French startup Rounded believes AI voice agents are going to become the default way customers interact with companies, so instead of building AI voice agents that are ready to use, the company is building an orchestration platform that lets companies build their own voice agent.
Rounded started off working on a web3 product before shifting its focus to explore AI voice agents in June 2023. “The idea was that we’d simply put ChatGPT after a transcriber and before a synthesizer, and it would be valuable,” co-founder Aymeric Vaudelin (pictured above, first from left) told TechCrunch.
But the team soon faced the usual product-market fit conundrum. “After a few months, we realized that the market wasn’t ready yet to hear about voice agents. So we created a product, and packaged everything to create a first agent,” Vaudelin added.
That effort resulted in Donna, an AI voice agent for anesthetists. While that seems a bit random, the startup picked that market because anesthesia secretaries have to deal with a large number of patients, and it’s usually a very transactional experience.
In France, when you schedule a surgical operation, you have to talk to the anesthetist beforehand so they can make sure you don’t have any allergies to anesthetic products or any potential complications.
Anesthesia secretaries have to handle a large volume of calls that are pretty straightforward. Typically, people just want to know when the anesthetist is available, schedule an appointment, or change the date.
Moreover, these aren’t sales calls, so an AI agent doesn’t have to be persuasive or extremely efficient. “In the early days, we struggled with latencies of sometimes 4, 5, 6 seconds,” Vaudelin said.
Nevertheless, with Donna, Rounded managed to convince 15 private hospitals to let an AI voice agent answer calls, and the company says the agent has handled hundreds of thousands of conversations so far. Over time, Rounded improved its product so that it would be more accurate, better integrated with other products and, importantly, faster.
“With a web call, for example, we now get a latency of less than 700 milliseconds — more like 600 milliseconds. You add 200 milliseconds for phone connectivity, more or less,” Vaudelin said.
More recently, Rounded broadened its remit with an orchestration product that other companies can use to build their own voice agents.
Rounded lets you pick and choose your off-the-shelf AI models, which can be a speech-to-text model, an LLM and a text-to-speech model. For instance, for your first voice agent, you could use Azure to transcribe the call, GPT-4o mini as the LLM, and ElevenLabs as your speech-to-text engine.
The platform then helps define the instruction trees and prompts that will make the LLM work for your specific use case. “Implementing an agent means finding the right prompting, the right parameters and the right variables in the prompt,” Vaudelin said.
“Our pitch is that we’re going to enable everyone to create great prompts or great agents, and our product will support them in this iteration process to create super-reliable agents,” Vaudelin said.
Rounded has raised €600,000 so far (around $620,000) from UC Berkeley’s deeptech accelerator SkyDeck, and several business angels. But given that artificial intelligence remains an extremely buzzy industry, I’m sure the startup will raise more money in the coming months.