French startup Rounded anticipates that AI voice agents will soon become the standard method for customer engagement with businesses. Rather than developing ready-to-deploy AI voice agents, the company is creating an orchestration platform that empowers businesses to create their own bespoke voice agents.
Initially, Rounded focused on a web3 product but transitioned to explore AI voice agents in June 2023. “Our concept was essentially to place ChatGPT after a transcriber and before a synthesizer to deliver value,” shared co-founder Aymeric Vaudelin (illustrated above, far left) in an interview with StartupSuperb.
However, the team encountered the typical challenge of achieving product-market fit. “After several months, we recognised that the market wasn’t quite ready to embrace voice agents. Therefore, we developed a solution and tailored it to launch our initial agent,” Vaudelin explained.
This endeavour led to the creation of Donna, an AI voice agent designed specifically for anesthetists. Although this choice may seem unusual, the startup targeted this sector because anesthesia secretaries manage numerous patient interactions, which tend to be largely transactional.
In France, patients must consult with the anesthetist before scheduling surgery to confirm that there are no allergies to anaesthetic substances or underlying health issues.
Anesthesia secretaries are responsible for managing a high volume of simple calls; typically, patients seek information about the anesthetist’s availability, wish to book an appointment, or need to reschedule.
Furthermore, as these are not sales inquiries, the AI agent does not need to demonstrate exceptional persuasion or efficiency. “Initially, we dealt with latency issues of four to six seconds,” Vaudelin remarked.
Despite these challenges, Rounded successfully persuaded 15 private hospitals to allow an AI voice agent to handle incoming calls. The company reports that the agent has now conducted hundreds of thousands of interactions. Over time, Rounded has fine-tuned its technology to enhance accuracy, integrations with other products, and, importantly, speed.
“For example, our web calls now feature latencies of under 700 milliseconds — more typically around 600 milliseconds. You can factor in an additional 200 milliseconds for phone connectivity,” Vaudelin explained.
Recently, Rounded expanded its offerings with an orchestration product that allows other companies to develop their own voice agents.
Rounded provides the flexibility to select from various off-the-shelf AI models, including speech-to-text, large language models (LLMs), and text-to-speech models. For instance, when creating your first voice agent, you might utilise Azure for call transcription, GPT-4o mini as the LLM, and ElevenLabs as your text-to-speech engine.
The platform assists in delineating the instruction trees and prompts tailored for your specific needs. “Creating an agent requires identifying optimal prompts, parameters, and variables within the prompt,” Vaudelin noted.
“Our proposition is to empower everyone to create effective prompts or agents, while our product will aid them in the iterative process of developing highly reliable agents,” Vaudelin added.
As of now, Rounded has successfully raised €600,000 (approximately $620,000) from UC Berkeley’s deeptech accelerator SkyDeck along with several angel investors. Given the ongoing excitement surrounding the artificial intelligence sector, there is a strong likelihood that the startup will secure additional funding in the near future.