On-Call Routing Only Works If the Schedule Is Readable

On-call routing sounds like a telephony feature and it is really a data problem. A patient calls a surgical practice at nine at night with a post-op question, and everything about whether that call goes well was decided weeks earlier, by whether anyone kept the roster honest.

Most practices discover this the same way. The routing works for a month, then a surgeon changes a call week by text message, or a new partner starts and nobody updates the list, or somebody covers a Thursday as a favor. The system keeps routing on what it was told. The call goes to a surgeon who is not covering, or to a voicemail that belongs to a person who left.

The patient does not experience that as a configuration issue. They experience it as a practice that did not pick up.

The roster is wrong before the router is

Start with the least glamorous step, because skipping it is what breaks every subsequent one. The provider list on the practice website and the provider list in the system disagree, at nearly every practice, and reconciling who actually works there and takes calls is step one before any routing rule can be written.

In a surgical group the divergence is usually worse than average. Partners have privileges at facilities the practice does not manage. Locums cover blocks. A retired partner still sees a half-day of follow-ups and is therefore in the system but should never receive a night call.

So the first artifact is not a call tree. It is a reconciled list: who is an active provider in the practice, which department each belongs to, and which of them are eligible to receive an after-hours escalation at all. That last column does not exist anywhere by default and it is the one that matters most.

Do that once properly, keep it current as part of onboarding and offboarding, and the routing becomes almost boring. Skip it, and every routing improvement gets built on a list that is quietly wrong.

Coverage is a schedule, not an attribute

The second failure is treating on-call as a property of a person rather than a property of a date. Systems that store an on-call flag on the provider record are always describing last week.

What the router needs is the same thing the practice manager keeps in a spreadsheet: a date, a department, a covering provider, and a fallback. Weeknights and weekends are different. Holidays are different again. Some practices split call between the practice and a facility after a cutoff hour, which means the correct answer changes at eight rather than at close of business.

The workable arrangement is that the schedule lives in one authoritative place and the router reads it rather than holding its own copy. The moment there are two copies, the swap that a surgeon arranged on Sunday exists in one of them.

And when the schedule has no answer, because the date was never filled in or the covering provider is inactive, the call does not get dropped into silence. It goes to the practice’s designated fallback with the gap stated plainly, and somebody fixes the schedule Monday. Failing loudly to a human is the correct behavior. Failing quietly to voicemail is how a practice discovers the problem from a complaint.

Surgical call volume is a chain, not a stream

General surgery calls are unusual in a way that changes routing design. Consult, pre-op, surgery, and post-op form a linked chain, and where a patient sits in that chain determines who should hear from them.

A pre-op patient calling about instructions belongs with the scheduling and pre-op staff. A patient four days out from an operation belongs with the surgeon on call for that case, or with the covering surgeon if the operating surgeon is away. A patient asking about a bill belongs nowhere near either.

So the router needs the patient’s position in the episode, which comes from the record rather than from the caller. Reading recent encounters and the department that owns them answers it. What it does not do, and must not do, is evaluate the caller’s condition. The call flow captures what the patient says in their own words and reads any emergency instruction first. It does not ask a series of questions designed to sort callers by how serious their situation sounds, and it does not advise anyone. That judgment belongs to a licensed clinician.

The handoff is explicit. Anyone who asks for the surgeon gets the surgeon, without being screened. Anyone using emergency language gets the emergency instruction immediately. Everything else is captured, cased in the record, and queued for the morning.

The calls that never should have reached a person

Once routing is honest, the next question is how many of these calls needed routing at all, and the answer in most surgical practices is a lot fewer than currently get it.

An MGMA Stat poll on where practices are pointing AI found prior authorization at 16% of responses, behind scheduling (31%), calls (27%), and registration and eligibility (23%). Calls sit near the top for a reason that is obvious to anyone who has staffed a surgical front desk: a large share of them are the same five questions.

Where is the office. What time is my surgery. Did my authorization come back. Can I move my follow-up. Am I still supposed to stop eating at midnight, which is a question about the written instruction the practice already gave and not a clinical question.

Handled automatically, those never reach the on-call surgeon and never reach a voicemail box that somebody clears in the morning. What is left going to a person at night is genuinely small, which is the only version of an on-call rotation that surgeons stop complaining about.

What to measure so it stays honest

Routing decays. The roster drifts, someone leaves, a coverage rule changes and nobody tells the person who owns the configuration. Assume decay and instrument for it.

Three numbers are enough. Calls that reached the intended provider on the first attempt. Calls that fell through to the fallback because the schedule had no answer. Calls escalated at night that a morning review says did not need to be. The second and third are the ones that tell you the configuration has drifted.

Access pressure makes this worth the discipline. When MGMA asked practice leaders where they would focus on patient access in 2026, no-shows were the largest single answer at 27%, with online scheduling at 24% and phone access at 22% close behind. A surgical practice that misroutes night calls is spending its phone access budget on frustration.

Review those three numbers monthly against the on-call schedule for the month that just ended. It takes fifteen minutes and it catches drift before a patient does.

Key Takeaways

Reconcile the provider roster against who actually works and takes call before writing a single routing rule.
Add an explicit eligible-for-after-hours-escalation column, because no system carries that distinction by default.
Store coverage as a dated schedule with a fallback rather than as a flag on the provider record.
Keep one authoritative copy of the on-call schedule, since a second copy is where last Sunday’s swap goes missing.
Fail loudly to a named fallback when the schedule has no answer instead of quietly to voicemail.
Route on the patient’s position in the surgical episode, read from the record, never on an assessment of the caller.
Track first-attempt reach, fallback rate, and unnecessary night escalations monthly to catch configuration drift early.

A surgical practice does not need a cleverer phone tree. It needs a roster that matches reality, a coverage schedule stored as dates rather than as a habit, and a call flow that finishes the routine questions so the small number that genuinely need a surgeon get through cleanly. Do that and the on-call phone stops being the thing partners dread. The calls that arrive are the ones that should have.

Sources