Somewhere in your program there is software shaping a clinical decision, and nobody can tell you what it costs.

That is not a failure of your finance team. It is how the software got in. The AI came inside a bundle, and it was the reason the bundle was approvable.

I spent twenty years building the vendor side of these packages. The AI module is the easiest line to include and the hardest line to price, and both of those facts are deliberate. Including it makes the proposal read like the future instead of a replacement purchase. Leaving it unpriced means the committee has nothing specific to push back on. You cannot negotiate a line that does not have a number next to it.

Then the capital deal ends and the software does not.

Here is the mechanic. The purchase covers the hardware and carries the software along with it. The machine gets commissioned. Warranty runs. Somewhere in the second or third year the software converts to an annual license, and it lands in the operating budget as a renewal rather than a purchase. It arrives carrying a number for the first time, and it arrives with no baseline, because there was never a separately negotiated price for it to escalate from. The vendor sets the anchor. You did not.

That is a structural problem, not a vendor problem. Nobody lied to anybody. The bundle did exactly what it was built to do.

The renewal table is where this gets expensive. A service escalator on hardware is at least arguing against a known original price. A software renewal on a line that was never separately priced has nothing underneath it. Your side of that conversation is whatever your team can reconstruct from a four-year-old capital file. The rep has the full line item history open in front of him. He is not hiding it. He just kept his copy.

So the evaluation has to happen before the software is in the building, and it starts with one question the demo will never raise on its own.

WHAT IS IT PRICED AT ON ITS OWN

Ask what the software costs as a standalone line, outside the bundle. A number, a term, an escalator. If the vendor has ever sold it standalone to anybody, that number exists, because the rep's own compensation is calculated against it.

If the answer is that it is not sold that way, you have found the line with no baseline, and you have found it before the signature instead of at the third renewal. The move at that point is not to argue. It is to write into the agreement that the software line gets separately stated on every invoice from day one. It costs the vendor nothing, they rarely refuse it, and it builds the baseline you were denied at signature.

That is the question with money attached. The rest of the test sits behind it.

WHO OWNS THE OUTPUT

Eighteen months out, this is the one that costs. If the software touches a clinical decision, it touches your documentation and your liability. A contour the physician accepted. A plan the system adapted. A flag the tool did not raise. Your medical staff bylaws, your peer review process, and your malpractice carrier all have a position on that, and none of them were consulted when the module rode in on a capital request.

WHICH CODE DOES IT GENERATE REVENUE UNDER

Most AI in radiation oncology improves the workflow. It does not move a payment line. Auto-contouring, adaptive planning support, automated scheduling, response monitoring: real work, and none of it generates a code you bill against. That is not a criticism. It is a category.

An operating expense is a fine thing to own, as long as you approved it as one. The problem is that it gets approved as capital and surfaces in the operating budget three years later, where nobody planned for it. The department that inherits the cost is rarely the committee that approved it.

Your vendor may point at the new 2027 outpatient proposal and say software finally has a payment pathway. It has a category, not a pathway. CMS calls the policy interim while it works out how to pay, it moves thirty-six already-paid codes at roughly the rates they already had, and none of the thirty-six are radiation oncology.

THE MOVE

It fits in an afternoon.

Pull your current planning system or OIS renewal and find the AI line. Not the total. The line.

If the AI is broken out with its own price, write that number down. It is your negotiating baseline, and most programs do not have one. If it is not broken out, that is the finding, and it is the more common outcome. You are renewing a price you never set for a capability you never separately approved.

In one review this spring the software line ran the same term as the hardware service agreement, and nothing in the capital file named what the AI cost.

Then, before the next demo, put the questions to the rep in writing, pricing first. In writing matters. A demo answers questions out loud, at the vendor's pace, in the vendor's order, and everyone leaves the room with a different memory of what was agreed.

THE AI PURCHASE TEST

A vendor who answers all three cleanly has a real product, and the questions cost them nothing. A vendor who routes you to a clinical champion has answered a different question than the one you asked.

What programs are doing instead of buying on the demo: they make the AI its own line before the signature, so it has a price to negotiate, an owner in the budget, and a paragraph in the agreement. That is not a harder purchase. It is the same purchase with the parts visible.

The demo told you what the software does. The renewal is what tells you what it costs, and it turns up about three years later, when the price has already been set for you. That renewal is sitting in a file right now with a date on it. Go see whether the AI on it has a number of its own.

The AI module, run against the renewal.

The Pitch: "The AI layer is included. Auto-contouring saves your dosimetrists twenty minutes a case, and it comes with the platform, so there is no separate line to approve."

The Reality: Included means unpriced, and unpriced means there is no baseline when the license converts to an annual renewal in the third year. The twenty minutes saved in dosimetry is real, and so is the time the physician spends editing what the tool produced, which lands in a different department's budget and never appears in the ROI slide. Total planning pipeline time is the number that matters, and it is not the number in the deck. Meanwhile the capability that made the purchase approvable is the one line in the agreement with no independent price, no named revenue code, and no clause saying who owns the output when a contour goes to a physician for signature.

With Heather Turner, RT(T), PMP

The Shiny Object Trap: The operational checklist for buying technology your team will actually use.

Radiation oncology is addicted to the latest and greatest. Every vendor pitch deck makes their new product look like magic. They show you a streamlined, hyper efficient clinical environment, and it's incredibly easy for an administrator to get swept up in the ROI projections.

My transition from radiation therapist to operational leadership taught me a painful lesson: An expensive piece of technology is only valuable if your team actually uses it.

This trap doesn't just apply to heavy machinery in the vault. We see it constantly with high priced software packages designed for dosimetry and physics, such as advanced treatment planning modules, automated contouring tools, or secondary calculation systems. I have seen departments spend six figures on software licenses that end up completely abandoned after three months. Why? Because the technology solved an administrative or marketing problem, but added tedious steps of friction to the daily clinical workflow. If a piece of software makes a physicist or dosimetrist feel like they are working for the computer rather than the computer working for them, they will quietly bypass it and default to their old, manual habits.

Before signing the capital purchase order or software contract for the next piece of shiny technology, run it through this clinical checklist first.

The Footprint Test: Where does the physical hardware live, and how does it change the room mechanics? If a new immobilization device or surface tracking hardware requires the therapists to constantly haul heavy equipment out of a crowded closet, or perform extensive manual indexing for every single patient, they will avoid it at all costs. If it isn't integrated seamlessly into the existing vault footprint and the natural physical movement of the setup, it is a bottleneck hiding behind the promise of an upgrade.

The Software Integration Test: Does this software directly talk to your primary OIS, or does it create a separate program that doesn't communicate with your main system? If a dosimetrist has to manually export a DICOM file, log into a separate third party software, upload the plan, and double document the parameters just to run a compliance check, you haven't bought efficiency. You have bought an expensive error generator that adds administrative labor to a department that is likely already stretched thin.

The Clinical Verification Test: Does the software actually save time across the entire department, or does it just shift the labor from one desk to another? An automated contouring tool that saves a dosimetrist twenty minutes of drawing organs at risk is completely useless if it requires a physician to spend thirty minutes of tedious editing at the screen to correct mistakes before signing off. True efficiency should be measured by the total time a patient's case spends in the planning pipeline, not just the speed of a single step.

The best technology implementations are invisible to the patient and frictionless for the staff. If a new tool doesn't fit the physical and digital reality of your treatment rooms, your dosimetry planning, or your physics QA workflows, leave it on the convention floor.

The answer key

For each of the three questions: what a real answer looks like, what each evasion actually means, and what to do next. One reference PDF. About twenty minutes to use, and you walk out of the demo holding a written record instead of a memory.

Reply or email me at [email protected] with 1 for the pricing question, 2 for ownership of the output, or 3 for the billing code, and I will send you the answer key for that one.

No call from me unless you ask for one.