Native or third-party voice AI on Genesys Cloud: how to choose

Genesys positions Genesys Cloud as a platform that runs its own AI agents and orchestrates other vendors' agents too, so the choice is rarely all one or the other. In practice an organisation can mix routes journey by journey: a scripted bot flow for balance enquiries, an agentic virtual agent for changes that need several systems, and a specialist third-party voice agent for a particular language or brand voice. The table summarises the five routes as Genesys and the vendors document them today.

The native routes live where the rest of your contact centre already lives. Voice bot flows are built in Architect for structured, predictable journeys and are documented as suitable for secure call flows. AI Studio adds Genesys virtual agents and agentic virtual agents, which use data actions as their tools, are grounded in knowledge you govern, and run within guardrails you configure, including rules that detect attempts to override their instructions. Agentic virtual agents are powered by Large Action Model technology that Genesys positions as action-oriented and less prone to invention than general-purpose language models.

The practical advantages are ownership and visibility. There is one vendor, one support path and one place where conversations, recordings, analytics and quality management sit. Changes go through the same flow publishing and change control as the rest of your routing. The trade-offs are equally clear: you use the models and voices Genesys provides, native agents are metered in tokens whose cost depends on interactions rather than minutes, and Genesys states that Virtual Agent and agentic virtual agents are not available in premise-only telephony configurations.

A third-party route earns its place when it brings something the journey needs and the native options do not: an existing bot estate and team you want to keep, a particular language or voice, a specialist capability, or a group-wide decision to standardise on one AI platform. Genesys makes this deliberate rather than awkward: the Lex V2 and Dialogflow integrations are called from Architect like any other action, and Audio Connector streams call audio in both directions to an engine of your choice while Genesys keeps the call, routing, queues and agents.

The cost of that choice is not only the vendor's price. It is a second contract, a second support path, a second place where customer audio is processed, and a hand-off that has to be designed and tested in both directions. Some routes also have documented limits, so they belong in the decision from the start, not after a pilot.

• Audio Connector is not supported on BYOC Premises or premises-based Edge, and allows five integrations per organisation by default
• The Audio Connector stream is active only in the IVR and does not continue when the call transfers to an agent
• Dialogflow CX is not supported on BYOC Premises or premises-based Edge; Dialogflow ES is not PCI compliant for secure flows
• Genesys support for the Nuance Mix Bot integration ends on 31 May 2027

We work through the same questions for every journey, with the people who own the risk as well as the people who want the capability. The answers usually decide the route before price is discussed, and they become the design decision log that explains the choice long after the project team has moved on.

Our voice AI cost calculator applies each vendor's published list prices to your own volumes, so here we describe only the patterns. Native voice bot flows are metered by the minute, while Genesys virtual agents and agentic virtual agents are metered per interaction: an agentic interaction consumes the same tokens as about twenty minutes of voice bot, and a Virtual Agent interaction about eight and a half. Third-party routes add a Genesys per-minute platform rate to the vendor's charge, which matters at volume and is easy to miss in a vendor's own pricing.

Two other patterns change the answer more than any rate. Containment is the biggest lever: a call the AI cannot resolve pays for the AI and then for the agent. And per-user AI features, such as Agent Copilot, share the same free token allowance as your bots, so a large contact centre may find every bot token is a paid one. That is why we compare routes by cost per resolved call, beside the cost of handling the call with an agent.

Control is the question boards ask most, and the answer differs by route. With native AI, the guardrails, knowledge, tools and versioning sit in Genesys Cloud: Genesys versions the agent model behind agentic virtual agents, with an opt-in period to test a new model before it becomes the default. With an external engine, the same questions apply to a second provider: which model it uses and where that model runs, whether customer audio is retained, how changes are released, and who can see what the agent said. Streaming customer audio to an external service is a data decision for your security and privacy owners, recorded before the first call, not after.

Whichever route you choose, the same disciplines keep AI dependable: an interface contract for every tool and hand-off, guardrails agreed with compliance owners, testing with real customer language and real calls, a fallback when a provider is slow or unavailable, and monitoring after go-live. Callers should be told when they are talking to AI, and always have a way to reach a person. These are design decisions, and they cost far less to make early than to retrofit.

QVCCS is a Genesys referral partner, not a reseller, and we receive nothing from any AI vendor, so our only interest is the route that works for each journey. We start with contact reasons, volumes and handle times, record an options analysis for each candidate journey covering capability, cost per resolved call, telephony fit, data and support, and then prove the shortlist with real calls placed from outside the platform, the way customers call. The decision, and the reasons for it, go into the design decision log, so the choice can be revisited calmly when prices, models or regulations change.