TasksAI receptionists and voice agents
Book appointments from missed calls
A call routes to the assistant. An authorized caller accepts an available slot and the calendar contains exactly one matching event.
- Human decision point
- Caller confirms slot, time zone, and details before a calendar write.
- What counts as passing
- Connected call, final approved slot, returned event ID, and matching calendar read.
- Listings
- 7 published listings
- Page data updated
Results on record
Third-party simulation: Cekura voice-agent workflows benchmark, frozen v1
Published by Cekura, last updated . 10 provider-selected configurations, 82 scenarios each (59 appointment, 23 Medicare), three retained runs per scenario: 2,460 retained runs. Caller and backend fixtures are simulated. Cekura sells voice testing and monitoring and discloses that interest.
- Repeatability: share of the 82 scenarios where all three runs passed the applicable scoring gates.
- Task completion: calls with an expected outcome and scored evidence.
- Infrastructure reliability keeps all 246 calls per configuration, including failed connections.
| Configuration | Repeatable scenarios | Share of 82 scenarios | Published completion and denominator | Context |
|---|---|---|---|---|
| PolyAI | 72/82 87.8% | 98.78% across 246 calls | Collected Oct 21, 2026. Repetition and termination were unscored, so gates differ in coverage. | |
| Retell AI | 62/82 75.6% | 93.9% reported | Interruption mean 5.00/5. One published example captured a number in the transcript but sent a different number to the tool. | |
| ElevenLabs ElevenAgents | 57/82 69.5% | 91.5% reported | Mean response 1.27 seconds. One published example continued with an invented result after narrating a tool call. | |
| Telnyx Voice AI | 56/82 68.3% | 97.56% across 246 calls | One published interruption example skipped completion of an opening disclosure. | |
| Vapi | 49/82 59.8% | 97.56% across 205 calls with outcome evidence | 41 of 246 calls did not connect; they stay in the infrastructure denominator. |
No cost per completed call is published, so none is shown. Sources: benchmark and method · disclosure · code license · test-case dataset (CC BY 4.0).
Vendor-reported results
Damaruka: Six of six checkups booked in the vendor’s sample.
Vendor test calls. The result applies to that sample and its disclosed method. Reported by Damaruka about its own product. Cost per completed booking and comparative outcomes are unknown. Published result
Listings and fit
Two different things are sold as an AI receptionist. A configurable voice platform gives you an agent, tools and telephony to assemble: you bring the calendar, the transfer destination and the exception policy. A managed receptionist service packages answering, scheduling and escalation as a plan. Compare them on the task, not the label.
| Listing | Documented for this task | Unknowns | Evidence | Published price | Sources |
|---|---|---|---|---|---|
| Configurable voice platforms | |||||
| Retell AIConfigurable voice platform · Retell AI | Documents checking availability, booking, rescheduling and cancelling through its current Cal.com integration. | Missed-call routing to the agent and booking accuracy are not tested here. | Third-party benchmarkPublisher documented | $0.07 to $0.31 per voice minute across configurations USD/voice minute Price source · checked Model, voice, telephony, transfer legs and add-ons affect the bill. |
|
| VapiConfigurable voice platform · Vapi | Google Calendar tools check availability and create events. | You configure the phone route and recipient. Booking accuracy is not tested here. | Third-party benchmarkPublisher documented | $0.05 per minute for Vapi hosting USD/hosting minute Price source · checked Speech, model, and transport costs are separate. |
checked |
| ElevenLabs ElevenAgentsConfigurable voice platform · ElevenLabs | Workflows can run a configured backend tool for a booking. You supply and authorize the scheduling backend. | No booking backend is included. Scoped to ElevenAgents, not the packaged Reception product. | Third-party benchmarkPublisher documented | $0.08 per extra voice minute, $0.16 per burst minute USD/overage minute Price source · checked Plan allowances and LLM usage are separate. |
|
| Telnyx Voice AIConfigurable voice platform · Telnyx | Inbound assistants with callable tools. A booking action depends on the scheduling backend you connect. | No calendar integration is documented in the recorded source; you supply one. | Third-party benchmarkPublisher documented | $0.05 per minute for voice engine, orchestration, hosted STT and TTS USD/voice minute Price source · checked LLM and telephony costs are separate. |
checked |
| Managed receptionist services | |||||
| Smith.ai AI ReceptionistManaged receptionist service · Smith.ai | Managed service: AI answering, call routing and AI scheduling with supported calendars. | Confirm calendar support and escalation behavior for your plan. | Publisher documented | Free tier lists 25 calls/month. Pro example $150/month for 75 calls, $2.50 per additional call. USD/month and USD/extra call Price source · checked Published example. Confirm plan and taxes. |
checked |
| DamarukaManaged receptionist service · Damaruka | Vendor describes booking in Google Calendar, answering calls and call summaries. | Live transfer and calendar error recovery are unverified for this record. | Vendor-reportedPublisher documented | $19/month for 150 minutes, $49 for 500, $129 for 1,500 USD/month with minute allowance Price source · checked Published tiers are separate from a measured completion cost. |
checked |
| Enterprise voice platforms | |||||
| PolyAIEnterprise voice platform · PolyAI | Documented healthcare flow looks up a patient, finds availability and creates an EHR appointment. | Requires enterprise onboarding and a supported EHR integration. General small-business calendars are not documented here. | Third-party benchmarkPublisher documented | Quoted per minute quote Price source · checked A separate model API rate does not establish the managed platform rate. |
checked |
Prices are shown in each publisher’s own unit (per minute, per call, per month, per page) and are not converted. A price is not a cost per completed task: that needs metered spend and a success count from the same comparable test.
Planned AgentsByTask test
Proposed. No observed result.
Pass means: An unanswered business test number routes the caller to the configured overflow agent. The agent reads a real available slot in the test calendar. After the caller accepts it, exactly one matching event exists with the right time zone and returned event ID.
Setup: Owned test number and a dedicated calendar. The caller confirms the write. A scheduled callback is a separate variant that needs consent and a record of the callback.