A concurrent inbound connection is one caller being handled by the AI receptionist at a time. New standalone Starter, Business and Scale plans allow up to two, three and five simultaneous AI conversations respectively. Those are company limits on shared fleet capacity, not dedicated hardware or a guarantee that AI is immediately available.
Think of it as the number of conversations your AI receptionist can hold at the same moment. It is not the number of phone numbers you own, the number of agents you create or the number of calls you receive in a day. For a business choosing a phone answering plan, it shapes how a busy hour feels to the caller who rings at the wrong moment.
What counts toward the limit
All numbers and agents in a company share its AI limit and included-minute allowance. Buying another number or creating another agent does not increase either.
Only a connected conversational AI session consumes AI allowance. Listening, silence and tool waits inside that session count; positive calls have one 30-second minimum, metered per second. A menu, team ringing or prerecorded closed-hours message outside it does not add conversational AI time, because a connection is a caller the AI is handling. Supported voicemail or a human transfer consumes phone time without conversational AI minutes. Managed non-AI incoming and connected transfer legs have their own started-minute prices. BYO carrier charges stay on the customer’s carrier bill.
Two meters, not one
The connection limit and the included-minute allowance measure different things. The limit is about how many conversations run at the same moment. The allowance is about how much conversational AI time you use in total. A company can reach its limit in a busy ten minutes while using a small part of its allowance, and it can use its whole allowance without ever reaching its limit.
A little arithmetic shows the difference. A limit of three, run flat out for an hour, can carry at most 3 × 60 = 180 minutes of conversation in that hour; the included-minute allowance is a separate figure on the pricing page. The 30-second minimum is a separate effect on the allowance. A call with 12 seconds of connected AI time is metered at 30 seconds, and a call with 95 seconds is metered at 95. Forty callers who hang up after 12 seconds therefore use 40 × 30 = 1,200 seconds, or 20 minutes, of allowance, not the 8 minutes their raw durations add up to.
Sizing it: a worked example
Start with your busiest hour, not your average day. The average number of conversations in progress is calls in the hour × average minutes per call ÷ 60. Thirty calls averaging four minutes is 30 × 4 ÷ 60 = 2. That unit is called an erlang, and it is an average: at some moments three callers are talking and at others none.
Calls arrive in clumps, so a limit equal to the average is not enough. The textbook Erlang B formula estimates the share of calls that arrive to find every connection busy, assuming calls arrive at random and a call that finds them busy goes elsewhere instead of waiting. For three busiest-hour loads and the three limits named above, the model gives these shares. It is a textbook calculation, not a measurement of any Vocapable account:
| Busiest hour | Average in progress | Limit of two | Limit of three | Limit of five |
|---|---|---|---|---|
| 12 calls of 3 minutes | 0.6 | 10% | 2% | under 1% |
| 30 calls of 4 minutes | 2 | 40% | 21% | 4% |
| 60 calls of 4 minutes | 4 | 62% | 45% | 20% |
Read the middle row. An average of two conversations in progress, with a limit of three, still sends roughly one call in five to your waiting or fallback path in that hour, because an average is a midpoint and not a ceiling.
Real traffic can be lumpier than the model, a waiting path can hold some callers instead of sending them elsewhere, and shared fleet capacity can be busy while your company is under its own limit. Use the table to decide how much your fallback has to carry.
When every connection is busy
A configured waiting or fallback path handles calls when the company limit or available fleet capacity is busy. What that path is, whether a short wait, voicemail or a human transfer, is something you set up in advance, and it decides what a busy hour feels like to the caller. The fallback decides where the caller goes. It does not add AI capacity.
Common mistakes
- Counting numbers or agents as capacity. They share one limit and one allowance, so another number or agent is another way to reach the same limit.
- Reading the limit as dedicated hardware or a promise. It is a company limit on shared fleet capacity, so a call can meet your waiting or fallback path even when your company is under its own limit.
- Sizing to the daily average. Connections run out in the busiest minutes. Size to the busiest hour and to the bursts that follow an ad, a mailing or a lunch break.
- Mixing up the two meters. Reaching the limit does not mean the allowance is spent, and a spent allowance is not the same event as a busy moment.
- Treating non-AI handling as free. Menus, ringing, voicemail and transfers add no conversational AI minutes, but they still use phone time, priced as described above.
- Assuming the new tiers replaced your contract. Accepted legacy and outbound-attached Receptionist contracts retain their connection-based entitlement and rates. The new standalone tiers do not replace those agreements automatically.
- Confusing it with a concurrent session. The concurrent session is the unit outbound plans are sold in. This is a separate entitlement for inbound answering.
Related
Compare the actual terms and supported fallback on the pricing page and the receptionist page. To work a month of calls through with your own numbers, use the AI receptionist cost calculator or the missed-call cost calculator. Related definitions: concurrent session, BYO telephony and answering-machine detection.