Genesys Adds ElevenLabs and Deepgram Voice AI to Its Contact Center Agents

What Genesys announced

Genesys, the customer-experience software vendor whose cloud platform now serves more than 7,500 organizations, opened its annual Xperience conference in Las Vegas on September 2, 2026 by introducing four new pieces of what it calls its "agentic orchestration platform": Genesys Cloud Navigator, Genesys Cloud Orchestrator, Contextual Intelligence, and an AI Control Plane. Contextual Intelligence and the AI Control Plane are available immediately; Navigator is targeted for general availability in Genesys's fiscal fourth quarter (November 2026 through January 2027), and Orchestrator is further out, targeted for the following fiscal year.

According to Genesys's own newsroom announcement, Navigator is meant to be the AI-native entry point for a customer interaction -- reading intent from the first message or utterance and routing the conversation to the right workflow, AI agent, or human employee. Contextual Intelligence sits underneath that, stitching live interaction signals together with customer identity, history and business events into what Genesys describes as "persistent enterprise memory." The AI Control Plane is the governance piece: centralized discovery, identity, policy and observability controls that let an enterprise define exactly where its AI agents are allowed to act, what data they can touch, and when a human has to sign off.

Source: Genesys newsroom, September 2, 2026.

The voice AI upgrade: ElevenLabs and Deepgram join the virtual agent

The piece of the launch most relevant to the voice AI industry is the update to Genesys's Agentic Virtual Agent, or AVA -- the conversational AI agent that handles live customer calls and chats inside Genesys Cloud. Genesys rebuilt AVA on Scaled Cognition's APT-2 large-action model, which the company says improves reasoning accuracy, factual grounding, task-execution reliability, response latency and multilingual handling. Layered on top of that model, AVA now runs Deepgram for real-time speech recognition -- available immediately -- with an ElevenLabs integration for more expressive, natural-sounding text-to-speech following in late October 2026, per CX Today's reporting on the announcement. Genesys says fuller native-voice capabilities, including end-of-turn detection (recognizing when a caller has actually finished speaking, not just paused) and contextual tuning, are planned for the first quarter of 2027.

That pairing is notable because it mirrors a pattern playing out across the industry: rather than building proprietary speech models, large CX platforms are increasingly wiring in dedicated voice-AI vendors as swappable components. Deepgram and ElevenLabs are both independent, venture-backed speech specialists -- Deepgram closed a $130 million Series C at a $1.3 billion valuation in January 2026, and ElevenLabs has been in talks for a secondary share sale that would value it at roughly $22 billion. Landing inside a platform with 7,500+ enterprise customers gives both companies a significant enterprise distribution channel without having to win each contact-center deal individually.

Sources: CX Today, September 2026; CMSWire, September 2026.

The numbers behind the launch

Genesys disclosed that Genesys Cloud generated $2.9 billion in annual recurring revenue in its fiscal Q2 (May through July 2026), up more than 30% year over year, according to CMSWire's coverage of the Xperience keynote. Cloud AI ARR specifically topped $400 million and is growing at roughly twice the rate of the overall cloud business. The platform now counts more than 600 customers each generating $1 million or more in ARR, and Genesys says it processes 33 billion conversations and a trillion API calls a year. Genesys's own newsroom post cites specific customer results behind those numbers: airline Virgin Atlantic reports $28 million in three-year value and a 25-point jump in customer satisfaction after deploying Genesys AI; Charles Sturt University saw a 29% improvement in first-contact resolution; and Brazilian retailer Trademark Global reported a 30% increase in service-level performance alongside a 65% cut in technology costs.

CMSWire also reported an even larger result from Brazilian fashion retailer Riachuelo, which said customer retention climbed from 30% to 84%, CSAT improved by 60 points, and agent productivity rose 400% without adding headcount, after adopting Genesys's AI tooling. Genesys says the orchestration layer is also built to plug into outside ecosystems -- the newsroom post lists deepened integrations with Adobe, AWS, Meta, Salesforce, ServiceNow and rival CX vendor Sierra, work Genesys attributes partly to its acquisition of integration-platform Pinkfish in June 2026, which it says brings 500-plus pre-built integrations and more than 25,000 MCP-compatible tools into Genesys Cloud.

Source: CMSWire, September 2026.

Why the governance layer matters as much as the voice upgrade

Mike Szilagyi, Genesys's SVP and Head of Product, framed the announcement around a shift from isolated chatbot deployments to something closer to a connected workforce: "The future of CX is AI that understands intent, preserves memory across resources and moves the interaction forward until the customer's problem is solved," he said, according to Genesys's newsroom post. CTO Glenn Nethercutt tied the AI Control Plane specifically to trust, telling CX Today that organizations "need the freedom to innovate across technologies without sacrificing continuity" -- the argument being that as voice agents get better at actually resolving calls end to end, the harder problem becomes proving to compliance and IT teams which data those agents can see and which actions they're allowed to take unsupervised.

CX Today's analysis also places the launch in a crowded field: Genesys is competing directly with Salesforce, ServiceNow, NICE, Five9 and Microsoft, all of which are positioning their own platforms as the central "control point" for enterprise AI agents. Genesys is betting that enterprises adopting several different specialist AI vendors -- a voice model from one company, a large-action model from another, an orchestration layer from a third -- will still want one governance and memory layer sitting above all of them, rather than rebuilding that plumbing themselves for every new AI agent they add.