By Bart Geerts

Estimated Reading Time: 7 min

Artificial Intelligence (AI) is currently viewed as a silver bullet in modern medicine, promising to improve outcomes and the experience for staff and patients, reduce workload, and cut costs. But while the potential is huge, the road from concept to clinical value is long—and in most healthcare institutions, still largely experimental.

That’s okay, it’s necessary.

AI in healthcare is still in its early stages, and most hospitals need time to learn how to use it effectively. Although there has been notable progress in certain regions, particularly in the field of radiology. AIFI is a great local example.

Unlike a lab test or a new surgical device, implementing AI often introduces new or different workflows, responsibilities, and expectations—sometimes even new roles. Institutions need to learn and iterate to get this right. We need to give each other time to become ‘AI ready’.

Also, if we do not give these changes the time and structure they need to land, we risk repeating an old mistake: digitising rather than digitalising. We’ve seen this before. Electronic health records and digital clinical pathways were once heralded as game-changers—but in many places, they simply digitised paper processes without rethinking or improving them. The result? More clicks, not better care. If we approach AI the same way—slotting it into old workflows without adapting those workflows—we miss the point and the potential.

If there seems to be a need, start with a business case (BC)

So what should we do before launching an AI initiative? Before purchasing and implementing an off-the-shelf solution or starting with data science and all the technical required tasks yourself, a business case needs to be assessed. One that clearly defines the clinical need, quantifies the current performance of care, and outlines what success looks like—clinically, operationally, and financially.

An effective business case should (and we are focussing on AI tools that er medical devices here) be built on:

1. Proven accuracy of the current way of working: Before implementing a model, do you know your baseline? What’s the current rate of readmissions, complications, or delays? These figures define the starting point and establish the stakes.
2. Proven performance of the AI tool: Accuracy, generalisability, and clinical relevance. Has the model been validated locally or just externally? Does it handle your patient population, your EHR system, your workflows?
3. Preclinical impact assessment or mini-HTA: These frameworks—such as EUnetHTA’s Core Model or the NASSS framework—help assess the value of an innovation before implementation. They give an early indication of whether an AI solution is likely to deliver measurable value in your context.
4. Evidence from similar clinical settings: If the tool has been used elsewhere, what happened? Did it lead to better outcomes? More efficient processes? Real savings? Peer-reviewed evidence or real-world case studies can strengthen your assumptions and help build internal support.

The business case is a shared effort

It’s tempting to hand over the responsibility of ROI calculations to your vendor—but resist that urge. Of course, any vendor should be able to provide you with a general or even more tailored outline. A business case is only as good as its grounding in your own clinical and operational reality. That means involving:

• Clinicians who understand the problem and how care is delivered;
• Operations managers, who understand workflows, throughput, and resource allocation;
• Finance teams, who understand costs, savings, and budget impact.

Clinicians don’t need to become financial analysts, just like they don’t need to become project managers. But some basic skills in understanding and assessing business cases are becoming essential for evaluating health innovations (in general). And just like reading a clinical trial, it’s not enough to take conclusions at face value—interrogate the assumptions, the methodology, the context.

Don’t skip the process: How to build a business case

If you’re wondering where to start, here are some practical steps:

1. Define the clinical problem: What exactly are you trying to improve? Admissions? Sepsis detection? Discharge planning?
2. Map the current process and metrics: Use local data to understand performance—length of stay, complication rates, cost per patient, etc.
3. Clarify the AI tool’s contribution: What does the model predict, how early, and with what accuracy? How would this change clinical decisions?
4. Identify required changes in workflow: Will roles change? Will new alerts be introduced? Will people trust and use the tool?
5. Estimate impact: Use a preclinical assessment, mini-HTA, or published case studies to estimate potential effects on outcomes, costs, and resource use.
6. Calculate ROI scenarios: Best case, worst case, most likely case. Include implementation and training costs, maintenance, and indirect effects.
7. Assign ownership and monitoring: Who tracks adoption, outcomes, and sustainability? How will success be measured over time?

Resources to help you build a BC

If you want to go further, here are some great places to start:

• The EUnetHTA Core Model: https://www.eunethta.eu
• NASSS Framework (Greenhalgh et al.): Helps assess complexity in tech implementation.
• HiMSS Value STEPS Model: A structured way to assess digital health ROI.
• Healthplus.ai Implementation Blueprint: We use a business case calculator that can be tailored easily to cover the local practices and volumes. It is highly transparent and referenced for everyone to check and alter the assumptions in the model (do reach out if you’d like a copy).

Final thoughts

AI in medicine won’t succeed if we just plug it into existing systems and hope for the best. It takes time, learning, and often a reimagining of care delivery. That’s why building a strong, local business case is not just a bureaucratic hurdle—it’s the foundation for success. It is the starting point. It provides alignment between vendor and hospital and forms the basis for the (periodic) evaluation of success too.

Vendors should support this work—but they can’t do it alone. Ask your vendor for a business case, and scrutinise it. Adapt it to your setting. Share it with your colleagues. The future of AI in healthcare is collaborative, and the first step is understanding what value looks like.

Let’s make AI not just promising, but proven.