By Arunima Rajan

A US hospital is about to hand its insurance appeals to AI. Its CFO is clear that nobody's day is about to get easier.

I asked Daron Hashir how much time her AI system had freed up, and how much work had disappeared from her billing team's week. She said she could not tell me, because the system has not gone live.

That should have been the end of a useful interview. Instead it turned out to be the interesting part. Spanish Peaks Regional Health Center pays about 90 per cent of its claims on first submission, and the rest come back as denials, each needing investigation, documentation and an appeal built from nothing. The administrative staff is small. Big denials get worked first. Small ones sit in a queue until, sometimes, the filing deadline passes and there is nothing left to appeal.

So the hospital is testing Iterate.ai's Generate for Healthcare, which reads a denial against claim data and payer requirements and drafts the appeal. What Hashir has, at this point, is not results. It is expectations, held before any data arrives to flatter or contradict them, and hers are not the ones the technology is usually sold on.

Chief financial officer, Spanish Peaks Regional Health Center, in conversation with Arunima Rajan.

Before the AI system, can you walk through what the manual appeals process actually looked like day to day, including how billing staff decided which denials to chase and which got dropped due to time constraints?

When I joined Spanish Peaks in February, claims were denied daily and every one of them required someone to manually investigate the denial reason. The process involved researching the denial, reviewing documentation, and assembling the appeal package. About 90% of our claims are paid on first submission, which is fairly typical for a rural hospital for our size. The remaining 10% represent important revenue, but with a small administrative staff, we have to prioritize our efforts. Higher-dollar and higher-priority denials are generally worked first, while smaller denials can remain in a queue. Appeals also have filing deadlines, so if staff don’t have time to get to a denial before that window closes, the opportunity to appeal can be lost. Like many rural organizations, we’re constantly balancing available resources against the volume of work that needs attention.

Now that the AI drafts appeals and pulls supporting documentation, what specific tasks disappeared from your team’s daily workload, and roughly how much time did that free up per case or per week?

We have not yet gone live with the solution, so I can’t speak to actual time savings or operational results. What we’re testing is Iterate.ai’s Generate for Healthcare platform, which is designed to analyze denial information, compare it against claim data and payer requirements, and generate a draft appeal with supporting rationale. Based on what we’ve seen during evaluation and demonstrations, the technology has the potential to reduce the amount of time spent researching denials and gathering supporting documentation. Instead of starting from a blank page, staff would begin with a drafted appeal that they can review and refine. One of the reasons we’re interested in the platform is the possibility of making smaller-dollar denials economically practical to pursue, but we’ll need to complete implementation and gather performance data before we can quantify those benefits.

What new tasks took their place, such as reviewing AI-drafted appeals for accuracy, and how does that review work differ from the origination work it replaced?

We expect the workflow to operate as follows rather than building every appeal from scratch, staff would review AI-generated draft appeals to ensure they accurately reflect the denial reason, claim information, and payer requirements. Human oversight remains critical. The role shifts from creating the appeal entirely by hand to validating the supporting logic, documentation, and recommendations generated by the system. That still requires personnel who understand denial management, coding, and payer rules. The work doesn’t become simpler; rather, we expect it to become more focused on quality assurance and decision-making.

Can you describe "ghost denials," the category of denials your team missed entirely before this system? How did the AI surface them, and what happened once they were caught?

One of the capabilities that attracted us to the platform is its ability to identify potential underpayments, payment adjustments, and other reimbursement discrepancies that may not appear in traditional denial reports. These are sometimes referred to as "ghost denials" because they can represent lost revenue without generating the denial codes our teams typically monitor. During demonstrations and early analysis discussions, the platform showed how AI can review large volumes of claims and payment activity to identify patterns that would be difficult to find manually. Since we are not yet live with the solution, we have not completed the process of identifying, validating, and appealing these cases within our own production environment, but it’s an area we’re particularly interested in exploring.

Has the shift from origination to review changed who on your team does this work, or the skills/seniority needed for the role?

We haven’t gone live yet, so we haven’t made staffing changes as a result of the technology. Our expectation is that the same revenue cycle expertise will still be required. If anything, human judgment becomes even more important because staff will need to validate AI-generated recommendations and ensure they are appropriate for the specific claim and payer. Strong knowledge of coding, documentation requirements, and payer policies will remain essential. The goal is not to replace expertise but to help experienced staff work more efficiently.

Looking back, is there anything about this deployment you’d change or would flag as an unexpected trade-off, something that added friction even as it saved time elsewhere?

Since the deployment is still underway it’s too early to identify meaningful trade-offs based on actual use. One area we’re paying close attention to is ensuring that staff maintain a thorough review process and don’t become overly reliant on automation simply because a draft appears complete. Human validation will remain an important safeguard. Another interesting opportunity that has emerged during discussions with Iterate.ai is the possibility of using AI to identify documentation patterns that contribute to denials and help target physician education efforts. While our initial focus has been revenue recovery, we’re also exploring whether these insights could help reduce denials before they occur.

Got a story that Healthcare Executive should dig into? Shoot it over to arunima.rajan@hosmac.com—no PR fluff, just solid leads.

The Work Doesn't Get Smaller. It Gets Sharper.