AI Receptionist for Allergy and Immunology: Shots on Schedule

Most of the calls hitting an allergy practice are not from new patients. They are from the same people, again and again, moving an injection appointment they already have. Allergen immunotherapy runs on a schedule: 1 to 3 injections a week during the build-up phase, generally 3 to 6 months to reach a maintenance dose, then one injection every 2 to 4 weeks, with the decision to continue usually made after 3 to 5 years of treatment.

That shape breaks the usual advice about front-desk automation. A clinic that books discrete visits can automate the booking and stop thinking about it. An allergy practice is booking a dosing ladder, so the same call can be a two-second reschedule or a clinical decision about what goes in the syringe, and the caller cannot always tell you which one it is.

This post covers what the recurring shot schedule does to your phone line, why the 30-minute observation period is a capacity constraint rather than a courtesy, the two places an AI receptionist has to stop cold, and how to set one up against your own injection schedule instead of a demo script.

Why an allergy phone line runs on repeat visits, not new ones

An allergy and immunology practice books a dosing schedule, not a series of separate visits. One immunotherapy patient returns 1 to 3 times a week for months, then every few weeks for years. Most of your calls move an appointment that already exists rather than creating a new one.

Every figure comes from the AAAAI and ACAAI practice parameter. One patient, dozens of booked visits.

Do the arithmetic once and the volume stops being abstract. A patient on a weekly build-up who reaches maintenance in 6 months books roughly 26 injection visits before the first maintenance appointment arrives. Run 50 patients through immunotherapy at the same time and you are booking, moving, and reminding well over 1,000 injection visits a year. That figure is modeled from the ranges in the practice parameter, so run it against your own panel, but the order of magnitude holds for any practice with a busy shot room.

Almost none of those visits needs a clinical conversation to schedule. The patient knows why they are coming. The slot type never changes. The same person calls every week for months, often about the same two things: moving Thursday to Friday, or asking whether they can come at lunch.

That combination, high frequency and low complexity, is the strongest automation candidate on a medical phone line. It is also the reason a generic AI receptionist demo looks great here and then falls over, because the exceptions in a shot room are not administrative.

The 30-minute wait after a shot is a scheduling constraint

The practice parameter tells patients to stay in the office for at least 30 minutes after an immunotherapy injection so that a reaction can be recognized and treated promptly. That turns a 2-minute injection into a half-hour of chair time, and it is the number your booking rules have to respect.

Higher-risk patients, including anyone with a previous systemic reaction, may be asked to wait longer at the physician's discretion. Cluster schedules push it further: 2 or more injections are given per visit, typically about 30 minutes apart, which can reach a maintenance dose in as little as 4 weeks but stretches one visit past an hour.

This is also the most common thing your front desk explains all day. A new immunotherapy patient asks how long it takes, hears "about five minutes", and is annoyed at minute 20. An AI receptionist that quotes the full block on every booking removes an entire category of complaint before it starts.

A missed shot is a dosing question, not a rebooking question

When a patient misses an injection week, the visit is easy to move and the dose is not. The parameter is blunt: no published evidence supports a specific dose change for treatment gaps, but it is customary to reduce the dose when the interval runs long. Your clinical staff owns that call.

How long is too long depends on the extract concentration, whether the patient has reacted before, and how far off the prescribed interval they are. A pilot study of 16 different missed-dose adjustment protocols found wide variation between practices, and half of them counted the delay from the last dose actually received while the other half counted from the missed scheduled date. Your protocol is one of many, and no phone agent can infer it.

Missed weeks are also where courses quietly die. In a retrospective study across 7 Italian allergy centers, 26.8% of patients on a conventional 7-injection build-up were still in treatment at 3 years, against 63.4% of those on an abbreviated 4-injection schedule, and first-year dropout was 46.4% versus 9.1%. That is one non-US study of schedule design rather than a benchmark for your practice, but the direction is worth taking seriously: the build-up phase is where patients fall out, so a gap that nobody chases is often the end of the course.

Reaction calls, and the line an AI receptionist never crosses

Systemic reactions happen in roughly 0.2% of injections on conventional schedules, and 27% to 50% of them start after the 30-minute wait has ended. So your practice does take reaction calls from patients who are already home, and every one of them needs a person immediately, with no triage attempt in between.

The reaction rule has to be absolute, because a phone script that asks "how severe is it?" has already started triaging. Epinephrine is the treatment of choice in anaphylaxis, and nothing about that decision belongs on a phone tree. The correct behavior is short: tell the caller to call 911, get a clinician on the line, ask nothing else.

That boundary is the same one every safe deployment draws, and it is worth understanding before you evaluate vendors. What an AI receptionist actually does and where it stops sets out the general version of it. The allergy-specific part is that your emergency call has a known clinical trigger, so you can test the routing against real scenarios instead of hoping.

Staffing to a pollen curve, and the biologics calls under it

Your call volume is seasonal in a way most specialties are not, and the curve is predictable. The base rates are large: 25.2% of US adults and 20.6% of children had a diagnosed seasonal allergy in 2024, per the National Health Interview Survey. When the pollen arrives, the phones follow.

The season is not fixed, either. EPA's ragweed indicator covers 11 North American monitoring sites and reports season length through 2015, with a statistically significant lengthening trend at 7 of them. Planning against last decade's calendar understates the shoulder weeks.

Underneath the seasonal noise sits a slower, heavier call type. Injectable biologics for asthma and chronic urticaria bring prior authorization and specialty pharmacy coordination, and those callers want status rather than a booking. Under the CMS Interoperability and Prior Authorization Final Rule, impacted payers must return decisions within 72 hours for expedited requests and 7 calendar days for standard ones, starting 1 January 2026. An agent can capture the request and route it. It cannot promise an approval or a date.

What to automate, and what to keep with the people who know your patients

The recurring injection booking is the right thing to automate and the shot room conversation is not. Volume and repetition make the booking a fit for an AI receptionist. Your staff know the weekly regulars by voice, and a change in one of them is often the first clinical signal anyone gets.

The volume argues for automation. The relationship argues against. Both are real.

A workable split looks like this. Automate the routine injection booking, the reschedule, the appointment reminder, the hours and location questions, and the new patient inquiry that ends in a testing appointment. Keep the dose, the reaction, the shot room conversation, and anything about holding a medication with your people.

The mechanics of the recurring booking and its reminders are worth getting right on their own, and we walk through them in AI appointment scheduling and reminders. One more thing to plan for: a large share of allergy patients are children, so the person on the phone is usually a parent booking for someone else, which changes both the verification and the tone. The pediatric version of this problem goes into that in detail.

Set it up against your own shot schedule

Configure against your real injection schedule rather than a demo script, then pilot narrowly. Start with routine injection rescheduling only, for 30 days, with the reaction path and the dose path hard-routed to staff from day one. Read the transcripts yourself in week one, and measure rebooked visits in your own schedule.

Six checks specific to an allergy line, not a generic go-live list.

The failure mode to watch for is not a rude agent. It is a helpful one. An agent trying to be useful will answer a dose question, reassure a patient about a missed week, or book past the number of observation chairs you have. Every one of those is a configuration choice you make before go-live, not a bug you find afterward.

Field note

Pre-flight for an allergy phone line

• A signed BAA is in place before any patient detail reaches the system.
• The reaction path is tested with real wording patients use, not clean sample scripts.
• Booking rules count the 30-minute observation period, and stop at your chair capacity.
• Someone reads a full week of transcripts before you widen the pilot.

On the compliance question, treat it as a contract and configuration matter. Any vendor touching patient details is a business associate and needs a signed Business Associate Agreement. Be skeptical of any page advertising itself as "HIPAA certified", because that certification does not exist.

Route by what the call is. Every call on an allergy line ends with a named owner.

Walk the flow once against your own calls. A reaction goes to 911 and a human. A dose or missed-week question gets rebooked and flagged for nursing. A routine injection gets the whole slot booked, wait included. A biologics question gets logged and routed with a callback. Everything else becomes a structured message.

If you would rather see the size of the leak before you look at tools, the free Growth Leak Audit sizes it from your own numbers, including the injection visits that never got rebooked.