Assess viability of buying or building AI 

Why does this matter?

The age-old question of “buy vs. build” is not a simple one when it comes to AI products. It is crucial to comprehend the advantages and disadvantages of both options to ensure the organization makes the right decision. This choice also influences the quantity and variety of resources an organization will need to make the selected approach viable for the organization and the specific problem it is addressing.

How to do this?

Step 1: Scope clinician interest

• Assess whether the organization has sufficient clinician interest. 
• Identify a strong clinical champion for the product who can help guide its development, ensuring that it is appropriately built for use within clinical care.
• If there is not a clinical champion for using AI to address a problem, consider buying over building the product. Lack of clinician engagement may also be due to a lack of understanding of the root cause of the problem. Consider revisiting the guide to determine dimensions of the problem to unpack the problem.

Step 2: Evaluate infrastructure and expertise

• Assess whether your organization has the right data and internal infrastructure to set up, integrate, monitor, and maintain the AI product. 
• Assess whether your organization has machine learning operations (MLOps), quality, systems engineering, regulatory, and compliance experts to support successful implementation. If these capabilities are not housed internally, consider services contracts from vendors or engaging external consultants.
• Assess the level of stakeholder enthusiasm for AI product adoption and how adaptive (or resistant) the organization will be to change. Ensure that AI product use aligns with organizational priorities (see the guide on prioritizing problems).

Step 3: Evaluate a solution against your patient demographic

• Commercially available products are designed to maximize growth and broadly solve problems targeted to a larger market. Evaluate how well a vendor can customize their solution to address the needs of the organization and fit the patient demographic, especially if the patient demographic differs from the AI product development environment.
• In-house developed products tend to be easier to customize with a faster turnaround time. Gain insight into what data is used and how it is applied for building the solution.

“If you have the sample size and the expertise to develop it on your own in your own setting you will tend to do better. The reason being that then you’re not working with a blackbox algorithm. You can know and see what you’re working with.”

Bias Key Informant

Step 4: Evaluate return on investment

• Analyze the costs and benefits of building or buying the AI product.

• Weigh the costs and benefits to evaluate the return on investment. 

Step 5: Consider long-term strategic goals

• If the organization has a solution with a unique competitive advantage, building the solution in-house can differentiate the organization as an innovation leader. 
• If the organization aims to reach a broader patient population, partnering with a vendor may make more sense, especially if the organization has limited time and resources. 

Step 6: Evaluate the impact on the workforce

• Consider potential resistance to change caused by the solution among the workforce. 
• Understand how the solution might affect employee engagement, satisfaction, and morale.

“If [an external AI product is] within a tolerance, they have some predictive value, and they’re inexpensive in terms of what you have to do to purchase and deploy it, then you should do that… You might as well spend your time and resources working through the hard stuff of implementation…”

Technical Expert