Buy the responsibility model, not the category label. The documented offerings below illustrate the available routes; the final column states our recommended procurement test.
| Provider type | Documented examples and scope | Buyer-fit assessment | What to verify before shortlisting |
|---|---|---|---|
| Vertical application or native operational-system agent | ServiceTitan documents job booking, confirmation, rescheduling and configurable escalation. Arini describes Open Dental scheduling integration. (help.servicetitan.com) | A sensible starting route when your workflow closely matches the supported sector and system | Exact account eligibility, location rules, permissions and supported transaction types |
| Configurable voice-agent platform | Retell advertises APIs, webhooks and simulation testing. Vapi prices hosting separately from model consumption. Bland publishes platform and connected-minute plans. (retellai.com) | Suitable when a buyer or implementation partner can own configuration and integration | Who builds and maintains the workflow; what production support is included |
| Contact-centre suite | Genesys documents virtual-agent licensing and token consumption. Five9 announced a new Voice AI Agents release on 23 June 2026. (help.genesys.cloud) | An important comparator for an existing suite customer | Incremental licences, consumption, routing changes and the quality of the actual human handoff |
| Enterprise managed specialist | Replicant describes a fully managed offer with AI engineers, conversation designers and implementation specialists. (replicant.com) | Relevant when the buyer wants an accountable delivery team rather than only a platform | Contracted deliverables, change capacity, monitoring ownership and incident responsibilities |
| Speech and agent infrastructure | ElevenLabs documents conversation simulation, next-response tests and tool-call tests within its agent platform. (elevenlabs.io) | Evaluate the specific agent product, not merely the supplier’s reputation for speech generation | Telephony, workflow integrations, permissions and support around the speech layer |
| Programmable telephony and custom application | Twilio Conversation Relay handles speech recognition and synthesis while connecting to the customer’s application. Amazon Nova Sonic documents external tool and API use. (twilio.com) | Appropriate when application ownership and custom engineering are intentional | Application hosting, authentication, tool execution, recovery and ongoing engineering costs |
| Human/hybrid reception | Smith.ai describes AI-first, hybrid and human-first options, including configurable routing to people. (smith.ai) | A useful baseline or alternative when callers and exceptions need human attention | Which calls receive people, their authority, coverage and additional charges |
| Integrator or managed automation agency | Treat the proposed implementation as its own procurement, separate from the underlying platform | Useful when the buyer needs outsourced technical ownership | Named platform dependencies, subcontractors, account ownership, acceptance tests and exit rights |
A native product should not receive an automatic quality advantage. Equally, a standalone specialist should not receive an automatic innovation advantage. Require both to demonstrate the same business-system changes and exception handling.