According to ElevenLabs, the number of calls an AI-based voice agent can handle depends on three factors: the concurrency limit of the subscribed plan, the capacity of the phone lines, and the response speed of the language model and CRM integrations. The publisher presents these criteria as the points to check before sizing a telephone reception system.
A human team absorbs simultaneous calls by adding staff, with each employee handling only one call at a time. For a voice agent, the key metric to look at is concurrency, meaning the number of calls handled at the same moment, rather than the total minutes advertised by many platforms, which says little about the real capacity of a setup.
A limit set by the subscribed plan
At ElevenLabs, concurrency starts at 4 calls with the free plan and reaches 40 lines with the Business plan, with higher limits available for Enterprise customers. “Burst pricing” allows temporarily exceeding the limit during periods of high demand, up to three times its level, with a cap of 300 simultaneous calls for non-Enterprise plans.
Other elements determine the actual capacity. SIP trunks, which are IP telephony lines, and carrier lines set the physical traffic limit and must be sized for peak calls rather than daily average. Language models, or LLMs, process callers’ requests, and the publisher states they generate responses “at speeds under a second.” Finally, integrations with the CRM, customer relationship management software, query databases during the call to check an account or a customer’s history.
To estimate the need, ElevenLabs proposes a three-step method. The example given is that of a SaaS company receiving at most 75 calls at peak hour, at 2 PM EST. By multiplying this volume by an average duration of 4 minutes (75 × 4 = 300), then dividing by 60, you get 5 simultaneous calls. The publisher recommends adding a 20% to 30% margin to absorb unexpected peaks, which in this example leads to provisioning 6 lines.
Activity peaks, transfers and pricing
Calls beyond the limit are billed at a different rate. Overage calls are billed at double the standard rate, i.e., $0.16 per minute, and are lower priority, which can result in slightly higher latency. If the queue is activated, callers wait and are connected as soon as a line becomes free, with no charge for waiting time.
ElevenLabs cites several situations likely to cause these sudden surges: a home insurer after a tornado, an e-commerce merchant whose niche product goes viral, an airline after flight cancellations, a telecom operator after an outage, or a bank facing a fraudulent email announcing the freezing of its customers’ accounts.
When the request exceeds the automated framework, the agent can transfer the call to an advisor using a tool to transfer to a number. With the native Twilio integration, it can send a brief message to the advisor before connecting, and, with Amazon Connect, a webhook can write a summary of the conversation before the transfer. Beyond reception, the publisher lists appointment scheduling, lead qualification, and payment reminders, but specifies that high-stakes negotiations should remain entrusted to a human.
On the pricing side, subscriptions determine the base capacity with no per-line fee. Additional minutes cost $0.08 per minute on every paid plan, while LLM and telephony costs are billed separately, based on usage.
Finally, compliance weighs on deployment. ElevenLabs cites the Telephone Consumer Protection Act (TCPA) in the United States, the Health Insurance Portability and Accountability Act (HIPAA) in the United States, the General Data Protection Regulation (GDPR) in the European Union, as well as Regulation (EU) 2024/1689, known as the AI Act. The publisher recommends consulting a legal advisor for the relevant markets.
