A referral has been sitting in a shared inbox for eleven days. Nobody decided to ignore it. It came in as a scanned fax, the patient's name was hard to read, the reason for referral was buried in a progress note, and the imaging the specialist needs was never attached. Someone meant to call the referring office. Then the schedule got busy, three more referrals arrived, and this one slid out of view.
This is the everyday shape of intake and referral work. A referral arrives through fax, a portal, a phone call, email, an EHR message, or an uploaded document. The packet is partly complete. Insurance needs checking. The clinical reason sits somewhere in a note. Something is missing, someone has to chase it, and the patient still has not been called. Every one of those steps is a place the referral can stall, and most of them stall quietly.
The instinct is to blame manual data entry, so a clinic buys a form tool or a scanning add-on and hopes the typing goes away. That helps a little. It does not fix the real issue, which is that intake gets treated like document handling when it is actually a question of status, routing, and communication: what came in, what it still needs, who owns the next move, and who has been told. When those four things are invisible, no amount of faster typing keeps referrals from aging in an inbox.
The job is to make where each referral stands visible
Before you touch tools or automation, get clear on what a working intake process actually has to answer. For any referral in the building, the team should be able to say, without opening three systems:
- What came in, from where, and for which patient?
- Is the packet complete enough for the specialist to act on?
- What is missing, who owns getting it, and who has already been asked?
- Which patient should be scheduled, triaged, or held right now?
- What has the patient been told, and what has the referring office been told?
Notice that only one of those questions is about typing data. The rest are about status and ownership. A referral is rarely stuck because someone has not finished entering it. It is stuck because the next action is unclear, or because it is waiting on a person nobody is chasing.
A quick test tells you whether your current setup passes. Open one referral that has been sitting too long and ask whether your team can see, in one place, the missing item, who owns the chase, whether the patient has been contacted, whether the referring office has been asked, the insurance status, and the next scheduling step. If that takes a walk to three screens and two people, the work is hiding, and hidden work is what ages in the inbox.
Follow one referral from arrival to the first appointment
Pick one common referral type and walk it end to end. Do this before you decide anything about software, because the walk shows you where the handoffs are fragile. In most clinics the path looks something like this:
- A referral arrives through fax, a portal, a phone call, email, an EHR message, or a web form.
- Someone identifies the patient, the referring provider, the reason for referral, and the requested service.
- Documents get uploaded, renamed, or attached to the right chart.
- Insurance, demographics, and contact details get checked.
- Someone reviews whether the packet is complete enough for the specialist.
- Missing information gets requested from the patient or the referring office.
- The patient gets routed to scheduling, triage, prior authorization, clinical review, or a wait queue.
- The patient and the referring office get contacted by phone, text, portal message, fax, or email.
- The referral gets scheduled, closed, redirected, or left waiting.
Nine steps, and a handoff between almost every one of them. The referral changes hands from the person who received it to the person who checks insurance to the person who books the appointment, and at each pass the context can drop. The single most useful thing you can do first is not automate any of these steps. It is to make each referral's status and owner visible so that a dropped handoff shows up instead of disappearing.
Where intake and referral work quietly breaks
The failure points are predictable. They look slightly different in a specialty group than in a hospital outpatient department, but the underlying patterns repeat.
The referral arrives in six different places
Fax, portal, phone, email, EHR message, uploaded document. Each channel has its own habits, and none of them naturally lands in a single view. A referral that comes by fax gets handled differently from one that comes through the state health information exchange, and a phone referral can live in someone's memory for an hour before it is written down. When the front door is six doors, the first thing lost is a reliable count of what is actually waiting.
Nobody owns the missing-information chase
A packet is missing recent imaging. Someone notices, sends a fax to the referring office, and moves on. There is no due date, no follow-up, and no record that the request was even made. Two weeks later a different staff member opens the same referral, sees the same gap, and sends the same fax. The chase happens twice and lands nowhere because it was never owned.
Insurance and authorization get discovered late
Eligibility often gets checked at the end, right before scheduling, which is the worst possible time. A coverage problem or a missing prior authorization surfaces after the patient has been contacted and a slot has been held, so the appointment gets unwound and rebooked. Discovered at intake, the same problem is a routine step. Discovered at scheduling, it is a cancellation.
Clinical review and administrative completeness get tangled
Two different questions get treated as one. Is the packet administratively complete, meaning it has the documents, the insurance details, and the contact information? And is it clinically appropriate and urgent, which only a clinician can judge? When those blur together, an administrative gap waits on a clinician who is busy, and a genuine clinical question gets buried under paperwork chasing. Separating them lets each move at its own speed.
The referring office never hears back
A referral is a relationship, not just a document. The referring provider sent a patient and, in most cases, hears nothing until the visit happens or does not. When a packet is incomplete, the referring office is often the fastest way to close the gap, yet it is the party least likely to be kept in the loop. Silence costs referrals: an office that never gets an acknowledgment starts sending patients somewhere more responsive.
Scheduling inherits work intake should have finished
When intake is loose, the scheduler becomes the last line of defense. They call a patient to book and discover the insurance is wrong, the imaging is missing, or the referral was never really complete. Now the person whose job is to fill the calendar is doing intake investigation on the phone while the patient waits. That is how a scheduling team ends up feeling permanently behind when the real problem lives upstream.
Build one intake queue you can actually work from
The central thing to build is not a new system to replace your EHR. It is a single queue that makes the work visible enough to route and finish. The EHR remains the record; the queue is the working view that sits on top of it and answers where every referral stands.
A useful queue holds, for each referral, the status, what is missing, who owns the next move, how old it is, and the next action with a due date. That is enough to turn a pile of aging faxes into a list the team can work down. It also does something for whoever runs the clinic: it shows where the bottleneck actually is. If forty referrals are stuck waiting on documents and three are stuck in clinical review, the problem is the document chase, not the clinicians, and you now know where to spend effort.
| Referral | Status | What is missing | Owner | Next action |
|---|---|---|---|---|
| Referral A | Needs documents | Recent imaging report | Intake coordinator | Request from referring office, set a two-day follow-up |
| Referral B | Needs clinical review | Nothing, packet complete | Triage clinician | Route to clinician triage today |
| Referral C | Waiting on patient | Scheduling preference, consent confirmation | Scheduler | Send approved message, hold a slot until Friday |
| Referral D | Needs eligibility check | Active coverage and prior authorization | Insurance verifier | Run eligibility, open authorization if the plan requires it |
The rows are illustrative, but the shape is the point. Every referral has one status, one owner, and one clear next action. Nothing is sitting in a state of general unfinishedness that everyone can see and no one can act on.
The states a referral moves through
The status column only works if the states are few and unambiguous. Too many states and staff argue about which one applies; too few and the queue hides the very distinctions that decide the next move. A workable set for most clinics looks like this:
- New: received, not yet looked at.
- Needs identification: the patient or referring provider does not match cleanly to a record.
- Needs documents: a required clinical or administrative item is missing.
- Needs eligibility check: coverage or prior authorization is blocking the next step.
- Needs clinical review: a triage or appropriateness call is required before scheduling.
- Ready to schedule: enough is in place to contact the patient and book.
- Waiting on patient: the patient has been contacted and a response is pending.
- Waiting on provider: the referring office or an outside party needs to respond.
- Closed or redirected: the outcome is recorded, whether booked, sent elsewhere, or declined.
The value of naming these states is that they separate reasons for delay. A referral waiting on the patient is not the same problem as one waiting on documents, and neither is the same as one stuck in clinical review. When the queue can tell them apart, the daily question stops being why is intake slow and becomes which of these specific piles do we clear first.
Give each referral type its own packet checklist
A single generic checklist fails in two directions at once. Make it thorough enough for the most complex referral and it buries the simple ones in unnecessary requirements. Make it light enough for the simple ones and it lets risky referrals through underprepared. The fix is to define what a workable packet means per referral type, and to tie each requirement to where a gap should send the referral.
The point of tying the checklist to routing is that a missing item becomes a decision, not just a flag. Missing imaging sends a referral to the document chase. A high-urgency clinical note sends it to triage now. A complete packet goes straight toward scheduling. The checklist stops being a form someone fills in and becomes the thing that decides what happens next.
| Referral type | What the packet needs to be workable | Where a gap sends it |
|---|---|---|
| Routine specialist consult | Patient identity and contact, reason for referral, relevant notes, active insurance | Missing insurance goes to eligibility; missing notes go to the referring office |
| Procedure requiring prior authorization | Clinical justification, prior imaging or labs, insurance and authorization details | No authorization on file goes to the insurance verifier before scheduling |
| Urgent or time-sensitive referral | Reason for urgency, key clinical context, a reachable contact for the patient | Anything flagged urgent routes to a clinician for a same-day triage look |
| Imaging-dependent specialty | The actual images or a report, not just an order, plus the referring contact | Order without images goes back to the referring office with a due date |
Keep the checklists short. The goal is not to collect everything a chart could hold. It is to collect the few items that decide whether the specialist can act, so that everything else can wait without holding the referral hostage.
Put eligibility and authorization inside the queue, not beside it
Insurance is where a surprising share of intake friction hides, and it hides because it usually lives in a separate step run by a separate person at a separate time. A referral looks complete on the clinical side, moves toward scheduling, and only then does someone check coverage and find the plan is inactive, the patient is out of network, or the procedure needs an authorization nobody started.
Moving eligibility into the queue means treating coverage as one of the states a referral can sit in rather than a gate it hits at the end. When a referral is logged, checking active coverage and flagging whether the requested service typically needs prior authorization becomes part of the first look, not the last one. That does two things. It surfaces coverage problems while there is still time to solve them, and it lets a referral that will need an authorization start that clock early instead of losing a week discovering the requirement after a slot was already held.
None of this asks intake staff to make coverage decisions they are not equipped to make. The queue records the status: coverage confirmed, coverage in question, authorization needed, authorization submitted, authorization approved. The insurance verifier still does the verifying. The queue just makes sure the referral does not travel toward a booked appointment while a coverage problem sits unseen behind it.
Keep the referring provider informed without extra phone tag
Intake has a second audience that is easy to forget under the pressure of the first. The patient needs to be scheduled, yes. But the referring provider sent this patient and, in most clinics, hears nothing back until either the visit happens or the patient calls them confused about why it has not. That silence is a quiet leak. Referring offices route patients to the specialists who make them look responsive to their own patients, and a specialist who acknowledges nothing loses that flow over time.
The intake queue is where this gets fixed, because the queue already knows the two moments worth a message. The first is acknowledgment: the referral arrived and here is its status. The second is a specific ask: the packet is missing recent imaging, please send it, and here is the two-day window we are holding. Both messages can be drafted from what the queue already holds, so keeping the referring office in the loop stops being an extra task someone remembers to do and becomes a byproduct of working the queue.
This is the part of intake that most directly earns the clinic more referrals, and it is usually the part that gets dropped first when the day gets busy. Building it into the status flow, rather than leaving it to good intentions, is what keeps the referring relationship warm without adding another round of phone tag to anyone's afternoon.
A worked example: an orthopedics group taking 200 referrals a month
Here is an invented example to make the shape concrete. The clinic and the numbers are made up, chosen only to show how the pieces fit; treat them as illustrative, not as a benchmark.
Say a specialty orthopedics group receives about 200 referrals a month across three surgeons. Roughly a third arrive by fax, a third through the hospital's referral portal, and the rest by phone, email, or the state exchange. Before any of this is organized, the front-desk team handles each referral as it appears, keys the details into the EHR, and calls patients to book. It feels busy but survivable. The problem shows up in the gaps: a handful of referrals age past three weeks because nobody was chasing the missing MRI, a few patients get booked and then bumped when the authorization falls through, and one referring office quietly stops sending patients because it never heard back on two of them.
Now put the same volume through one intake queue. Each of the 200 referrals lands in the queue with a status the moment it arrives. On any given Monday the queue might show 30 in New, 45 needing documents, 20 needing an eligibility check, 15 in clinical review, 60 ready to schedule, 20 waiting on a patient, and 10 waiting on a referring office. The team no longer works from a vague sense of being behind. They work from a list. The intake coordinator clears the New pile and the document chases, each with a two-day follow-up so nothing is asked once and forgotten. The insurance verifier works the eligibility pile before those referrals reach scheduling, so authorizations start early and slots stop getting unwound. The triage clinician sees only the fifteen that genuinely need a clinical look, not all 200. The scheduler works the ready pile and stops discovering surprises on the phone.
The point is not that a queue conjures more hours. It is that the same 200 referrals become visible and sorted, the missing-information chase gets owned instead of repeated, coverage problems surface early, and the referring offices hear back. The referrals that used to age past three weeks now show up as the oldest rows in a specific pile, which means someone can actually clear them.
Where AI helps inside intake and referral work
Once the queue exists and the states are defined, there is real, safe room for AI to take the grunt work out of intake, which is the whole promise of doing this without adding admin. The key is that AI works on the administrative layer only: reading, extracting, sorting, and drafting, always with a person reviewing before anything moves. Useful applications include:
- Reading a referral document and pulling out the patient, provider, insurance, and requested service into the queue, so staff correct fields instead of typing them from a scanned fax.
- Suggesting the referral type and the matching packet checklist, so the right requirements attach automatically.
- Flagging what is missing against that checklist before a person starts the manual chase.
- Drafting the acknowledgment or the missing-information message to the referring office for a staff member to check and send.
- Summarizing a complete packet for the triage or scheduling handoff, so the next person reads a page instead of a folder.
- Finding prior communication or documents already tied to the same patient or referral.
Every one of these speeds up a step a person already does by hand. None of them decides anything clinical, and none of them acts without review. That boundary is what lets AI reduce the typing and chasing without becoming a source of errors nobody caught. Interoperability and administrative burden are real, documented pressures in healthcare, not abstract ones; HealthIT.gov's interoperability materials and the CMS burden reduction initiative both point at the front door as a place where better process design matters more than another inbox.
Where clinical judgment stays with your clinicians
It is worth being explicit about the line, because the fastest way to make an intake project fail is to let it drift into clinical territory it has no business in. The queue and the AI around it handle administrative work: what arrived, whether the paperwork is complete, whether coverage is in place, who to contact, and when. They do not decide whether a referral is clinically appropriate, how urgent it is, whether a patient needs to be seen sooner, or what the specialist should do. Those are clinical judgments, and they stay with clinicians.
In practice that means the queue routes a referral to clinical review when the packet is complete or when something is flagged urgent, and then a clinician makes the call. AI can summarize the packet to make that review faster, but the summary is an input to a clinician's judgment, not a substitute for it. Patient information stays inside your approved systems throughout, and none of this is a substitute for your own compliance and legal guidance on how patient data is handled. The workflow exists to get the right, complete referral in front of the right clinician faster, not to make clinical decisions on anyone's behalf.
What to measure once the queue exists
The queue gives you numbers you could never see before, and a few of them tell you almost everything about where intake is stuck. You do not need a dashboard project to start. You need to watch a handful of measures and let them point you at the next fix.
| What to measure | What it tells you | Where it usually points |
|---|---|---|
| Time from referral received to status assigned | How fast the front door works | A backlog here means arrival channels are not being triaged |
| Share of referrals missing a key item on arrival | How incomplete referrals really are | A high number points at referring-office templates and packet rules |
| Average age of referrals by status | Which pile is aging | The oldest pile is your real bottleneck, not the loudest one |
| Time from complete packet to booked appointment | Whether scheduling keeps up once a referral is ready | A gap here is a scheduling-capacity or contact-reachability issue |
| Referrals waiting on patient, provider, or eligibility | Who the delay is actually parked on | Waiting-on-provider volume points at referring-office follow-up |
| Referrals closed, redirected, or lost to follow-up | How many patients never make it to a visit | Lost referrals are the number worth reducing first |
These measures also make the cost of manual intake visible in a way that helps you decide what to redesign. If you want to put rough numbers on the manual steps before committing to changes, Ubisar's guide to estimating the cost of manual work is a practical place to start.
The first month should make the queue visible
Do not try to automate every referral channel in the first month. Start with one high-volume or high-friction referral type and get the queue working for that before you widen it. A sensible first-month path looks like this:
- Choose one referral stream, location, specialty, or intake team.
- Map how those referrals arrive, who touches them, and where they stall today.
- Define the states and the packet checklist for that referral type.
- Build the queue with owner, age, missing item, and next action.
- Connect the few source documents or exports the team actually needs to work it.
- Write the two message templates that matter: acknowledgment to the referring office, and a missing-information request.
- Add AI only where staff review the output, starting with reading documents into fields and summarizing packets.
By the end of the month, the team should be able to say how many referrals are new, incomplete, waiting on a provider, waiting on a patient, ready to schedule, or held on eligibility. That visibility alone tends to cut a lot of the daily chasing, because the referrals that used to disappear now show up as the oldest rows in a named pile. Only after the queue is working for one stream is it worth widening it to the next, and only then is it worth adding heavier automation.
Common traps
A few mistakes turn up again and again, and knowing them in advance is cheaper than learning them live.
The first is treating referrals as documents instead of work with a status and an owner. A scanning tool that files faxes faster still leaves them aging, because filing is not the same as moving. The second is letting administrative completeness and clinical review blur together, so a missing fax waits on a busy clinician and a real clinical question waits on a paperwork chase. The third is making the scheduler chase context that intake should have captured, which turns the person filling the calendar into an intake investigator. The fourth is letting AI extraction run without a person checking it, which trades slow-but-correct for fast-but-wrong. The fifth is losing the communication trail, so the patient and the referring office cannot tell whether the referral is moving. Each of these is avoidable once the queue makes the next action obvious and puts a name next to it.
How Ubisar would implement this workflow
In week one, Ubisar would choose one referral stream and map how it arrives, which packet items matter, who checks eligibility and authorization, when a clinician needs to see it, and how the scheduling handoff happens today. The first thing we would build is the intake queue: status, missing-item reason, the owner of the patient or provider follow-up, the age of the referral, urgency, and the next action.
In weeks two and three, we would connect the minimum source documents, EHR exports, scheduling data, and coverage status needed to work the queue, plus the two message templates that keep the referring office informed. AI would sit behind reviewable steps only, reading documents into fields and summarizing packets, never making a clinical call. By week four, staff should be able to work the queue, see which referrals are ready to schedule, and know which party needs the next nudge.
At the end of the first month, we keep going if the queue is making intake clearer and cutting the missing-information chase, and we narrow or stop if the referral stream first needs a staffing or policy decision that no workflow can make for you. If patient intake and referrals are where your clinic loses time, that is a good first workflow to fix, and you can tell us about it or read more about the monthly AI, Data & Tech Implementation Service it fits inside. This is a healthcare workflow; you can also browse the wider workflow guide library or start with how to choose the first workflow to improve with AI.