After-hours calls: what to automate and what to escalate

In one academic primary care practice that tracked its physician-staffed after-hours phone line for 15 months (January 2020 to March 2021), 63% of calls came in on weekends or holidays, and another 14% arrived between 5 and 7 p.m. on weekdays.[1] That pattern may look familiar to anyone who runs a practice: the phone keeps ringing after the lights go off, and in that study the people calling used more care than the practice's other patients.

The question for a practice owner is not whether to answer those calls. It is what happens after the answer. Some of what comes in after hours is routine administration that never needs a clinician.

Some of it is a patient who needs a person with medical training, fast. Treating the two the same way creates two risks: on-call clinicians can spend their evenings on refill requests, and an urgent call can sit in a voicemail box until morning.

This post lays out a practical split: what is reasonable to automate, what must be escalated, and how to handle the gray zone in between. It is general guidance for practice owners. Your clinicians decide what counts as urgent.

What actually comes in after hours

In that 15-month study, callers were older and used more care across visit types than the practice's other patients, and about 25% of patients phoned more than once. In total, 2,208 calls came from 1,397 patients, about 5% of the practice population. Patients who called more than once had more chronic illness than those who called once.[1] This is one practice in 2020 and 2021, so read it as an example, not a forecast.

What callers want varies. Some calls involve symptoms or worries. Others involve refills, appointment changes, billing questions or logistics such as holiday hours and directions. Your own call log will show the real mix better than any study.

Who takes the call matters too. A 2001 survey of 91 primary care offices found that more than two thirds used answering services to take their calls.[2]

The risk of asking callers to self-triage

In that same survey, 93% of the practices required the patient to decide whether the problem needed immediate physician notification.[2] Four family physicians, who did not know which calls had not been forwarded, later reviewed calls that had not reached the on-call physician. They judged 50% of those calls to be emergencies needing immediate contact with a physician, with individual reviewers ranging from 22% to 77%. The authors concluded that all clinical patient calls should be sent to appropriately trained medical personnel for triage decisions.[2]

When a phone menu, an answering service script or a recorded prompt asks a worried, untrained caller "Is this an emergency?", it hands a clinical decision to the person with the least information to make it. A worried caller may decide a serious symptom can wait, or may not want to bother the doctor. The survey is more than twenty years old, so check how your own system works today.

Why getting after-hours right matters

This is not only about safety, because it also shows up in how patients use care. In a 2010 national household survey analyzed in Health Affairs, roughly one in five people who tried to reach their primary care practice after hours said reaching a clinician was very or somewhat difficult. Patients who reported more difficulty also reported more emergency department use (37.7% versus 30.4%) and more unmet medical need (13.7% versus 6.1%) than patients who reported less difficulty. These are associations in survey data, not proof that difficulty causes the difference.[3]

The emergency department connection matters because so much emergency use happens when practices are closed. A 2012 article in the Journal of General Internal Medicine, citing national emergency department data, notes that almost 65% of all emergency department visits occur between 5 p.m. and 8 a.m. or on weekends.[4]

One health system reports what a staffed service can do. Stanford Health Care runs a dedicated after-hours triage call center staffed by registered nurses and non-licensed clinical assistants who work from triage protocols. In materials published by the AMA, Stanford reports that over 35% of callers said they would have gone to the emergency department without the service, that less than 10% of primary care calls needed escalation to a physician, and that 92% of patients had all their needs met.[5] These are one organization's own figures, so they show what is possible with trained staff and protocols, not what another practice will see.

What to automate

The reasonable layer for automation is everything administrative: calls where the right answer comes from a schedule, a policy or a form rather than a clinical assessment. A simple test: if the correct response could be read off a document your front desk keeps, it is a candidate.

In practice, that layer can include:

• Hours, locations, parking, holiday schedules and which services are offered

• Appointment requests, confirmations, reschedules and cancellations

• Prescription refill requests, captured with medication name, pharmacy and patient details and sent to the right queue for staff or clinician review

• Billing and insurance questions, captured for a callback from the billing team

• Structured messages for the morning: who called, when, callback number, reason and best time to reach them

Two things keep this safe. First, none of it involves judgment about a patient's condition. Second, every item produces a written record the team can act on. Done well, it does what a good receptionist already does at 11 p.m. on a Sunday, and it leaves a record behind.

What to escalate

Escalation covers anything where the right answer depends on clinical judgment. The caller should not be asked to make that call, and neither should an automated system acting alone. Your clinical lead should write and approve the list of situations that trigger escalation. The examples below are illustrations, not a protocol.

Some situations are possible emergencies:

• Chest pain or pressure

• Trouble breathing

• Signs of stroke, such as face drooping, slurred speech or weakness on one side

• Severe allergic reaction

• Uncontrolled bleeding

• Thoughts of self-harm or suicide

For these, the practice's approved script should tell the caller to call 911 or go to the nearest emergency room now. It should not leave the caller waiting for a callback.

A second group needs the on-call clinician rather than an ambulance: worsening symptoms, problems after a procedure, a possible medication reaction, pregnancy concerns and new severe pain. These calls should reach the on-call provider under whatever protocol the practice already uses.

A third group is about the caller rather than the symptom: someone who is confused, very distressed, hard to understand or unable to say what is wrong. If the person on the line cannot be understood, the protocol should send the call to a human, not leave it to a guess.

The gray zone rule

Between the obvious refill request and the obvious emergency sits a wide gray zone: the stomachache that might be nothing, the rash that might be a reaction, the dizziness in an older patient on several medications.

A sensible rule for the protocol is this: when a call does not clearly fit an administrative category, treat it as clinical and escalate it. Escalating a call that turns out to be minor costs a few minutes of a clinician's time. Failing to escalate one that turns out to be serious can harm a patient. Your clinicians should decide exactly where that line sits.

Call type	Examples	What happens
Administrative	Hours, directions, appointment changes, refill requests, billing questions	Handled by automation or front-desk staff, with a written record for the morning team
Needs clinical judgment	Worsening symptoms, problems after a procedure, a possible medication reaction	Escalated to the on-call clinician under the practice's protocol, not left in voicemail
Possible emergency	Warning signs your clinical lead lists, such as chest pain or trouble breathing	The practice's approved emergency message tells the caller to call 911 or go to an emergency room, with no wait for a callback

There is evidence that trained humans can handle much of the gray zone safely. In a randomized trial of 14,492 out-of-hours calls to one general practice cooperative in Wiltshire, England, specially trained nurses using decision-support software managed 49.8% of calls without referring them to a general practitioner. The trial, published in 1998, found statistical equivalence with usual care for deaths within seven days, emergency admissions and accident and emergency attendances.[6] The calls were handled by nurses, not by an automated system, so the trial does not show how automation would perform.

What a real escalation looks like

"Someone will call you back" is not an escalation. A real escalation has four parts, and each one should be written down before the system goes live:

1. A logged record. Who called, when, the callback number and what the caller said, captured at the time of the call.

2. A named human. The on-call provider or designated staff member, following the practice's own rules, actually receives the escalation.

3. A defined response window. The practice sets response expectations in its own clinician-approved protocol. Possible emergencies always get immediate guidance to call 911 or go to the emergency room, never a wait-for-a-callback message.

4. A morning handoff. Everything from the overnight period lands in one list the team reviews at the start of the day, so nothing depends on one person's memory.

Test the path on a regular schedule, for example monthly. Have someone call after hours with made-up caller details and a scripted red-flag phrase, and watch what happens. If the escalation path has a hole, that is how you find it before a patient does.

Where AI answering fits

AI phone handling can take on the administrative layer described above. 2Care AI describes itself as an AI admin workforce for outpatient clinics, covering front-desk and back-office work.[7] Its website also says urgent calls are escalated on the practice's own protocol,[7] which matches the principle in this post: the practice's clinicians define what counts as urgent, and the software follows that definition.

Whatever tool a practice uses, it should not diagnose, give medical advice or decide on its own that a clinical-sounding call can wait. When you evaluate any vendor, including us, ask what a caller hears when they describe a possible emergency, who receives an escalation, how the practice confirms it was received and what the practice can see in the log the next morning.

A practical rollout plan

If you are setting this up or tightening what you have, one approach is to work through these steps over about a month:

1. Audit your last month of after-hours calls. Sort each one into administrative, clinical or gray zone. The split in your own log shows what your volume actually looks like.

2. Write two lists: what the system handles on its own, and what always goes to a human. Ask your clinicians to sign off on the second list.

3. Define the on-call protocol. Who gets escalations, how, and within what response window, for each day of the week.

4. Test with scripted calls that use made-up caller details, including red-flag phrases, before you rely on it.

5. Review the call logs weekly for the first month. Look for calls that landed in the wrong category and tighten the rules.

Frequently asked questions

Who decides which calls get escalated?

The practice's clinicians do. A vendor, a front-desk script or the caller should not set the line. Ask your clinical lead to write the list in plain language, approve the patient-facing wording and review it on a set schedule, because contact details and clinical instructions change over time.

How can we test an after-hours setup without real patient information?

Use made-up callers and scripted situations: an appointment change, a refill request, a call that does not fit any category and a red-flag phrase. Check what the caller hears, who receives the escalation and what the morning log shows. Keep real patient details out of the test.

Does automation replace an answering service or on-call schedule?

That is a decision for the practice, and it depends on your protocol. Automation can handle the administrative layer and record what it heard. Someone with clinical training still needs to receive escalations, so confirm who that person is, and how the practice knows they got it, before any change.

What should we ask a vendor about emergencies?

Ask what a caller hears when they describe a possible emergency, and whether anything in that path depends on a callback. Ask who is notified, how the practice can see that it happened and how you can change the wording.

Where this guide stops

This guide covers how to split administrative work from clinical escalation in a primary care setting. It fits practices that already have, or are ready to write, an on-call protocol approved by their clinicians.

It is not medical, legal or compliance advice, and it does not tell you what counts as urgent. Those decisions belong to your clinicians and your own counsel. The studies cited come from specific settings and years, so they illustrate risks and possibilities and should not be read as guarantees for your clinic.

After-hours calls are two problems sharing one phone line. The administrative half benefits from speed, accuracy and automation. The clinical half needs trained human judgment under a protocol your clinicians approve, because the caller cannot reliably tell which half they are in.

If you want to bring your own after-hours protocol to a conversation about front-desk and back-office work for your clinic, book a demo with 2Care AI.

Sources

• [1] Utilization Patterns of an After-Hours Physician Triage and Advice Telephone Service in an Academic Adult Primary Care Practice, Journal of General Internal Medicine (2024)
• [2] After-hours telephone triage affects patient safety, Journal of Family Practice (2003)
• [3] After-Hours Access To Primary Care Practices Linked With Lower Emergency Department Use And Less Unmet Medical Need, Health Affairs (2012)
• [4] After-Hours Care and its Coordination with Primary Care in the U.S., Journal of General Internal Medicine (2012)
• [5] Improving Transitions of Care by Reducing After-Hours Call Volume, AMA STEPS Forward
• [6] Safety and effectiveness of nurse telephone consultation in out of hours primary care: randomised controlled trial, BMJ (1998)
• [7] 2Care AI, homepage