AI Prior Authorization Follow-Up for athenaOne Practices

Prior authorization rarely fails at submission. It fails in the two weeks afterward, when a request sits pended at the payer and nobody has an open hour to call and ask what happened to it. The submission is a form. The follow-up is a phone queue, and the phone queue is where earned revenue quietly dies.

The scale of that queue is now measured. Practices complete 40 prior authorizations per physician per week and spend 13 hours a week completing them, and 40% of physicians have staff who work exclusively on prior auth (AMA). The same survey names the method: phone is the most commonly used way physicians complete prior authorizations for medical services, and only 24% report that their EHR offers electronic prior auth for prescription medications. The work is a person on hold, several times a day, every day.

The follow-up gap is a staffing gap, and the survey says so

The most useful number in the 2025 AMA survey is not about denials. It is about why denials go unchallenged. Among physicians who do not always appeal an adverse determination, 52% report insufficient practice staff resources or time (AMA). Not disagreement with the payer. Not a judgment that the appeal would be wrong. Hours.

That reframes the problem for an administrator. If the constraint were clinical or legal, software would not help. The constraint is that structured follow-up is repetitive, unglamorous, and always loses to whatever is ringing right now. An authorization coordinator with a hundred open requests works the ones a patient called about today, and the quiet ones age.

This is what practices mean when they say the backlog is frozen revenue. A pended authorization is work already done, sitting in a queue, waiting on a status call nobody has time to place.

Where prior auth actually breaks inside athenaOne

Generic automation descriptions miss the places this work goes wrong. Four of them come up repeatedly in practices running real authorization volume.

Procedures get booked outside the authorization window. At one multi-site pain practice, roughly 6 to 8% of procedures were being scheduled outside their prior-auth window by human schedulers. Nothing catches that at booking. It surfaces later as a denial, after the visit happened and the claim was filed. The fix is to check the window at the moment the slot is offered rather than at checkout, which is a scheduling-time job, not a billing-time job.

Rescheduling silently breaks the authorization link. At the same practice, when an appointment carrying an authorization is rescheduled, the new appointment is not automatically linked to the existing authorization. Their answer was to staff a human team responsible for re-linking. There is a second-order rule inside it: if a patient reschedules an appointment whose authorization has not come back yet, the appointment can only be pushed later, never pulled earlier, because there is no approved date to schedule inside of.

Which services need authorization is a specific list, and staff get it wrong. One OB/GYN practice’s rule is memorizable: a pap smear needs no authorization, contraception needs no authorization, procedures and follow-ups do. Meanwhile at a pain practice, call review caught staff telling one patient the practice did not take their plan and offering cash-pay when the plan was in fact contracted, and telling a PPO patient they needed a referral their plan did not require. Both errors cost a booking.

Referral authorizations expire on a clock. For a chronic condition a specialist referral might be good for a year, and then somebody has to notice. Nobody is assigned to noticing.

What the automation does, and what it does not

Pretty Good AI runs front-office work inside athenahealth, with production access to 820+ athenaOne APIs. For prior authorization that depth matters, because the job touches appointment types, scheduling slots, the order and tickler queue, document classes and the eligibility check, and a tool that only sees one of those cannot align the other five.

On the payer side the agent places outbound status calls on pended requests, captures the determination, the reference number and any stated reason, and writes it back into athenaOne so a coordinator opens a record that is already current. On the scheduling side it checks the authorization window before a procedure slot is offered, and flags the mismatch when an appointment and its approval do not line up. When a patient reschedules, it surfaces the broken authorization link as a work item instead of letting it pass silently. On the expiry clock it works the renewal list, calling on referral authorizations approaching their end date.

Here is the boundary, and it is absolute. The agent does not decide whether a service is medically necessary, does not argue clinical criteria with a payer, and does not write or sign an appeal. Medical-necessity judgment belongs to a clinician, and a peer-to-peer review is a conversation between two licensed physicians. What the agent removes is the hold music in front of that judgment, so the people qualified to make it spend their hours making it.

A practical example of the handoff: a status call comes back denied for a documentation gap. The agent captures the reason and the reference number, attaches it to the record in athenaOne, and routes it to the authorization coordinator with the payer’s stated requirement already written down. A human decides what clinical documentation answers it. The agent then places the resubmission status calls once that decision is made.

What to ask any vendor about prior-auth follow-up

Buyers evaluating this capability should get concrete answers to five questions, and the answers are more revealing than any demo.

Does the authorization check happen when the slot is offered, or after the visit? Only the first one prevents the denial.
What happens to the authorization link when a patient reschedules, and is that surfaced as a work item or left to a person to remember?
Where does the captured determination land in athenaOne, and can staff act on it without re-reading the call?
How does the system handle a service that does not require authorization, so patients are not delayed for approvals nobody needed?
At what point does a request stop being administrative, and what exactly does the handoff to a clinician contain?

Ask each vendor to walk one real pended authorization all the way to the booked visit. The gaps show up in that walkthrough and nowhere else.

The math an administrator can run

Prior-auth automation is usually sold on hours saved. Hours are the wrong unit for a finance conversation, because the recoverable money is in the denials nobody had time to work and the procedures booked outside their window.

Take your own book. Count the authorizations that aged past a payer deadline last quarter, and the procedures denied for a window mismatch, and multiply each by the average allowed amount for those services. That is the estimated opportunity for that quarter, not an annual figure. Then compare it to the 13 hours a week the survey says your physicians and staff already spend on this process. The second number tells you why the first one exists.

Both numbers come from the same shortage, and it is a shortage of hours, not of skill.

Key Takeaways

Prior authorization fails on follow-up, not submission, and 52% of physicians who do not always appeal cite insufficient staff time or resources as the reason (AMA).
Practices complete 40 prior authorizations per physician per week across 13 hours, and phone remains the most common method for medical-service requests.
The expensive breaks are operational: procedures booked outside the auth window, authorization links severed by a reschedule, and referral authorizations that expire with nobody assigned to notice.
An AI agent places status calls, captures determinations into athenaOne, checks the window at booking time and works the renewal clock, while medical-necessity judgment and peer-to-peer review stay with clinicians.
Measure the opportunity in denied and expired authorizations rather than in hours saved, then ask any vendor to walk one pended request all the way to the booked visit.

Prior authorization will not get simpler in the next year. What a practice can change is how much of the chasing lands on people whose time is better spent on the determinations that actually require their training. Give the status calls, the window checks and the renewal clock to automation, and keep the clinical judgment where it belongs.

Sources