EXCLUSIVE: SoundHound CEO's Winning Formula: 'AI Resolves What It Can, Humans Handle What They Should'

SoundHound AI (NASDAQ:SOUN) is navigating a significant expansion in 2026. After launching the OASYS platform in Q2, the firm completed the strategic acquisition of LivePerson earlier this month. Exclusively for Benzinga, CEO and co-founder Keyvan Mohajer breaks down post-merger cross-sell into Fortune 100 contact centers, taking on cloud giants, and why on-device AI will drive gross margins. BZ: With the LivePerson transaction officially closed on September 4, 2026, you are pairing voice AI with enterprise digital messaging to create a unified omnichannel agentic stack. What does the concrete product integration roadmap look like over the next 6 to 12 months, and where do you expect the earliest cross-sell conversions—bringing SoundHound voice into LivePerson’s Fortune 100 contact center base, or pushing omnichannel messaging into your automotive and restaurant accounts? This is about building one world-class omnichannel agentic platform for enterprises across all industries. OASYS is the single system LivePerson’s digital messaging and orchestration will be unified with, so customers no longer maintain separate voice and digital stacks — they get one agentic layer across both. The combination gives us something rare: a voice-native AI platform paired with an enterprise engagement platform and an established human-agent footprint, in a single system. Much of the market is selling either pure AI deflection or a legacy contact center with AI bolted on. We believe the winning formula is a stack where AI resolves what it can, humans handle what they should, and the two are designed to reinforce each other. Owning both sides is what lets us deliver that. On cross-sell, SoundHound’s footprint already spans financial services, healthcare, retail, restaurants, energy, hospitality, telecom, and automotive, so the earliest conversions run in both directions: voice-native capabilities into LivePerson’s Fortune 100 contact-center base, and digital messaging into our existing verticals. The contact center is where the full-stack advantage is clearest, so that’s our center of gravity. BZ: OASYS was launched to transition enterprises from rigid rules and experimental LLM wrappers to autonomous, self-learning agent orchestration. Across key verticals like automotive, healthcare, and retail/QSR, where are you seeing the highest autonomous containment rates, and how has that performance impacted enterprise sales cycles compared to legacy deployments? We don’t share containment rates by vertical, because performance depends on the use case, how long a deployment has been live, and how much containment the customer actually wants. Containment is a design choice as much as a capability ceiling: a healthcare customer may deliberately route more interactions to a human because of the nature of the conversation, not because the AI couldn’t handle it. What’s consistent is the shift from scripted, rule-matched flows to agentic reasoning that determines intent and resolves the request end-to-end. Quálitas, a leading Mexican auto insurer, runs our agents on roadside assistance and partial-theft claims. More than 74% of roadside calls are now resolved start to finish with no human on the line, with containment up 30%; on partial-theft claims, containment has tripled. These are emotional, high-volume calls the industry assumed still required a person. That also changes sales cycles. Because these agents reason rather than follow hand-built decision trees, they stand up in weeks rather than months, so we can get a pilot live and prove results early. BZ: SoundHound’s placement as a Leader in the 2026 Gartner Magic Quadrant for Conversational AI Platforms puts your tech stack head-to-head with hyperscalers. When enterprise procurement and IT architecture teams evaluate agentic solutions today, what specific technical advantages are leading them to choose SoundHound’s voice-native orchestration over bundling with their incumbent cloud provider? The advantage is architectural, and it holds up on the dimensions procurement teams evaluate: multimodality, guardrails, and human-in-the-loop design. SoundHound owns the full voice stack — speech-to-text, reasoning, and text-to-speech — built to work together natively, so a request is processed in a single pass instead of being handed between a bolted-on speech vendor and a text-first platform. On guardrails, our Agentic+ framework pairs generative reasoning with deterministic execution for regulated or high-stakes actions. We just added Human Assisted Resolution, a proprietary capability that lets an AI agent ask a human colleague for an approval or policy check without handing the conversation over, so the customer never has to repeat themselves. OASYS is also a working platform, not a build-it-yourself kit; we don’t leave assembly, integration, and maintenance to the customer’s team. And unlike hyperscalers, which tie agentic capabilities to their own cloud and models, OASYS runs on-premise, private, hybrid, or public cloud and is model-agnostic. BZ: With EU AI Act enforcement now active and biometric voice and deepfake fraud surging, enterprise Chief Information Security Officers (CISOs) are demanding rigorous audit trails and consent frameworks. How is SoundHound architecting its Agentic+ and OASYS frameworks to guarantee auditability, provenance, and zero-data-leakage for highly regulated healthcare and financial enterprise clients? Auditability starts with how the platform is built, not a policy layered on afterward. Agentic+ gives every interaction three modes: autonomous reasoning for nuance, rule-based execution for anything regulated or high-stakes, and Human Assisted Resolution when a person should weigh in. In financial services, before an agent completes a money transfer, rule-based steps verify identity, check the amount against thresholds, and route to approval where required. In healthcare, a prescription refill runs identity verification and a safety check first, and anything requiring clinical judgment routes to a clinician. The path is defined, repeatable, and logged — which is what an auditor or CISO needs to see. Underneath that, a dedicated AI governance function spanning Product, Engineering, Compliance, and Legal delivers deterministic PII masking and redaction, plus an evaluation framework that tests agent behavior for sensitive-data exposure before deployment. That sits on HIPAA, ISO 27001, SOC 2 Type 2, and PCI certifications, and we’re working to stay ahead of EU AI Act requirements, including watermarking. BZ: SoundHound has long differentiated itself through edge and on-device capabilities, particularly in the automotive cockpit. As compact models and edge inference advance, what portion of your agentic workloads is shifting on-device versus cloud-hosted? How does that split affect your gross margin expansion? We just launched OASYS Edge, available now to automakers, device manufacturers, and service providers. It runs agentic reasoning directly on the device and calls the cloud only when a request genuinely needs live, online information. Our vision is that every business and every product will have an AI agent. Businesses will lean toward cloud; products toward edge. We’ll soon have robots among us that are always listening, and you don’t want that ambient audio streaming to the cloud. Three things drive on-device: privacy, because data stays local; latency, because local processing responds fast regardless of connection; and control, because more of the experience stays in the device maker’s hands. On the economics, the split cuts two ways. For customers, on-device processing avoids per-interaction cloud costs that compound at scale. For us, running our own models at the edge reduces cloud dependency and its cost. As edge workloads grow, that mix works structurally in favor of our gross margin. Photo: Tigarto / Shutterstock