Concurrent calls is the number of callers an AI phone agent can talk to at the same moment, not how many it handles in a month. A busy clinic should size it from its busiest 15 minutes, not its monthly total, and ask each vendor exactly what the next caller hears once every line is in use.
Most AI phone agents answer far more calls than a front desk can. But each one still has a ceiling on how many calls run at once, set by the plan or the platform. For most of the day you never reach it. For a clinic, the moment you might is predictable: Monday morning, when the weekend's symptoms, the week's bookings and the reschedules all call in at once.
This guide explains what concurrent calls means, how to estimate the number you need from your own call log, what different products do when the limit is hit, and what to ask before you buy.
What does concurrent calls mean for an AI phone agent?
Three numbers often get mixed up on pricing pages:
| Term | What it measures | Example |
|---|---|---|
| Total calls | How many calls arrive in a month | 900 calls a month |
| Minutes | How long all those calls last added together | 2,700 minutes a month |
| Concurrent calls | How many calls are live at the same moment | 5 at once |
A product can handle unlimited calls a month and still only answer five at a time. Concurrency is a width, not a length. Think of it as the number of phone lines into the building. A wider road doesn't make the trip longer; it lets more cars through at rush hour.
For a human front desk, concurrency is the number of people free to pick up. For an AI phone agent, it is a limit set by the plan, the provider's infrastructure and sometimes the phone carrier.
How to estimate the peak you need from your call log
You don't need a formula sheet. You need your call log for the last month or two, which most phone providers can export with a start time and length for each call.
Step 1: Find your busiest 15 minutes
Sort calls by start time and look for the 15-minute window with the most calls. For clinics, it is almost always early Monday or the first hour after a holiday.
Step 2: Find your average call length
Add up the minutes of answered calls and divide by the number of calls. Booking calls are usually two to four minutes.
Step 3: Work out the average number of calls live at once
Multiply the calls in your busiest window by the average call length, then divide by the window length.
Calls in busiest 15 minutes × average minutes per call ÷ 15 = average calls live at once
Step 4: Add room for bunching
Calls don't arrive neatly spaced. Some minutes get none, others get three at once. This is the intuition behind the Erlang models call centers use: random arrivals mean the real peak sits above the average. A simple rule of thumb is to add about twice the square root of your average. You don't have to be exact; you want a number that covers almost every moment, not the average one.
A worked example: a clinic on Monday morning
A two-doctor family clinic exports its log. Most of the week is quiet, but between 8:00 and 8:15 on Mondays it sees 14 calls. The average call lasts 3 minutes.
| Step | Calculation | Result |
|---|---|---|
| Calls in busiest 15 minutes | From the call log | 14 |
| Average call length | From the call log | 3 minutes |
| Average calls live at once | 14 × 3 ÷ 15 | 2.8 |
| Room for bunching | About 2 × √2.8 | about 3.3 |
| Lines to plan for | 2.8 + 3.3, rounded | about 6 |
So this clinic should plan for roughly five to six calls at once at its peak. With two lines, several callers in that window would hear a busy message. With five, a rare caller might at the very worst moment of the week. With six or more, almost nobody would.
If the front desk picks up first and the AI takes the overflow, the AI's share of the peak is smaller. Two staff on the phones at 8:00 take two of those six lines off the AI.
Use your busiest real week, not an average one. Flu season, the first Monday after a long weekend and the week a new doctor joins are the moments that test your limit.
What happens when an AI phone agent hits its limit?
Every product has to do something with the caller who arrives once every line is in use. The common options:
| What happens | What the caller experiences | What to watch for |
|---|---|---|
| Busy message | A short message, then the call ends | Is the call logged so you can call back? |
| Queue | Hold music until a line frees up | How long is the wait, and does it time out? |
| Voicemail | A recorded message from the caller | Who checks it, and how quickly? |
| Overflow to staff | The call rings a person or another number | Is someone free to answer it? |
| Paid burst capacity | The call connects above the limit at a higher rate | What it costs per minute in a busy month |
None of these is wrong. What matters is that you know which one you are buying, and that no caller simply disappears. Some developer platforms publish their approach. Retell AI, for example, lists 20 concurrent calls on every account, extra lines at $8 each per month, and an optional burst setting at an extra $0.10 a minute that connects calls above the limit instead of rejecting them (as listed in November 2026). Vapi and Bland AI also publish concurrency per plan. Many receptionist products don't state a number at all, so ask.
What Aiva Chat does when every line is busy
Aiva Chat's limit is set by your plan: Starter answers 2 calls at once, Growth 5 and Pro 20.
When a call arrives while every line on your plan is in use, the caller hears:
"All of our lines are busy right now. Please try again in a few minutes. Goodbye."
The call is not dropped silently. It is logged in your call history as declined, with the reason, so your team can see who tried to reach you and call them back. Calls are not queued, which means nobody sits on hold not knowing whether anyone will pick up.
A few settings help you keep that message rare:
- Ring desk phones first. During business hours, your front desk phones ring first and the AI answers only if nobody picks up. Staff take what they can and the AI takes the overflow.
- Transfers to people. The AI can transfer a call to a staff member or a department. If the transfer isn't answered, it takes a message and creates a follow-up task, so the caller isn't stuck.
- After-hours choices. Out of hours, you can let the AI answer every call or send calls to voicemail.
- Live call monitor. On a busy morning you can see every live call in the dashboard, listen in, whisper to the AI, or take over a call yourself.
Every answered call has a recording, transcript and summary, and the AI books into its own calendar or Google Calendar during the call. More on the voice agent and on Aiva Chat for clinics.
Monthly volume decides the bill. Your busiest 15 minutes decide whether patients get through.
How should a busy clinic compare AI phone agents on concurrent capacity?
Put the same questions to every vendor and get the answers in writing.
- How many calls are answered at the same moment on the plan I'm pricing? A number, not "multiple" or "unlimited".
- What does the next caller hear when that limit is reached? Ask for the exact message or behaviour.
- Is that call logged? You want the caller's number and time so someone can call back.
- Can I raise the limit, and what does it cost? A higher plan, a per-line fee or a burst rate.
- Can staff answer first? Ringing the front desk before the AI spreads the peak.
- What happens with transfers during a rush? If a transfer isn't answered, does the caller get stuck?
- Does the limit cover outbound calls too? If you run reminder or follow-up calls, they may share the same lines.
Then test it. During a trial, have four or five people call the number at the same moment and see what each one hears. It takes ten minutes and tells you more than any spec sheet.
For the bigger picture on buying for a clinic, read our buyer's checklist for clinics. For how AI capacity compares with a staffed team, see AI call agent vs. call center.
Matching the plan to your peak
Using the worked example, here is how the clinic might think about plans:
| Peak lines needed | Aiva Chat plan | Calls at once | Price |
|---|---|---|---|
| 1 to 2 | Starter | 2 | $49/month |
| 3 to 5 | Growth | 5 | $99/month |
| 6 to 20 | Pro | 20 | $199/month |
The two-doctor clinic needing about six lines has two sensible choices. Growth with desk phones ringing first, so staff absorb a line or two of the Monday peak. Or Pro, so the AI covers the whole peak alone. Both are reasonable; the call log and how many staff are on the phones at 8:00 decide which.
Every plan includes every feature, has no setup fee and can be cancelled any time, effective at the end of the billing period. Call minutes and AI usage are billed by Twilio and OpenAI on your own accounts at their published rates, with no Aiva Chat markup, so check their current rates against your volume. For a fuller look at how AI phone products charge, see AI answering service pricing.
Common mistakes when sizing concurrent calls
- Sizing from monthly totals. 900 calls a month tells you nothing about Monday at 8:05.
- Using the average week. The week that matters is the busiest one.
- Forgetting call length. Longer calls hold lines longer. A clinic whose calls average 5 minutes needs more lines than one averaging 2.
- Assuming "unlimited" means unlimited at once. Ask for the number.
- Not checking what the next caller hears. A silent drop and a logged, polite busy message are very different outcomes for a patient.
Ready to check your Monday morning?
Export last month's call log, find your busiest 15 minutes, and run the sum above. You'll know within ten minutes whether you need two lines, five or twenty.
To hear Aiva Chat answer, call +1 (970) 972-0974. Then compare the plans, start your 14-day free trial with no card and test a few calls at once, or book a demo and bring your call log so we can size it with you.
See how it works: the Aiva Chat AI receptionist
Written by
Shyamsinh Parmar
AI Solutions Architect
Shyamsinh designs production AI systems at Woyce AI, the team that builds Aiva Chat, including the real-time voice architecture behind it. Shyamsinh works from the business process down: where AI should act, what needs a person, which models to use, and what each call costs to run.
More posts


