The asymmetry no one prices in
Most radiology AI is sold per study. You pay the same fee every time the tool runs — when it produces a clean, accurate draft, and when it produces a wrong one. The price doesn't move with the result.
That creates a quiet asymmetry. To improve accuracy, you're expected to stack more tools and pay more. But when the AI gets it wrong, you absorb the cost in full — and silently. No one reimburses the radiologist for the minutes spent untangling a bad output. The upside is sold to you; the downside is yours to eat.
When AI is wrong, it's double work — not time saved
Picture the workflow. A radiologist opens an AI-prefilled report. The findings are off. Now they have to re-read the study from scratch and correct the AI output. That's not time saved — it's two passes instead of one, slower than if the AI had never been involved.
This isn't a fringe complaint. In a February 2026 comment letter to the U.S. Department of Health and Human Services, the American College of Radiology — representing more than 40,000 physicians — wrote that radiologists "must spend added time reviewing, validating, and interpreting AI outputs," so the short-term effect is frequently an increase, not a decrease, in workload.
False positives are where the cost accumulates. In one real-world evaluation of a commercial lung nodule algorithm, 39% of flagged findings were not true nodules, at a rate of 1.36 false positives per scan (AJR; DOI: 10.2214/AJR.26.34524). Each false alert is a finding the radiologist has to open, evaluate, and dismiss — paid for at the same per-study rate as a result that genuinely helped.
The rest of software is moving to outcome-based pricing
Outside radiology, pricing is shifting from "pay to access the tool" toward "pay for the outcome it delivers." AI customer-support vendors increasingly bill per resolved ticket rather than per seat; AI coding and sales tools are experimenting with fees tied to merged code or booked meetings. The common thread is charging for the result, not the run. Radiology AI, by contrast, still mostly charges per study — decoupled from whether that study was made faster or slower. The question worth asking is why one of the highest-stakes uses of AI in medicine is among the last to align price with result.
What performance-based pricing could look like
A result-based model for radiology AI doesn't have to be complicated. Three principles cover most of it:
1. A minimum performance threshold, agreed up front
Before deployment, vendor and practice agree on a measurable accuracy floor for the specific task and patient population — not a marketing number, but a threshold tied to how the tool will actually be used.
2. Reimbursement when accuracy drops below it
If the tool falls below the agreed threshold in production, the vendor credits or reimburses. The financial risk of underperformance shifts back toward the party that built and sold the model.
3. No full price for tools that create rework
If a tool consistently forces radiologists to redo work, that is negative ROI. A practice should not pay full price for an output that costs more time than it saves.
| Dimension | Per-study pricing | Performance-based pricing |
|---|---|---|
| What you pay for | The tool running | The result delivered |
| Cost when AI is wrong | Same fee, you absorb rework | Reduced / reimbursed |
| Who carries accuracy risk | The practice | Shared with the vendor |
| Incentive for the vendor | Volume of runs | Quality of output |
| Alignment with ROI | Weak | Direct |
Why it all comes back to trust
The real issue underneath pricing is trust. Efficiency gains from AI only exist when radiologists trust the output enough to not re-read everything from zero. Without that trust, you're not buying a productivity tool — you're buying an expensive second opinion that still needs a first opinion.
Performance-based pricing is, in effect, a way of putting that trust in writing. A vendor willing to tie its fee to results is making a statement about how much it believes in its own accuracy. This is why a no-edit guarantee — where the provider stands behind a report and reimburses when a radiologist has to materially correct it — matters more than any single benchmark on a slide. It aligns the vendor's incentive with the one thing the practice actually needs: output it can sign without redoing.
What to ask before you sign
Even before result-based contracts become standard, the framing changes the pricing questions worth asking a vendor: What accuracy level do you commit to, and what happens to my bill if you fall below it? Does your pricing reflect results, or only usage? And do you stand behind your output financially when a radiologist has to materially fix it? The answers tell you who is carrying the risk — you, or the vendor that built the model. (For the full vendor-evaluation checklist beyond pricing, see our 2026 buyer's guide.)
Frequently asked questions
How is radiology AI priced today?
Most radiology AI is priced per study or per case — a fixed fee charged each time the tool runs, regardless of whether its output actually saved the radiologist time. The fee is the same when the AI produces a useful, accurate result and when it produces a wrong one that forces the radiologist to re-read the study from scratch and correct the output.
What is performance-based pricing for AI?
Performance-based (or result-based) pricing ties what you pay to the outcome the tool delivers, not just the fact that it ran. In a radiology context, that could mean a minimum performance threshold agreed before deployment, a reimbursement or credit when accuracy falls below that threshold, and a reduced price for tools that consistently force radiologists to redo work. It mirrors the shift toward outcome-based pricing already underway across enterprise and generative-AI software.
Why does inaccurate AI cost radiologists time instead of saving it?
When a radiologist opens an AI-prefilled report and the findings are wrong, they have to re-read the entire study from scratch and fix the AI output — which is slower than reading from zero. The American College of Radiology told HHS in February 2026 that reviewing and validating AI outputs often increases, rather than decreases, cognitive and workflow demand. The radiologist absorbs that cost silently, because no one reimburses the wasted minutes.
What should imaging centers ask AI vendors about pricing and accuracy?
Ask what accuracy level the vendor commits to, and what happens contractually if the tool falls below it. Ask whether pricing reflects results or only usage. Ask for published accuracy evidence rather than a single headline metric, and clarify whether the vendor stands behind its output — for example, with a no-edit guarantee that reimburses you when a radiologist has to materially correct the report.