Before You Buy Another AI Tool, Find the RCM Problem Worth Solving
In this second and final post of a two-part series, I build on the conversations explored in Part 1 to examine how radiology practices can identify the right revenue cycle challenges to address—and determine where AI can deliver meaningful value.
One question came up in different ways after my Mid-Atlantic Radiology Business Management Association (RBMA) presentation: Where do we start?
I understand why practices are asking. There are more AI capabilities to consider, more claims about what they can do, and no shortage of revenue cycle work that could use attention. But buying a tool before defining the problem puts the decision in the wrong order.
Start with the work. Then decide whether AI belongs in the solution.
A Denial Is a Signal, Not a Diagnosis
Denials and appeals drew particular interest when I spoke with attendees. That makes sense to me. A denial can delay reimbursement, and preparing an appeal can require staff to gather information from several places, understand the payor’s reason, assemble support, and act before a deadline.
But I do not want a team to stop at “we have too many denials.” A high denial count tells you something needs investigation. It does not tell you why it is happening.
Are the same issues recurring with a particular payor, procedure, location, or referring source? Does the problem begin with an order, eligibility information, authorization, documentation, or a claim edit? If we only get faster at working each denial after it arrives, we may miss the chance to prevent the next one.
This is where analytics and AI can work together. Analytics can help a team see a pattern and investigate its cause. AI may help interpret information, prioritize exceptions, or support parts of a complex appeals workflow. But neither is a substitute for understanding the process and assigning someone to act on what the information reveals.
Give the Technology a Defined Job
I would begin with one consequential problem and bring in the people who know that workflow best. Map what happens today. Find where time is lost, information is missing, or work is repeatedly sent back for correction. Then ask what kind of intervention fits.
Some work may need a clearer process or better data. Some repeatable steps may be handled with established rules and automation. Other work, particularly when information varies or staff must sort through documents and exceptions, may be a better candidate for AI assistance. The point is not to use the most advanced tool available. It is to use the right approach for the job.
Take appeals. A focused test might examine whether an AI-supported workflow helps staff assemble the information needed for review more quickly and consistently. It should also make clear what a person checks, what happens when information is incomplete, and how the team identifies an error before it moves further through the process. Those are part of the workflow, not details to figure out after launch.
Measure the Result, Including the Work That Remains
Before a pilot, I would want to know the starting point. How long does the work take now? How many items require manual handling? What happens to reimbursement, cost, turnaround time, or patient service? Without a baseline, it is hard to separate a useful improvement from an impressive demonstration.
Then I would compare the results with the full cost of making the change. If the process is faster, what happened to staff time? Was that capacity put to better use, or did review and rework consume it? If revenue improved, how does that compare with the cost of the solution? A change might still be worthwhile if it addresses a serious workflow or patient-service problem, but leaders should be able to say clearly what value they are buying.
I would also look closely at security, compliance, accountability, and the people who will manage the process. Early on, staff may need to review outputs closely. As the workflow develops, their role may shift toward monitoring results, handling exceptions, and finding problems the technology cannot resolve. That is not a case for taking people out of the picture. It is a case for making better use of their judgment.
Start Small Enough to Learn
The advice I gave at Mid-Atlantic RBMA was simple: pick a problem you can define, choose a piece of it you can measure, test the approach, and learn from the result before expanding it.
Radiology practices do not need to wait until every question about AI has been answered. They also do not need to apply it everywhere at once. They can begin with a real source of revenue cycle friction, understand its cause, and ask whether a carefully governed change improves the outcome.
That is a better starting point than asking what an AI tool can do. It asks what the practice needs to accomplish.
Find the Problem Worth Solving First
If you are unsure where to start, begin by finding out where revenue cycle friction may be costing your practice time or revenue. XiFin’s RCM Opportunity Explorer offers a research-informed, directional view across revenue recovery, operational efficiency, and patient engagement and access. A guided Revenue Opportunity Assessment can then test those assumptions against your own data and help identify which problem is worth addressing first, whether the answer involves process change, analytics, automation, or AI.