integrate technology simply because considerable resources have been invested up to this point. Make decisions based on the overall usefulness and benefits of the tool, as well as the potential ROI.

How to do this?

To make a final decision on integrating AI technology, follow these steps:

Step 1: Establish consensus among clinical, operational, and technical stakeholders

• Decision-making should include centralized personnel familiar with technology adoption challenges across the organization and on-the-ground professionals familiar with the specific context of use for the AI product under review.
• There should be a consensus between the centralized committee and on-the-ground professionals to adopt the AI product.
• The decision-making process for deciding on AI tool integration should be standardized and updated over time. 

“I will tell you my view of Phase One [of adoption of an AI tool under consideration]. I couldn’t scale this to phase two outside the confines of a clinical trial. That’s how that ended.”

Clinician

Step 2: Examine the intended use of the product

• Evaluate if testing and user feedback meet the healthcare delivery organization’s needs. Refer to guides on design and test workflow for clinicians and define the role of AI for more details. 
• Testing and user feedback could reveal a different root cause for the problem being addressed. Perhaps the problem is simpler, no longer exists, or requires re-prioritization based on new information.
• Ensure that insights generated during local evaluation (described in the guide on locally validate prior to clinical integration ) lead to improvements and refinements in the workflow design and AI product intended use.

“You would not attempt to check the gas tank of your battery operated lawnmower if it wasn’t working. So why would you ever suggest it? Why would anyone suggest that you could train a mode on the dentistry department and think that that no show set of coefficients for the variables in the no show would be the same as for oncology, or for dermatology, or for general practitioners, or any other specialty?”

Technical Expert

Step 3: Evaluate the return on investment

• This includes not just financial returns but also harder-to-quantify benefits like improved patient outcomes or contributions to future research and development activities. 
• Quantifying all the potential benefits of an AI product using a cost-benefit analysis can be challenging. However, consider comparing a technology’s advancements (e.g., improved accuracy in diagnosis, enhanced patient outcomes, reduced wait times) to the costs of additional resources and staffing, as well as the target scale of integration. 
• Assess how easily the tool can adapt to changes in regulatory and clinical landscapes. Would new information render the tool obsolete, necessitating more resources for new integration, or could it be re-trained through an update/change management? (Refer to the guide on determine if updating or decommissioning is necessary)

Step 4: Quantify other returns on investment

• Such as:
  • Reducing clinical burnout by automating clinical decision-making or repetitive tasks.
  • Improving transparency between clinicians and their patients potentially leading to better patient adherence to treatment plans or self-care.
  • Lowering healthcare costs for patients or the healthcare system.

Step 5: Account for ongoing maintenance and monitoring costs 

• Remember to account for costs associated with ongoing maintenance and monitoring, which can demand significant resources and time, particularly for regulated products.

Step 6: Ensure and account for a long-term AI product owner

• Ensure the integrated product has a lifelong connection to an AI scientist and clinical champion to prevent immediate obsolescence. For more details, refer to the guides under monitoring the AI solution under the lifecycle management phase.