There’s a new player in the field of voice AI for restaurants, and it believes it has solved one of the technology’s biggest challenges: poor audio quality.
Incept AI was founded last year by veterans of Amazon and Presto Automation. It offers an AI-powered voice bot that can take restaurant orders in the drive-thru and over the phone. Since May, its system has been able to complete 97% of orders accurately, without human help.
That number helps explain why, though it is barely a year old and has just four employees, Incept has already landed a pilot with a 1,000-plus-unit restaurant chain and just raised $3 million to continue developing its technology.
“No one has really cracked the nut on figuring out accuracy without involving humans,” said Justin Foster, Incept’s co-founder and chief revenue officer, in an interview. But Incept believes it has done just that.
According to Foster, most AI order takers top out at about 83% accuracy before a human gets involved. That means about 2 out of every 10 orders get escalated to a person, either in the restaurant or in a call center. That level of human intervention is not ideal because it adds to the cost of the technology and makes it harder to scale.
One of the biggest things preventing AI bots from handling more orders on their own is bad audio, especially in the drive-thru, Foster said. The technology can struggle to parse multiple people talking in a vehicle, for instance. Wind, rain and traffic noise create interference. And the AI’s own voice can echo off of the waiting car, causing it to start hearing itself and get confused. All of this makes it harder to get customers’ orders right every time.
“It’s sort of a garbage in, garbage out situation,” Foster said.
That is where Incept believes it has an advantage. The company’s roots are in audio science. Co-founder Umut Isik was previously an applied scientist at Amazon Web Services, where he built software that addressed issues like background noise and echo. When he left, he began developing similar products for restaurants. Foster, who previously worked for AI voice provider Presto, said he was “blown away” by the accuracy Isik’s model was achieving.
While voice AI has made a lot of progress in recent years thanks to large language models like ChatGPT, Foster saw an opening in the restaurant market for a product that could solve the “last-mile” challenge of audio input.
“There are players on the field, but they’re all kind of scattered on the edges,” he said. “If you can drive unit economics down and drive accuracy up, that will open the field wide up.”
Incept launched in January 2024. The company is technically based in New York, but the team is spread out. Isik lives in New York and Foster is in Seattle. They bridge the distance by working side by side over video all day long.
“We wanted to have the intensity of a startup, where we’ll all in the same room together,” Foster said of the unusual setup.
Incept just completed a startup rite of passage with a $3 million pre-seed funding round. It was led by Rally Ventures, with participation from 10VC. Ben Fried, the former CIO of Google and a partner at Rally, is also joining Incept’s board of directors.
“We see a massive opportunity in Incept AI’s novel approach to solving voice AI’s biggest challenge—accuracy and reliability in real-world environments,” Fried said in a statement. “The team has a strong technical thesis and overall vision, and they’ve demonstrated impressive early traction.”
Incept plans to use the investment to hire more audio scientists to continue improving its technology. It also wants to bring on more customers. But it intends to move at a deliberate pace on that front.
“We really want to make sure that we’re going really deep with a few customers and doing a really good job rather than signing up 30 logos,” Foster said.
Members help make our journalism possible. Become a Restaurant Business member today and unlock exclusive benefits, including unlimited access to all of our content. Sign up here.