In the world of artificial intelligence and tech constantly hunting for the next thing over the horizon, there is definitely a kind of vibe in the air at the moment that voice is the new interface.
When GetVocal CEO and Co-Founder Roy Moussa hears this talk and thinks about the evolution of voice technology, however, he doesn’t reach for a futurist metaphor. Instead, he goes straight back to the Stone Age.
“Voice, I would say it's the oldest interface, right?” he said. “When we were cavemen, we passed on stories by mouth. Now we just did a loop around and went back to the richest modality we have in explaining our thoughts.”
That mix of basic human perspective and cutting-edge tools has become something of a signature for GetVocal, the Paris-based conversational AI startup that has just raised $26 million in Series A funding led by Creandum, with Elaia and Speedinvest participating. It’s the second round this year for the company, an indication of both its own momentum as well as the growing interest in the Voice AI space by venture capital.
Moussa recognizes that the space is getting heated. At the same time, beyond his confidence in the company's technology, he also believes that it is taking a fundamental approach that will set it apart from competitors. Rather than obsessively trying to deliver fully autonomous AI agents to enterprises, the company instead wants to create ones they can trust.
For GetVocal, that means finding the right balance between humans and AI.
By developing a hybrid system where AI agents work under human supervision while gradually learning to handle more complex tasks, GetVocal aims to create a service that delivers actual gains in efficiency without overpromising full-scale automation.
"It's the best of both worlds approach," Moussa said. "It's how to make decisions in real-time conversations and take actions that affect your revenue or affect your customer relations and brand perception. How to make that confidently, responsibly, and according to your business logic and business protocol, while still leveraging the best of what LLMs today give us."
The Voice Renaissance
GetVocal's timing coincides with a broader resurgence of voice-based interfaces, powered by advances in speech recognition and natural language generation.
Over the summer, CB Insights noted that Voice AI has become "the new battleground in the race to build the future of human-machine interactions." It pointed to Meta's acquisition of PlayAI for an undisclosed amount this year, as well as the surge of VC funding in the space that began last year.
Advances in technology have made voice AI more reliable. Such AI agentic chatbots can respond almost instantly with a flow that comes close to mimicking human conversation.
In Europe, ElevenLabs has been red hot, gaining a global footprint in the voice-audio AI space from its perch in Sweden. Meanwhile, just last week, Paris-based Beside announced it had raised a round of $32M for its AI-powered telephony & messaging assistant for small businesses that captures inbound calls and texts and turns them into actions (bookings, follow-ups, notes, scheduling) by rebuilding the full telephony stack with AI for the “real economy.”
Moussa said that when it comes to any kind of human interaction, voice offers a powerful advantage because it conveys tone, emotion, and nuance that text cannot.
"The maturity of the technology has arrived at a stage that it can enable a lot," he said.
While projections for just how far this trend might go can vary quite a bit, there is general agreement that there is a big wave of growth coming.
The Founders' Journey
Moussa and his GetVocal Co-Founder Antonin Bertin aren’t newcomers to conversational AI.
"I'm an engineer that's gone rogue, an engineer that went to the business side," Moussa said. "My co-founder, CTO, and longtime best friend, he went even deeper into tech, into becoming really one of the most technical and smartest people I know."
Almost a decade ago, the duo co-founded Qopius, a retail tech startup that helped "Brick and Mortar focus back on customer experience by automating all tedious shelf monitoring work using computer vision in robots and IoT cameras." They sold that company to Trax Retail in February 2020.
They teamed up again to start Kompanion Care, a healthtech company that uses the "latest in AI technology to enhance the lives of families and individuals. Our app, Lilia, leverages the power of generative AI to turn routine digital communication into enriching experiences, fostering cognitive engagement and combating social isolation."
With the rapid advancements in LLMs, they decided to spin off the healthtech business and focus on enterprise customer service. The founding team also brought along seven colleagues who had been working together for years, creating what Moussa describes as unusual efficiency and speed for a startup.
The Protocol Problem
While many competitors wrap large language models with basic instructions and hope for the best, GetVocal took a different approach to its architecture. After 13 years of experience scaling enterprise machine learning systems, the team concluded that pure LLM-based approaches couldn't deliver the reliability enterprises need.
Instead, the company built what Moussa describes as "protocol automation." This graph-based system encodes business rules and procedures with mathematical precision. It broke down all the parts of the customer service job into their most elemental parts to understand what tasks could be performed by AI agents and what needed to be handed back to humans.
"We really took [LLMs] to the limit of their capabilities," Moussa said. "What are their actual limitations if they are to provide a customer service rep's job? What are their capabilities?"
Learning Like Humans
Once that system is established, it doesn't remain static. The agents continue to learn. This architectural choice reflects GetVocal's broader philosophy about AI deployment.
"We really tried to mimic how humans get on the job, learn, and improve," Moussa explains. "You never hire a person and expect them to be the same as day one. Everything we saw out there is pretty static. They put out an LLM, they give it a document and an instruction. That's not how it works."
GetVocal's agents improve through multiple mechanisms that mirror human learning. First, there's data-driven optimization.
"Every time a conversation speaks, every time a person speaks in front of that AI, we're measuring," Moussa said. "We split conversations. We turn conversations into a science."
The system also incorporates direct human feedback through real-time shadowing. This hybrid approach is much like junior employees learning from supervisors. Over time, these interactions transfer skills to the AI agents.
"It takes that human feedback in the most intuitive way, just like shadowing in real time, and it improves upon it," Moussa said. "An AI agent today, on a particular conversation that he's having, he can go and refer to a human supervisor, his manager, to either ask for a pending approval in real time and go back and finish that customer journey, to ask for help when a conversation's gone critical, either good or bad."
Market Validation
Last summer, an MIT study showed that 95% of AI pilots at enterprise companies failed to deliver value due to poor governance and integration. It has become a centerpiece in the ongoing debate over whether there is an AI bubble.
GetVocal has specifically tailored its hybrid system in a way designed to demonstrate that such systems can deliver ROI in ways that are clear to executive decision makers by targeting those narrower sets of tasks and then expanding them.
“Companies are very excited that there are LLMs," Moussa said. "How can we fulfill it in the most reliable way?”
That approach has helped GetVocal grow quickly in Europe. It now serves 23 markets with a 60-person team, counting Vodafone and Movistar among its customers, with a pilot underway at Deutsche Telekom.
The company claims its AI agents drive 31 percent fewer live escalations and 45 percent more self-service resolutions compared to existing enterprise solutions, achieving a 70 percent deflection rate within three months of launch. Glovo, the food delivery platform, grew its agent fleet from one to 80 AI agents in less than 12 weeks, achieving a five-fold increase in uptime and a 35 percent increase in deflection.
The business model will evolve along with the technology. The company currently operates on consumption-based pricing.
"What we want to get to very quickly is an outcome-based model," Moussa said. "It's the only way to align customer and service providers on the same page."
Even as that advances, the company intends to keep the human-AI approach at its core. Rather than talking about replacing people, Moussa speaks in terms of transitions, apprenticeships, and supervisory loops.
He believes that the future of AI in the enterprise will be gradual and deeply human. He said the company's mission is to help guide enterprises and employees through a fundamental workforce transition.
"Do we obsess over whose AI agents are better?" Moussa said. "No. We obsess over ensuring that [customers] do this transition of the workforce in the easiest, most trusted, and human way."
