How the Chinese Are Learning English Using AI Tutors
Some of our American readers may recall an animated series that was once quite popular in ‘Murica called King of the Hill. One memorable skit from that series was when Hank Hill – best described as a slightly redneck, beer-swilling propane salesman – meets his Asian-looking neighbor and asks him “are you Chinese or Japanese?“, to which the neighbor responds that he’s from Laos, and then goes on to describe his country in great detail. After this rather confusing diatribe about “being Laotian”, Hank pauses, then asks him again “so, are you Chinese or Japanese?” It’s a great exposé into the typical knowledge base most Americans have of “Asians”, as they try and ball up half the world’s population into a single label.
While today’s hip American youth may be more in tune with Asian cultures outside of “the big two”, most of America seems rather oblivious to the fact that the Chinese (not the Japanese) are seriously becoming contenders to win the global technology race. We’re now finding that almost every week, there’s some interesting Chinese technology company that’s emerged as a “global leader” in something America should be leading in. Just last week, we talked about how Chinese Electric Vehicle Maker NIO Filed an IPO, and a few days ago we talked about A $100 Million Chinese Artificial Intelligence IPO. Today, we’re going to talk about another Chinese AI startup that just filed to raise another $100 million from the U.S. markets – and maybe much more than that.
Liulishuo – Means “Speaking Fluently” in Mandarin
Founded just five years ago, Chinese startup LAIX Inc. (also called Liulishuo) had raised almost $129 million for their AI-powered English tutor that teaches Chinese people how to speak English. Now, the company is looking to raise $100 million in an Initial Public Offering (IPO) that some say may even raise more, up to $300 million more, if the demand is there. Given that Liulishuo is being represented by both Morgan Stanley and Goldman Sachs, it’s safe to assume that their road show is pretty solid. (Notice how these last three Chinese IPOs have the biggest names in global banking running their IPOs.)
We once postulated that artificial intelligence will have reached maturity when it can translate Chinese into English without creating Engrish. (We also call B.S. on any startup claiming to have mastered this problem if the English version of their website still contains Engrish.) It seems like AI is at least ready to begin teaching people how to speak English, because Liulishuo has been using it to teach English to over 83 million Chinese people:
The average amount of time per day that these users spend on the app is over an hour. That’s because Liulishuo figured out the ultimate product feature that no Asian person can resist – gamification combined with elements of social media – and then embedded it into their app. They also started to provide enterprise learning services last year, and have already landed more than 100 corporate clients.
The Liulishuo Business Model
The Liulishuo app is offered under a “freemium” business model, which means you attract users with free services and then convert them into paying users. With 83.3 million registered users and just 1 million paying users, how do we know if the business model is working as expected? According to an article by TechCrunch back in 2012 – a time when they were at least trying to be apolitical – the following useful metric was offered up:
Most freemium companies have customer conversion rates of between 1 percent and 10 percent (with the average around 2 percent to 4 percent).
Liulishuo has a 1.2% conversion rate which is below average, but maybe that’s not where the real value lies. Time and time again, we see companies offering up a subsidized product or service that’s really about gathering the big data. We talked before about a similar situation in the translation industry. Translation companies – like Korean startup Flitto – are barely making money off the translation tasks they perform (it’s become a commodity after all), but in fact are making money by selling collected language data called “corpus” to companies like Baidu, Microsoft, and Tencent, who use this data to train their own algorithms. That could be the real value on offer from Liulishuo, and here’s why.
Liulishuo’s Big Data Advantage
Liulishuo claims to have the world’s largest database of English spoken by Chinese, covering a broad range of geographic distribution and proficiency levels. As of mid-2018, the company had recorded approximately 1.3 billion minutes of conversation and 17.5 billion sentences. To put that into perspective, it would take you 2,473 years of constant listening to get through all that banter. If an average sentence contains 15 words, then that means in order to read all the sentences that Liulishuo has gathered on their platform, it would take you about the same time as it would to read War and Peace – 446,970 times:
The point is, all that big data is being used to help train their AI algorithms to teach Chinese people English faster and faster. How much faster? Well, some of that is up to you. The more time you spend studying on the platform, the better it gets at teaching you. The end result is something like a 3X learning efficiency over human teachers who are now forced to spend the better part of their traveling years “begpacking” instead of teaching English in China:
When it comes to who uses the app, more than 92% of users originate from China. That’s no surprise. Neither is the fact that females comprise the majority of their user base – when you consider that more than 60% of self-made billionaires are Chinese women – and you can’t run a billion dollar startup these days without knowing a bit of the Queen’s English. Around 14% of their users are college students, 31% are students in K-12 education, and another 38% are employed or working as freelancers.
For many of these business models that focus on growing a large user base rapidly, it’s the growth metrics that interest prospective investors the most. Liulishuo is no different. By all measures, their user growth is accelerating rapidly, and so are revenues. While we can’t be asked to translate these numbers into U.S. dollars, here are the monthly metrics showing the rapid growth of revenues:
Nothing too surprising here. The strong revenue growth represents paying user growth, which is a proxy for overall user growth as mentioned earlier. The increasing “cost of revenues” is a result of the more than 1,600 employees that now work for the firm. The equally large R&D spending number reflects the +50 research scientists building the deep learning algorithms that power the platform – led of course by a former Google AI research scientist. Finally, we see heavy sales and marketing expenditures as the company tries to grow their user base as fast as possible in the face of a rapidly growing Chinese English learning market that’s only expected to grow faster in the years to come:
Given the growth of China on the global stage, it may not be a bad idea to think about building a startup that teaches Americans Mandarin. (Based on some of the emails we get, we may want to start by teaching them English first.) While the Americans are still having problems deciphering whether or not someone is Chinese or Japanese, the Chinese are using one of the world’s most promising technologies to teach themselves English. They’re descending upon America in droves, wandering around in large tour groups looking confused, while snapping pictures of nearly everything, and buying up properties before they head home to bore their families to death with all their pictures. They’re also using our capital markets to raise money while we flounder around with our ICO “utility tokens” like a bunch of monkeys trading shiny rocks with each other. Maybe it’s time we Americans became a bit more familiar with the Chinese – because they’re becoming awful familiar with us.