What happens to calls after 5pm
Most practices have one of three arrangements, and most have never explicitly chosen between them. Voicemail, which is cheapest and loses the most patients. A traditional answering service, where a non-clinical operator — usually covering several practices at once — takes a message and follows a script for what counts as urgent. Or a clinician carrying the phone, which works until it is somebody’s third night in a row.
The calls themselves are not evenly distributed and not mostly urgent. The bulk is routine: appointment changes, opening hours, whether to come in tomorrow. That mix is the whole opportunity — it is a large volume of answerable questions arriving at the least-staffed hour.
On-call rotation and escalation
The escalation rule is the actual product here. Two failure modes, and they pull in opposite directions: escalate too little and a clinical call sits in a message queue overnight; escalate too readily and the on-call rota is woken for parking questions, stops trusting the system, and starts ignoring it — which is the more dangerous failure, because it is silent.
What works is a short list of explicit triggers rather than a judgement call: stated symptoms, a named condition, an explicit request for the doctor. Everything else is logged. And when a call does escalate, it should arrive as a warm transfer with the transcript attached — a clinician woken at 2am should not have to start the conversation from nothing.
Triage vs message-taking
These get sold as the same thing and are not. Message-taking captures who called and why, and defers every decision to the morning — the patient is no better off at 2am than they were before they dialled. Triage classifies the call, acts on the routine ones, and escalates the rest.
Worth being precise about a boundary: triage in this sense means routing by stated urgency, not clinical assessment. An answering service — human or AI — is not diagnosing anyone, and any vendor implying otherwise is selling you a liability. The job is to get the right call to the right person quickly and to answer the ones that were never clinical to begin with.
One practical detail that decides whether any of this works overnight: voicemail detection on the callback leg. An escalation that reaches a clinician’s voicemail and treats it as answered is the same as no escalation at all.
Cost per after-hours call
After-hours is expensive for a structural reason: the volume is low but the coverage has to be continuous, so you are paying for availability rather than for calls. A fully-loaded staffed seat runs around $300 per day1, and overnight that seat is idle most of the time by design. Staffing to a healthy occupancy band — sustained load above about 83% is associated with burnout and attrition2 — makes the arithmetic worse, not better, because it means deliberately buying slack.
That is why the market bears a $126 cost per click on this term. It is also why per-minute pricing changes the shape of the problem: an agent idle at 3am costs nothing, and the same agent answers six simultaneous callers at 8:55am without a queue. Size the staffed alternative honestly first — the Erlang C calculator gives you the seat count your call pattern actually implies, which is usually the number that settles the argument.
Sources
- 1 Fully-loaded rep-day cost: Derived from BLS (US) and published Indian call-centre salary data, retrieved 2026-07 (reference). Loaded cost includes benefits, seat, tooling and supervision — not wage alone. Adjust to your own payroll.
- 2 Occupancy ceiling, 83% of logged-in time: Call Centre Helper — industry standards for call centre metrics. Retrieved 2026-07. Sustained occupancy above 85% is associated with burnout and attrition.
