Validating AI Investments in Radiation Oncology

 

Workflow fit, unbundled pricing, staffing reality, and what AI costs after the demo.

 

Medsolve Dynamics validates AI claims in radiation oncology against workflow fit, staffing reality, and what the software actually costs once it is unbundled.

 

Short field notes, recorded as they came up. Newest first.

 

 

Price the AI, Not the Demo: Two Questions Before You Buy Theranostics Response-Monitoring Software

July 13, 2026 · Listen

Yoel Bakas, computer science and an MBA in artificial intelligence, twenty years configuring and pricing this class of technology from the vendor side, on the two questions radiation oncology and theranostics programs should ask before buying AI response-monitoring or workflow software: does it have a billing pathway (or is it an operating expense with no code to bill against), and who owns the output when the software touches a clinical decision. For hospital and freestanding radiation oncology administrators, CFOs, and theranostics program leads evaluating AI, software renewals, and capital vs operating spend.

 

Why AI Efficiency Disappears Between the Demo and the Staffing Model.

June 11, 2026 · Listen

Vendor AI demos show real efficiency that never reaches the staffing model. The workflow redesign and oversight cost behind AI value in radiation oncology.

 

Why AI Is Not a One-Time Purchase in Radiation Oncology

June 4, 2026 · Listen

AI purchasing differs fundamentally from traditional capital and software decisions. The cost is not the license, it is the recurring monitoring as the model drifts in ways no tolerance table catches. From hype to value: how to evaluate AI for long-term cost structure, clinical accountability, and organizational readiness, not the demo.

 

Ask The Vendor What The Same Software Costs Without The AI. The Unbundled-Price Question

May 21, 2026 · Listen

From 20 years on the vendor side of oncology software sales, the one question that exposes the AI price premium baked into software renewal quotes in 2026. Treatment planning systems, contouring software, registry products, SaaS modules. The AI features were priced higher than the same software would have priced two years ago. Get the unbundled price.

 

When I Was On The Vendor Side: The AI Demo That Skipped The Workflow Question

May 19, 2026 · Listen

A vendor-side story Yoel Bakas has watched play out dozens of times in radiation oncology. A software demo runs without anyone asking what is actually broken in the customer's workflow. The tool gets relabeled AI-enabled a year later. Until the customer understands what is broken, it does not matter what the tool is. AI in radiation oncology, workflow design, vendor evaluation, capital decision.

 

AI Procurement Accountability in Healthcare: 3 Questions Before You Sign

April 14, 2026 · Listen

The conversation highlights the pitfalls of AI procurement, the consequences of procurement failure, misleading claims in AI sales, and key questions for procurement. It emphasizes the importance of considering adoption requirements and the impact on workflow when purchasing AI solutions.

• Procurement failure.

• Adoption requirements.

 

Defining the problem before pricing is crucial.

April 1, 2026 · Listen

The conversation emphasizes the importance of answering key questions before requesting a quote from a vendor, highlighting the vendor's influence on unanswered questions and the significance of defining the problem before pricing. It also mentions an upcoming presentation at the AMAC® Spring Symposium.

• Answer key questions before engaging with vendors.

• Defining the problem before pricing is crucial.

 

Who supports an AI tool when it flags something your team cannot explain?

March 17, 2026 · Listen

The vendor selling an AI tool often did not build the model underneath it, so support for an unexplained output rarely runs where hardware support runs.

That gap arrives at a specific moment. It is a Friday evening, the algorithm has flagged something nobody on the team has seen before, the physicist cannot account for it, and the vendor support line has no answer. Three things settle this before signature. Who picks up the first call when the software does not behave as expected. How the tool reads from and writes back to the EHR or OIS. Who validates an output after the physicist flags it. A tool being extraordinary does not solve the support problem. It makes it harder, because the better the model, the less often anyone has practiced the exception.

Ask the vendor to walk through exactly what happens the first time something goes wrong. The answer tells you whether a support path exists or whether you are it.

Validating the IT and physics support path before signature is the read Medsolve runs on an AI or software agreement.

 

What should you answer before an AI purchase goes to signature?

March 17, 2026 · Listen

Two answers decide an AI purchase. Whether the model was validated on data matching your treatment and modality mix, and who still reviews and signs off after it runs.

AI adoption does not start with a demo. It starts with a map. Most published benchmarks are built around the highest use case, and the highest use case is usually not yours, so an accuracy figure validated on another program's modality mix is a number about their practice. The second answer is the one programs skip. If the review and sign off stay where they are, the tool adds cost without moving margin. The test is sixty seconds. If those two answers do not come inside a minute, the questions have not been asked yet.

Ask them before the paperwork, not after.

Validating a model's accuracy claim against a program's own modality mix is the read Medsolve runs before signature.

 

The challenges and considerations related to the implementation of AI tools in Oncology programs.

March 17, 2026 · Listen

The conversation explores the challenges and considerations related to the implementation of AI tools in programs. It delves into the evaluation framework, task automation, liability, KPIs for success, workflow changes, and the impact on schedule and review processes.

• AI tools require a clear evaluation framework.

• Workflow changes are essential for AI implementation.

 

Why does AI that speeds up one step not reduce the total work?

March 17, 2026 · Listen

AI that speeds up one step does not reduce total work unless the review and sign off behind it change too, because the bottleneck moves rather than disappears.

The 2026 reimbursement change is the largest radiation oncology has seen in over a decade, and tighter margins make the system less forgiving of this particular miss. Programs are looking at AI to protect capacity without adding headcount. The trap sits in what happens after the fast step. If the downstream review and sign off stay as they are, work does not decrease. Sometimes it increases, because a faster upstream step delivers more cases into the same unchanged queue. In 2026 math, a small workflow miss becomes a large financial one. AI can protect margin, but only when the workflow is rebuilt around it.

Before the purchase, map which step gets faster and which step absorbs the output.

Rebuilding the workflow around the tool rather than around the demo is the work Medsolve does on an AI deployment.

 

More field notes: Theranostics & service line growth · Program operations · all seven topics