PG&E

Beyond the clunky IVR toward a Jarvis-like reality The ambition of Siri was to create a digital Jarvis, an all-knowing assistant that lived in your pocket. However, as Nikola Mrkšić, Co-Founder and CEO of PolyAI, notes, the technology simply wasn't ready for that leap. The gap between the dream of a conversational companion and the reality of a frustrating, button-pushing IVR system left a void in the market. PolyAI was born to fill that middle ground, focusing on high-volume, high-stakes human-computer interaction that actually works. By moving away from the "cost-saving" mindset that plagued early automation, Mrkšić is building a world where AI doesn't just deflect calls—it manages them with a level of care that signals a brand's commitment to its customers. Today, the vision has scaled into a platform that has processed over half a billion conversations. The goal is no longer just to prevent a phone from ringing; it is to ensure that when it does ring, the response is immediate, intelligent, and capable. This isn't just about software; it’s about a fundamental shift in how enterprises communicate with their base. Whether it's a casino in Vegas or a utility giant like PG&E, the reliability of the voice interface is becoming the new standard for operational excellence. The contact center as an enterprise nervous system Most executives view contact centers as a necessary evil—a cost center dedicated to failure management. When things go wrong upstream, the phones light up downstream. Mrkšić challenges this narrative by positioning voice AI as the "nervous system" of the enterprise. When PolyAI handles a million calls for PG&E during a biblical flood, it isn't just delivering ETAs; it is gathering real-time data on where the business is hurting. This creates a "diagetic enterprise" where information flows back to the brain, allowing companies to fix billing issues or operational errors before they escalate into social media outrages or PR scandals. In hospitality, the impact is even more direct. For restaurants, missing a call is missing revenue. By implementing voice agents that never miss an appointment, businesses see a top-line increase of 5% to 10%. In an industry where the average lifespan is only five years, that margin is the difference between survival and bankruptcy. This shift from "pinching pennies" on labor to "expanding the top line" through availability is the hallmark of a truly disruptive technology. It turns a reactive department into a proactive intelligence layer. Why verticalized AI agents are a distraction A recent trend has seen the rise of hyper-specific AI agents, such as Linda AI for dentists. While these niche players find traction by solving a single problem, Mrkšić remains skeptical of verticalization as a long-term moat. An appointment for a dentist is fundamentally the same as an appointment for a vet, a restaurant, or a hotel. The complexity doesn't lie in the industry jargon, but in the backend integrations. Once a platform like PolyAI productizes the ability to sync with various scheduling and loyalty systems, the industry itself becomes secondary to the capability of the agent. The real battle isn't over who can talk to a dentist; it's over who can navigate the "archaeology" of enterprise software. Large companies often don't know how their own legacy systems work. The documentation is lost, and the experts have retired. A voice AI company that can step into that messy environment and successfully integrate with a homegrown loyalty system or a custom CRM builds a moat that is quadratically proportional to the number of its integrations. This stickiness makes it nearly impossible for a competitor to rip and replace the solution, regardless of how specialized they claim to be. The trap of the AI wrapper business model There is a brewing conflict between "full-stack" AI companies and those building on top of third-party models like OpenAI. Mrkšić is blunt: companies that rely solely on other people's tech lack strategic autonomy. These "wrapper" companies—such as Sierra or Decagon—are effectively value-added resellers. They are betting that model costs will plummet, but they are vulnerable to the whims of their suppliers and the demands of their customers' IT departments. Outcome-based pricing—charging for a successful result rather than time—often looks like a genius move until the first renewal. When a vendor charges $2.00 for an outcome that their customer realizes they could build internally for $0.30 using OpenAI and some clever prompting, the pricing power evaporates. PolyAI avoids this by owning its models and maintaining transparent, consumption-based pricing. This approach ensures healthy gross margins and provides a "retreat position" that resellers simply don't have. In the long run, the companies that build their own technology will have the leverage to survive the inevitable commoditization of the model layer. Engineering a partnership with Nvidia Defensibility in AI is increasingly tied to the depth of technical collaboration. PolyAI has cemented its position by becoming a key partner for Nvidia, running massive volumes of real-time conversations on the Nvidia Riva framework. This isn't just about buying GPUs; it’s about a technical congruence where Nvidia provides the hardware and software primitives that PolyAI uses to advance its specific conversational data sets. This "big data moat" is built from years of enterprise deployments. While off-the-shelf models are becoming impressive, they cannot match the performance of a model trained on specialized, high-quality conversational data. This is why Jensen Huang and Nvidia have leaned into the partnership. By focusing on being the most technical player in the space, PolyAI isn't just riding the AI wave—it is helping to build the surfboard. For investors and founders alike, the lesson is clear: long-term success requires more than just a cool demo; it requires control over the full stack and the courage to take the hard path of technical innovation.