Radiology Has Two AI Conversations. Both Matter.
In this first post of a two-part series, I explore what recent conversations with radiology leaders reveal about these different perspectives—and what they mean for making informed AI investments.
Spend time with radiologists and revenue cycle leaders, and you will hear plenty of agreement about AI. Both groups see its potential. Both are dealing with pressure to do more with the resources they have. And neither can afford to make decisions based on a polished demo alone.
But they are not asking exactly the same questions.
That became clearer to me after attending the Strategic Radiology Fall Summit and speaking at a Mid-Atlantic Radiology Business Management Association (RBMA) chapter meeting. At the summit, I heard radiologists and practice leaders discuss what AI means for clinical work, staffing, and the future of independent groups. At RBMA, the conversation leaned more toward business operations: where AI fits, what problem it solves, and how a practice would know whether it is working.
The Clinical Question: Does It Make the Work Better?
At Strategic Radiology, declining reimbursement was never far from the conversation. Groups are managing higher costs, recruiting challenges, and growing workloads while trying to remain independent and continue to grow. AI is part of that discussion because leaders are looking for ways to relieve pressure on their radiologists and their practices.
What interested me was the candor around clinical AI. Radiologists described using new tools without necessarily seeing the substantial time savings they had hoped for yet. Their work still requires careful review, and that review takes time. If a tool helps with more straightforward studies but leaves the radiologist with the most complex cases, we have to ask whether the overall workload has really become easier. We also have to account for the cost of implementing and using the technology.
I do not take those questions as a rejection of AI. The message I heard was to keep exploring it, while being honest about what it can deliver today and what may take longer. When the work involves interpreting images and making clinical decisions, the standard for accuracy and oversight has to reflect what is at stake for patients.
The Business Question: Where Do We Start?
At the Mid-Atlantic RBMA meeting, the questions came from a different vantage point. Some attendees were exploring AI but were not sure which workflow to tackle first. Others wanted to understand what a solution could actually do, how much review it would require, and how they could tell whether it had improved results.
Those are good questions. A revenue cycle team may be dealing with incomplete information, changing payor requirements, denials, and appeals that take too much time to prepare. The opportunity is not to put “AI” over all of it. It is to find a specific point of friction and ask whether technology can help the team handle that work more accurately, consistently, or efficiently.
The distinction between clinical and administrative AI is not that one needs oversight and the other does not. Both do. The risks, decisions, and measures of success differ. In the revenue cycle, a practice can define a workflow problem, establish how it performs today, and test whether a change improves turnaround time, staff capacity, reimbursement, or the patient experience. That makes it possible to have a more concrete conversation about value, even while keeping people involved in review and exception handling.
One Practice, One Bigger Decision
These conversations are connected by the same business reality. Independent radiology groups need strong teams and sound operations to navigate reimbursement pressure and keep growing. Clinical and revenue cycle leaders may approach AI from different starting points, but both need to understand the work before changing it.
My view of AI has not reversed since the spring. I remain optimistic about its role in radiology. What I am more conscious of now is the need to be precise. Which work are we asking it to improve? What will people still need to do? What will it cost? And what evidence would tell us it was worth the investment?
Those questions are not obstacles to innovation. They are how we make innovation useful.
In the next post, I’ll take that question into the revenue cycle: how to choose a starting point and determine whether AI has earned a larger role.