The AI language tutoring app Speak, founded by former Thiel fellows, began its journey in Seoul, South Korea. However, when entering the US market, it would need to compete with better-known competitors resembling Duolingo.
Ona trip to Seoul in 2018, Connor Zwick drove past what looked like normal skyscrapers. But he soon discovered that there have been many classrooms teaching English, and there have been advertisements for such courses on taxi roofs and billboards. It became clear to him: the most important marketplace for his then-young language learning app Speak was here and never in Silicon Valley, where his headquarters were. “Learning English was like an obsession there,” he remembers. “There was such a latent demand.”
But the language training offered was often ineffective. Students spent hours learning the fundamentals through textbooks and pre-recorded videos, under the supervision of teachers who lacked language skills. The outdated methods focused on learning grammar and vocabulary fairly than actually speaking out loud. “Everything was so academic and there was this fear of making mistakes,” Zwick said.
For Zwick, Seoul’s bustling downtown and infinite demand for English lessons were each inspiration and opportunity – to develop a real AI teacher that may replace the human language teacher and supply learners with a judgment-free space to make and proper mistakes when speaking one other language. The promise: “No one will know if you say something embarrassing,” he said.
Today, Speak has a voice-based AI trainer that role-plays scenarios like ordering drinks at a restaurant, asking for directions to a tourist spot, or making small talk with a classmate to practice speaking a brand new language in real-world situations. Based totally on OpenAI’s models, the goal is to encourage users to practice saying commonly used words and phrases out loud. Learners may create their very own situations from scratch using AI software. Features like leaderboards and streaks that track each day usage help ensure consistency. And it isn’t only for English learners. Speak offers its role-playing games and lesson plans for five additional languages: Korean, Spanish, Japanese, French and Italian.
Today, around 15 million people have downloaded Speak to practice conversation with the interactive AI teacher and speak a foreign language more fluently and confidently. This momentum has helped the $1 billion startup reach a recent milestone: The company announced today that it has over $100 million in annual revenue. This is essentially because of the strength of its consumer business. Users use the app without cost and pay between $80 and $200 to access additional content.
But Speak also began operations in 2024 as a few of its consumers asked employers to cover the prices. Around 500 firms, including KPMG and HD Hyundai, now offer Speak subscriptions for workers, primarily in South Korea. And after gaining traction in other countries like Japan and Taiwan, Speak launched a concerted push into the US market in early June this 12 months.
This is impressive progress, but we still have an extended strategy to go to meet up with our biggest rival. Speak’s revenue is dwarfed by that of Duolingo, which had revenue of $724 million last 12 months Projects its revenue will reach $1.02 billion by the tip of this 12 months. The company can be betting big on AI and now offers occasional video calls for language learning with its AI nature. In doing so, Lily fulfills billionaire CEO Luis Von Ahn’s mission to develop an automatic tutor that may teach anyone a foreign language easily and inexpensively.
Zwick is keenly aware of Duolingo’s dominance, particularly within the US, but believes Speak is well differentiated. Duolingo is designed to enhance grammar and vocabulary, he argues. Speak’s focus is on conversation fluency, constructing linguistic muscle memory, and refining pronunciation so people feel comfortable speaking out loud. The name says all of it – Zwick claims that individuals speak five to 10 times more on Speak than on other language learning apps.
And Zwick doesn’t shrink back from how these differences impact people’s experiences with Duolingo in comparison with Speak. “Mobile games that also teach you language are ultimately a way to feel less bad about using your phone when you’re bored. It’s a healthier alternative, if you will, to Candy Crush,” he said Forbes. “And I think what we’re trying to do here … is something more serious.”
Speak has attracted attention from VC heavyweights resembling Khosla Ventures, Accel and OpenAI Startup Fund, raising a complete of around $160 million in capital. But in 2016, when Speak was founded, collecting fun wasn’t easy. Not many individuals were fascinated about language learning or AI, and a few investors even asked the founders if the app was a secret ploy to gather and sell data, Zwick said. But his insight into South Korea’s promise was prescient. Accel partner Ben Quazzo, who flew to Taipei for the startup’s user conference and led its $78 million Series C round last December, said Speak’s decision to deal with the South Korean market was intentional – a “sandbox” to check the product in one of the crucial competitive language learning markets. “We don’t sell ourselves like a dopamine hit or engagement bait app,” Quazzo said. “We sell an educational journey to lead you to an end result.”
Zwick met his co-founder Andrew Hsu through the Thiel Fellowship in 2012. He had just dropped out of Harvard for a so-called “deep learning sabbatical” and was constructing his first startup, a learning tool and mobile app called Flashcards+, which was later acquired by education technology company Chegg. In Hsu he found a kindred spirit, a person who dropped out of Stanford in 2011 to hitch the community. Fascinated by AI, the 2 secretly attended courses on the University of California, Berkeley and Stanford and deepened their understanding of key machine learning techniques resembling reinforcement learning. They even listened to talks by OpenAI co-founder John Schulman without ever getting caught.
Today, OpenAI is key to the corporate’s growth. As the AI models proceed to enhance, so does Speak’s app, which uses the models to mechanically update lesson plans based on user performance and do things like accent correction, something that wasn’t possible when Speak was founded. Sven Strohband, enterprise partner and managing director of Khosla, who was one in all the primary investors, remembers versions of Speak from early 2017 that might not yet perform full-fledged conversations with AI, but could appropriately understand foreign accents. “To be honest, they were AI-first before it was really cool to be AI-first,” Strohband said.
More from Forbes
