The After-Hours Triage Script: Telling an Answering Service What Counts as Urgent

Short answer: Your answering service routes by the categories you give it, so the script is a routing document, not a clinical one. Open with the emergency instruction, define three or four urgency tiers in observable language the operator can hear rather than diagnose, name exactly who each tier reaches, and state what happens when nobody answers.

Every answering service will ask you to approve a script. Most practices skim it, change the greeting to include the practice name, and send it back the same afternoon. That document then decides, for every night and weekend that follows, which calls wake a physician and which ones wait until Monday.

It is worth more than an afternoon. The script is the only place where your clinical judgment about urgency gets encoded before the fact, so that a person with no medical training can act on it correctly at three in the morning without having to improvise.

This page is about what goes in the script. The vendor diligence that has to happen before any script matters — the signed BAA, the subcontractor chain, where messages are delivered — is in what to require before you sign an answering service, and the cost side of the after-hours decision is in our after-hours answering comparison.

The operator routes; the operator does not assess

This is the line that every other decision in the script hangs from, and it is the one most often blurred. An answering service operator is not performing triage in the clinical sense. They are matching what a caller said against a list you wrote and sending the call somewhere.

The practical consequence is that every category has to be recognizable from the caller's own words. A tier defined as "respiratory distress" asks the operator to make a judgment they are not qualified or insured to make. The same tier defined as "caller says they are having trouble breathing" asks them to listen and match. The second one works at 3 AM with a tired operator on their eleventh call.

Write the categories the way a transcript would read. If you cannot express a tier without a clinical term the caller would not use, the tier is doing assessment work that belongs to the on-call provider, and it needs rewriting or escalating by default.

Start with the sentence before the script

Before any categories, before the greeting, the first thing in the document should be the emergency instruction — and it should be unconditional.

The standard form is simple: if this is a medical emergency, hang up and dial 911. It belongs in the recorded greeting the caller hears, and it belongs as the operator's first branch. No tier, no message, no callback promise. A script that routes a possible emergency into a message queue has introduced a delay that no compliance posture can justify.

Two details that get missed. First, the instruction should be at the front of the greeting, not appended after a menu — a caller in distress will not listen to forty seconds of hours-and-location information first. Second, the operator needs explicit permission to give that instruction again mid-call, at any point, regardless of which tier the call started in. Callers frequently understate at the open and escalate as they talk.

Three tiers is usually right; five never is

Practices that write elaborate scripts tend to write them once and then discover that the distinctions do not survive contact with real callers. The categories that work are few, mutually exclusive, and obviously different from one another.

• ▸Emergency. Not a tier so much as an exit. The caller is told to hang up and dial 911, and the practice is notified after the fact rather than consulted during.
• ▸Urgent — reach the on-call provider now. The set of situations where a clinician needs to hear about it tonight. This is the tier that costs money to get wrong in both directions, so it deserves the most specific language.
• ▸Same-day or next-business-day. Real clinical questions that do not need a 2 AM answer. Captured as a structured message, worked first thing.
• ▸Administrative. Scheduling, hours, billing, directions, refill requests that are not time-critical. Most of the volume, none of the escalation.

The administrative tier is worth designing rather than tolerating, because it is where the savings are. Reschedules and hours questions do not need a human operator at all — they are the calls an auto-attendant or an automated response resolves outright, which is the argument in setting up an auto attendant and, for the messages that arrive as missed calls rather than conversations, missed-call text-back.

Write the urgent tier as a list, not a definition

The urgent tier is where scripts fail. A definition — "situations requiring prompt clinical attention" — hands the decision straight back to the operator. A list does not.

Build the list from your own practice's history rather than a template. The specialty determines almost everything here: what constitutes an overnight call for an obstetrics practice, a dermatology practice, and a behavioral health practice have almost nothing in common, and a generic list will be wrong for all three.

A useful exercise is to pull the last six months of after-hours calls that did reach a provider, and sort them into the ones where the provider was glad to be called and the ones where they were not. The first pile, described in the callers' own words, is your urgent tier. The second pile tells you which phrasings are currently over-triaging.

One structural rule worth adopting whatever the specialty: when a caller's description does not clearly match a tier, the script should escalate rather than defer. Ambiguity resolving downward is how a practice ends up explaining why a call waited until Monday. Ambiguity resolving upward costs a phone call.

Name people, not roles the operator has to resolve

A tier that says "route to the on-call physician" is incomplete unless the operator has an unambiguous way to know who that is tonight. This sounds obvious and is the single most common operational failure in after-hours coverage.

• ▸Who maintains the rotation, and where does the service read it from? A calendar the practice updates and the vendor subscribes to beats a monthly PDF emailed to a shared inbox, which beats nothing.
• ▸What is the deadline for changes? Swaps happen. There should be a stated cutoff and a named person who can make a change after it.
• ▸What happens on holidays and coverage gaps? The weeks around major holidays are when rotations are least reliable and call volume is least predictable.
• ▸Who is the fallback when the primary does not answer? This deserves its own treatment, and it is the subject of the next section.

The mechanics of getting a call from the service to the right phone — and what the provider sees before they authenticate — are a phone-system question rather than a script question. Our cloud voice handling keeps the content inside the platform and places the return call through the business number, so a provider returning a call at midnight is not exposing their personal number to do it.

Say what happens when nobody answers

Most scripts describe the happy path in detail and go silent on failure. The failure path is the one that generates incidents.

The script should state, in plain terms, how many attempts the operator makes, how long they wait between attempts, who they try next, and what they tell the caller in the meantime. A chain that has no defined end is not a chain — we go through how to build one that terminates in when the on-call provider does not answer.

The caller-facing half matters too. "The doctor will call you back" is a promise the practice has to keep. "I have sent this to the on-call provider and marked it urgent; if you have not heard back in thirty minutes, call us again" is accurate, sets a checkable expectation, and gives the caller a defined action if the system fails.

What the script should not collect

A triage script is a natural place to over-collect, because more information feels like better care. It is not, and the minimum-necessary standard applies to the script as much as to anything else.

The routing decision needs a name, a callback number, whether the caller is the patient, the tier, and a short reason in the caller's words. It does not need a date of birth, an insurance identifier, a medication list, or a symptom narrative — those create a store of sensitive detail in a vendor system that serves no routing purpose. The full treatment of what an operator may write down, and where it may go afterward, is in the answering-service requirements guide; the rules for who the returning clinician may then discuss it with are in who am I allowed to talk to?

There is one important exception in the other direction. If your script asks the operator to determine whether a caller is a current patient before routing, be aware that this is often unanswerable at 2 AM and frequently wrong. Routing an urgent call from a non-patient to the on-call provider costs a phone call. Turning away a patient the operator could not find in a list costs considerably more.

Test it before it goes live, then again in ninety days

Scripts are approved on paper and fail on the phone. The only meaningful validation is to call the service yourself, out of hours, as a caller rather than as the client.

1. 1.Place an emergency-sounding call and confirm the operator gives the 911 instruction immediately rather than taking a message.
2. 2.Place an urgent call and time how long it takes to reach the on-call phone. Look at what appears on that phone's lock screen.
3. 3.Place an ambiguous call — a real one, phrased the way a worried caller would phrase it — and see which way it resolves.
4. 4.Place an administrative call and confirm it does not escalate.
5. 5.Place an urgent call and do not answer the on-call phone, so you can watch the failure path actually run.

Repeat the exercise a quarter later. Answering-service staff turn over, scripts get amended by whoever picked up the account, and a script that was accurate in March is a hypothesis by September.

If you are evaluating an AI answering system rather than a human service, the script becomes configuration rather than training, which changes the failure modes but not the requirements — what those systems can resolve on their own versus route is covered by our sister company at AI voice agents for business. And the practice-wide compliance program the after-hours setup plugs into is IT territory: the HIPAA compliance guide for healthcare IT.

Frequently asked questions

What should an after-hours answering service script include?
An unconditional emergency instruction to hang up and dial 911, three or four urgency tiers written in observable language a caller would actually use, the named destination for each tier, the minimum fields the operator collects, and a defined failure path stating how many attempts are made and who is tried next when nobody answers.

Can an answering service operator triage a patient call?
Not in the clinical sense. Operators match what a caller says against categories the practice wrote and route accordingly. Any category that requires interpreting symptoms rather than hearing them is asking for a clinical judgment the operator is not qualified to make, and it should be rewritten in the caller's own vocabulary or escalated by default.

How many urgency levels should an after-hours script have?
Three or four, including the emergency exit. Practices that write five or more find the distinctions do not survive real calls, and operators default to the tier they are most confident about. Emergency, urgent, next-business-day, and administrative covers almost every practice.

What information should the operator collect on an after-hours call?
The caller's name and callback number, whether they are the patient, the urgency tier, and a brief reason in the caller's own words. Scripts should not solicit dates of birth, insurance identifiers, medication lists, or symptom histories, because none of that is needed to route the call and all of it creates sensitive data in a vendor's system.

How often should the script be reviewed?
Quarterly, with live test calls rather than a document read-through. Answering-service staff turn over, accounts change hands, and amendments accumulate — so a script that was accurate when it was approved is an assumption a few months later until someone dials the number and checks.

The script and the phone system are one build, not two. Our pricing is public, and the plans that carry a BAA are marked on the healthcare page.