A young college student receiving Ocrevus, a drug used to treat multiple sclerosis, came to one of our ambulatory infusion centers after being treated in another city. We did the work that should have cleared him for care. Our team called his Blue Cross Blue Shield plan, verified benefits, obtained prior authorization, confirmed medical necessity, and scheduled the infusion.
Months after we treated him, the claim was denied.
The payer said the plan had a specialty pharmacy carve-out. Under that arrangement, the drug was supposed to come through a designated specialty pharmacy, often with pressure to move the patient to home infusion. The carve-out had not been disclosed during the benefits call. More confusingly, the payer had approved the authorization.
We later learned that the same thing had happened at the patient’s previous infusion center. Their team had also received an approval, infused the patient, and then been denied payment.
Because we had an authorization on file and the patient met medical necessity, the manufacturer agreed to replace the drug. That protected us from the full acquisition cost, but we could not continue treating him. The manufacturer eventually worked with the insurer to pursue an exception, pointing out that a college student living in a dorm room was not an appropriate candidate for home infusion.
That case changed how I think about infusion benefits verification. Intake can appear successful at every visible checkpoint and still fail months later. An authorization number is important, but it is not a guarantee of payment.
Authorization is not a payment guarantee
An authorization can fail to translate into payment for several different reasons, and the problem may not become visible until after treatment. In my experience, the most common breakdowns fell into three categories: the payer provided incomplete or incorrect information, the patient’s benefit or coverage rules changed, or the treating provider was not properly credentialed. Each creates a different operational problem, but the financial risk is the same. The organization treats the patient believing the reimbursement pathway has been cleared, only to discover later that one part of it was not.
Incorrect payer information was one recurring source of risk. We would sometimes call and be told that a patient owed only a small specialty copay. Based on prior experience with the plan, we knew that answer was unlikely. The team would call again, speak with another representative, and sometimes receive a different answer.
At one point, I estimated that roughly one in ten benefits investigations returned information we believed was wrong, even with callbacks. That is not a formal benchmark, but it reflects what we saw. The staff members who caught these errors were relying on pattern recognition developed over many calls with the same payer.
That tribal knowledge was our safety net. It was also evidence that the process was fragile.
In infusion care, a small administrative miss can create a five-figure loss. A specialty pharmacy carve-out may not surface until after administration. A provider may see a patient before the payer finishes credentialing that provider. A drug may move to a self-administered list while patients remain on the infusion schedule.
We experienced the last scenario with a medication that had routinely been covered in the clinic. The drug was later reclassified as self-administered, which meant doses given in the clinic were no longer covered. But before every clinic had identified the affected patients and removed them from the infusion schedule, a few still received injections that would not be reimbursed. At roughly $15,000 per injection, two missed patients could create about $30,000 in write-offs.
Credentialing creates similar exposure. Some payer contracts are facility-based. Others require each provider to be credentialed. Credentialing can take 30, 60, or 90 days. If a new nurse practitioner sees a patient and the claim is billed under that provider’s name before enrollment is complete, the payer may treat the encounter as out of network even though the facility is contracted.
These examples look different, but they point to the same operational lesson: prior authorization confirms only one part of the reimbursement pathway. It does not necessarily confirm the required site of care, the correct drug sourcing channel, the provider’s network status, or whether the drug still falls under the same benefit. “We have an authorization” and “we are in network” are therefore not sufficient operating answers. Before treatment, the organization must confirm that the facility, provider, drug, site of care, and sourcing method are all valid for that patient’s specific plan on the date of service.
Treat denials as process data
We reduced these failures by treating denials as process data rather than isolated billing problems.
Our billing team brought a denial list every week. We reviewed each drug-payer combination and asked what caused the failure. Was the payer information wrong? Did intake miss a necessary question? Had a policy changed? Was the provider still pending credentialing? Did the authorization cover the drug but not the site of care?
After the Ocrevus denial, we added a required question to the benefits investigation workflow: “Is there a carve-out to a specialty pharmacy?” It sounds obvious in retrospect. It was not part of the standard call script before the denial exposed the gap.
We used biweekly intake meetings to distribute changes to the standard operating procedure. The cycle was straightforward: identify the root cause, update the workflow, train the team, and watch for recurrence. Over time, individual failures became organizational knowledge.
The difficult part was consistency. Updating a policy document is easy. Making sure all intake staff members remember a new question while handling calls, patients, faxes, and portal work is harder. Even experienced people omit steps when interrupted.
Make operational knowledge executable
This is where I see practical value in AI-assisted benefits verification. The main benefit is consistent, auditable work; speed is secondary.
A human team member may document a reference number and the payer representative’s name, but that record often does not show exactly what was asked or answered. A call transcript changes the post-denial investigation. When a payer later cites a carve-out, the team can confirm whether the question was asked, what the payer said, and whether the authorization conflicts with the denial.
That record can also strengthen an appeal. A transcript and authorization do not ensure recovery, and a low-value claim may cost more to pursue than to write off. For a high-cost biologic or repeated infusions, however, the evidence can materially improve the organization’s position.
Automation also makes process changes easier to scale. Once an operator learns that a question must be added, an AI agent can apply it to every relevant call. A bot does not forget because the front desk is busy or assume a familiar plan works as it did last month.
At Mandolin, the benefits verification product combines call execution with transcripts and audit trails. The useful principle is broader than any product: operational knowledge should live in a system that can apply it consistently, not only in the memories of experienced employees.
Human intake teams will remain essential. Edge cases require judgment and coordination with patients, manufacturers, payers, and providers. The opportunity is to remove repetitive work and give those teams better evidence when something goes wrong.
Resilience does not mean eliminating every denial. It means being able to explain one, change the process quickly, and prevent recurrence without relying on someone to remember what happened last time.
