It’s 9:40 on a Tuesday morning, and somewhere in a fax inbox, a referral has been sitting for twenty minutes. A physician sent it over first thing, right after seeing a patient who needs a specialist, soon. The front desk hasn’t gotten to it yet. There’s a stack ahead of it: insurance paperwork, a prior auth request, two more referrals that came in overnight. Someone will open it, read it, and start typing the patient’s name, phone number, and insurance details into another system by hand. That will take another ten minutes or so. Only then does anyone call the patient.
By the time the phone rings, the patient may have already left the parking lot. They may have picked their kids up from school, gone back to a meeting, or quietly decided to look elsewhere. Nobody did anything wrong. The process just wasn’t built to move at the speed the moment required.
A few miles away, the physician who wrote that referral has moved on to the next patient, and the next after that. She won’t think about it again until, weeks later, she happens to ask a colleague whether that patient ever got seen, and finds out they never did. Nobody called her about it. There was no dramatic failure to point to, no error message, no denied claim. The referral simply moved through a process too slow to catch the moment it mattered most, and by the time anyone might have noticed, there was nothing left to notice.
This is the story behind almost every conversation about AI in healthcare right now, except it usually gets told backwards. Leaders ask whether a new AI pilot can help. The better question is whether the organization has the operational capacity to handle what’s already arriving, referrals, faxes, forms, prior auths, fast enough that a pilot program isn’t even the bottleneck.
The burden hiding inside a normal day
Ask almost any physician’s office what eats their time, and prior authorization comes up fast. It’s the kind of task that never feels like the main event, a form here, a phone call there, and yet the American Medical Association’s 2025 Prior Authorization Physician Survey found it adds up to an average of 13 hours a week per physician and staff. Ninety-three percent of physicians said the process delays patient care.
It’s tempting to picture that time as one dramatic bottleneck, a single overwhelmed fax machine or an understaffed call center. The reality is quieter and more evenly spread across the day than that. It’s a referral read here, a field re-typed there, a record matched to the wrong patient and corrected, repeated all day, every day, across an entire staff. None of it looks like a crisis in the moment. All of it adds up to hours, and every one of those hours is an hour a patient waits.
What changes when the process actually keeps up
Zoom out to the industry level, and the picture is the same story at scale. The CAQH 2025 Index found that U.S. healthcare avoided an estimated $258 billion in administrative costs in 2024 through automated transactions, and identified a further $21 billion in savings still sitting on the table, tied to manual work that hasn’t caught up yet. That remaining number isn’t a technology gap. The tools already exist. It’s a lot of organizations still routing documents and re-typing data the way that front desk did on Tuesday morning.
This is the specific gap Documo’s Intelligent Document Processing is built to close. Instead of a person opening a fax, deciding what it is, and typing its contents into another system by hand, Documo automatically classifies inbound documents, separating referrals from prior auth requests, insurance paperwork, and everything else, then extracts the data that used to be re-keyed and routes it directly into the right workflow. The referral that used to sit in a queue while someone got to it is instead read, sorted, and acted on the moment it arrives.
Classification and extraction are only half of what closes the gap, though. The part that actually changes a patient’s experience is what happens immediately after: Documo’s workflow automation takes that extracted data and puts it to work, updating the record in the practice’s existing EHR or practice management system or flagging the referral for scheduling, all without a staff member manually carrying the information from one screen to another. The only thing that reaches a person is the exception that genuinely needs judgment, not the routine handoff that used to eat up the bulk of the time. The outcome is simple to state even if the mechanics aren’t: a patient gets contacted in minutes instead of hours, and staff get their day back for the work that actually requires them.
That’s where workflow efficiency stops being a nice-to-have and starts being the actual lever on patient access. When inbound referrals and documents are automatically classified, routed, and entered into the right system, the time between “referral arrives” and “patient is contacted” collapses. Fewer referrals stall in an inbox. Fewer patients give up waiting and go elsewhere. The economics of that shift aren’t abstract, they’re the direct, dollar-for-dollar difference between a referral that converts and one that quietly doesn’t.
The tail end of the problem, not just the average
It’s worth sitting with why “our average response time looks fine” isn’t always reassuring. An average can hide a lot. If most referrals get a call within ten minutes but a meaningful share sit for two hours because they arrived during a busy stretch or after the front desk went home, the average still looks respectable while a real slice of patients quietly disappear. Those are usually the patients on the losing end of the story, the ones for whom the delay ran long enough that the moment passed and they didn’t call back.
Fixing the average is easy to celebrate and easy to overstate. Fixing the tail, the referrals that take far longer than everyone assumes, is where the actual patients being lost usually are.
There’s a second, quieter cost to that lost referral, one that doesn’t show up in any single month’s numbers. The referring physician who sent it doesn’t necessarily know it stalled, but over time, a pattern like that shapes where she sends her next patient. Referring relationships are built on the assumption that a patient handed off will actually be seen, and every referral that goes nowhere chips away at that assumption a little, even when nobody ever says so directly. Practices rarely lose a referral source over one bad experience. They lose it slowly, the way trust usually erodes, one referral that went quiet at a time.
The staff working that fax inbox feel a version of this too. Prior authorization is widely reported by physicians as a major driver of burnout, and the staff members doing the manual entry underneath that frustration are living a version of the same grind, just without a survey asking about it. Retyping the same fields off a fax, over and over, for referrals that may or may not turn into a scheduled patient, is not the kind of work that keeps people in a role. Turnover in these positions means retraining, means inconsistency while someone new gets up to speed, and means the whole cycle of delay described above gets worse before it gets better.
Where to start, if this all sounds familiar
The instinct, once a problem like this feels real, is to want to fix everything at once: scheduling, staffing, documentation, the works. That’s rarely necessary.
The highest-leverage place to look first is the single narrowest point every referral has to pass through no matter what kind of visit it’s for: the step where a person opens a document, figures out what it is, and types its contents into another system. That step is purely mechanical. It doesn’t require clinical judgment, and it’s exactly the kind of task Documo’s Intelligent Document Processing was built to remove. Fix that one step, and faster patient contact, more consistent data, and freed-up staff time tend to follow on their own.
Go back to that Tuesday morning, the fax sitting in the queue, the physician who won’t think about that patient again until a colleague mentions it weeks later. The technology to close that gap already exists and has for a while. What’s usually missing isn’t a smarter model. It’s the operational capacity to put the tools already available, like Documo’s IDP, to use on every referral, every fax, every single day, not just in a pilot that proves it’s possible.
What actually changes
Strip away the mechanics, and the outcomes are what leadership, staff, and patients each end up feeling directly. Patients get contacted in minutes instead of waiting an hour or more, while the reason they were referred is still front of mind. Staff spend their day on outreach and judgment calls instead of retyping the same fields off a fax, which shows up as less turnover in roles that used to burn people out. Referring physicians see their patients actually get seen, which keeps the relationship, and the next referral, intact. And leadership gets a referral pipeline that converts more of what it already generates, without adding headcount to do it.
None of that requires betting on a new AI initiative. It requires the operational capacity to keep up with what’s already arriving, and act on it while it still matters.
Frequently Asked Questions
Isn’t administrative burden mostly a billing and coding problem?
Billing and coding are part of it, but a lot of administrative time happens earlier, at referral intake and prior authorization, before a claim is ever generated. That’s also the part most likely to directly delay patient care rather than just delay payment.
Does fixing this require replacing EHR or practice management systems?
No. The bottleneck sits upstream of those systems, in how documents get read, classified, and entered in the first place. The goal is getting accurate data into the systems that already exist faster, not replacing them.
Is this only relevant for large health systems?
The scale changes, but the story doesn’t. A referral sitting unprocessed for an hour matters just as much to a single-location practice’s patients as it does to a large system’s, even if the total dollar impact looks smaller on paper.
If our average referral response time already looks reasonable, is there still a problem worth fixing?
Often, yes. A reasonable-looking average can hide a long tail of referrals that took far longer to reach than anyone realizes, and those are frequently the ones most likely to have already gone elsewhere by the time contact happens.
Does this apply the same way to specialty referrals as it does to primary care referrals?
The mechanism is the same either way, a document arrives, gets read, gets entered, gets acted on, but the stakes can be higher on the specialty side, where the patient is often waiting on a diagnosis or a treatment decision rather than a routine visit.
How would an organization know whether referring physicians are quietly losing confidence in the process?
Most organizations don’t track this directly, which is part of the problem. A practical proxy is referral volume from a given source over time. A referring physician who stops sending patients without any explicit complaint is often the clearest sign that referrals stopped converting reliably.
Is staff burnout from manual data entry really connected to patient access?
Indirectly, but meaningfully. Staff turnover in referral processing roles means retraining time, inconsistent handling while someone new learns the process, and a higher chance of the exact delays described throughout this piece.
