Return to site

Who Needs an AI Teacher? – with Liulishuo (流利说) Founder & CTO Ben Hu

· The Boss Interview,Cherubic Ventures,Liulishuo,AI,Education
broken image

Adam Bao interviews Ben Hu, Cofounder and CTO at Liulishuo, an AI-powered English learning solution with nearly 100 million registered users that IPOed in September 2018 (as LAIX). On Liulishuo’s platform, AI technologies are integrated with learning content, well-established language learning pedagogies, gamified features and social elements to deliver an interactive and adaptive learning experience. Ben shares his perspective on a number of topics, including: his journey from China to NYC/SF and back, what is Liulishuo and how does its AI work, the importance of integrating cutting-edge tech with a keen understanding of consumers and quality education content, how Liulishuo compares to other edtech giants in China, Liulishuo’s growth strategy, and more.

TL;DR 👇🏻

  • Liulishuo is an AI + education company, with its in-house AI built to emulate what a high quality teacher does: listen (ASR – automatic speech recognition), understand (NLP – natural language processing), think and offer feedback (via adaptive learning and TTS – text to speech)
  • Via its AI teacher, Liulishuo offers a consistent high quality education experience (since offline teacher quality varies greatly), makes this accessible to millions of students all across China (geographic limitations offline), and at reasonable cost (software solution variable cost goes to $0)
  • For a consumer company (especially edtech), the product design, experience, and educational content are just as, if not more important, than the technology behind the product
  • Language learning, especially English learning, is an enormous market
  • Building an in-house infrastructure for AI allows potential expansion into subjects beyond English learning
  • Liulishuo's global growth strategy will explore developing markets similar to China 5-10 years ago, with an appetite to learn English, like Southeast Asia, South Asia, and Latin America
  • Demand for Chinese language education remains a niche interest in the U.S

Edited by Tina He, Shaolong Lin;

Link to SoundCloud (here), also available via Apple Podcasts, Google Play, etc.

[Editor's note: this interview has been edited and condensed for clarity. The opinions expressed in this article are Ben's own and do not reflect the views of Liulishuo; Adam Bao is VP at Cherubic Ventures, the fund is a seed investor in Liulishuo]


Ben: I was born in Shanghai, and went to Shanghai Jiaotong University as an undergrad studying computer science. After graduation, I went to University of Arizona for the PhD program. One year into the program, I figured that I would not pursue the PhD program further and so quit the program and joined a startup in NYC, which was quite interesting. Prior to that, I only worked for big corporations. I interned at Microsoft Research Asia when I was in the college, and Bank of America when I was a student in Arizona. Right after college, I got interested in learning about how startups work, especially how they build stuff from scratch.

Before going back to China, I worked in 2 startups – the first one was in New York, and at the time I joined there were only seven people. The whole experience was enjoyable, and when I left the company, there were about 35 people from 15 countries. It was a very diverse environment, and everyone worked super hard. After New York City, I went to Quantcast in San Francisco. It's a big data-based advertising company — helping the marketers to optimize their advertisement especially digital ads. When I first arrived there, there were only 60-70 people. The company was around 4-5 years without any revenue stream, but rather they had a tool that helps the websites analyze traffic. By having that tool alone, the company was able to collect a huge trove of data. During my time at Quantcast, with the help of the business team and our technology team, we figured out a way to monetize the data we've accumulated for the past 4-5 years. When I left the company, it had about 150-160 people and started to show a steady revenue stream by optimizing digital ads for lots of large clients like BMW, Toyota, etc.

I went back to China in 2011. By that time, I'd worked for four years at two startups. When you’re in San Francisco, everyone talks about starting something new all the time. 2011 was a time when I saw China growing a lot even if I were in San Francisco. There were lots of news from China through social media showing all the startups in China. We were very excited about this trend, and we had an eager to start something new on our own – therefore a few friends and I decided to go back to China. It's such a big market, and as someone Chinese, I figured I might have a bigger shot there. A month later, I was back in China. It was quite a big surprise for my parents as they were not expecting me to go back to China after only five years in the States.

In 2011, I started a company doing a very geeky thing, kind of like a personal search engine, but the technology was very hard, and there weren't enough user needs. By early 2012 we decided to shut down the shop. Afterwards I took some time off, and after the summer and autumn of 2012, I started company Liulishuo (LAIX) together with other two co-founders.

Adam: What is Liulishuo?

Ben: Liulishuo is a world-leading AI + education company. At this moment we are focusing on language learning (mainly English). We developed a lot of the in-house AI technology to create an AI teacher to teach students English in a more efficient way. That's what we are offering to most of the users, at this moment we are focusing on English learning, our flagship app (英语流利说) in China has accumulated over 97m registered users. Over the past 30 months of our monetization period, we've accumulated more than 1m paid customers.

The technology we have in-house, more specifically AI-related technology, is quite natural. Think about what a human teacher can do. When he (or she) wants to teach a student, the first thing he will need to do is to be able to listen to his student. That's why we developed our in-house Automatic Speech Recognition system (ASR). At this moment we probably have the most extensive audio database for Chinese speaking English. With that amount of data, we have our ASR to recognize Chinese accent English, and are probably #1 in the world in that regard. After a human teacher knows what you are speaking, they need to understand what you are saying, and then comes the Natural Language Processing part, NLP or NLU (Natural Language Understanding). The ASR translates the audio data to text, and NLP/NLU understands what the text means.

After understanding what they are saying, the teacher needs to process the information and give students feedback – after he understands the students' performance, ability and progress, he should be able to recommend the best content for the students to learn. This involves a technology well-known in the online education world, called adaptive learning. The system is able to adaptively recommend suitable contents to the students depending on the students' current level. That's a dream for many educators, as it is personalized learning. With this adaptive learning technology, the learning path is 100% adaptive to you. The teacher then gives students feedback via TTS (test to speech).

Basically, we mimic the whole process of a human teacher: listen, understand, think and offer feedback. These are some of the technologies we have in-house related to the AI.

In the English learning theme, we observed that the user behaviors take time to change — therefore lots of different models co-exist at the same time. Unlike most area of the Internet, social media, social network or even bike sharing, it's very rare that several models will co-exist at the same time. But education is quite different. In China and even globally, language learning is an area where lots of companies are trying to tackle with different products. Here I only compare us with other online education products. Lots of online education platforms are still leveraging human instructors, with one-on-one education or even one-on-many education.

Adam: What are key differentiators between Liulishuo and other edtech solutions in China?

Ben: The key difference between Liulishuo and other education solutions is cost. Our solution is a purely software-based solution, the marginal cost of our content to an additional student is low (approximately 0), that's why we can offer an affordable price for our students. Second is about scalability. There are only a few teachers that are good – if comparing our solution to the top 1% of human teachers right now, we still have some work to do. However, the fact is not all teachers are the top 1%. When you have many students, you will need a massive supply of teachers, and it's very difficult to control the quality. You are not able to ensure the best quality, and that would have a negative impact on customer experience as well as customers' studying efficiency and effectiveness. Our solution is consistent. Once it's running on the cloud, it's quite predictable.

The 3rd point is efficiency and effectiveness. Even though lots of potential customers still have doubt, as human behavior towards education or other areas like healthcare, their behavior will change slowly. Both education and healthcare are difficult decision for a consumer to make. We are constantly conducting research to compare our solutions with other solutions, to see how efficient and effective our solution is vs. other solutions. We did a study in 2015, and another study with our much larger student pool in late 2018. The 2018 report was not officially released yet, but the initial data we saw was consistent with the report in 2015. Students using our software solution are able to learn English efficiently and effectively. That differentiates us from lots of other players.

Technology & Product

Adam: What does it take to build such a product? Is it driven more by the technology side, or the product side?

Ben: We are still in the hype of this wave of Artificial Intelligence. We are a little bit different from other AI companies, especially in China. Most of the AI companies B2B companies. They are providing technology solutions to business customers. We are directly providing our products to the end customers. For us, the core technology is essential, but for the end customers, they don't care about the technology behind it. A successful technology product needs more than the technology itself. Especially when it comes to education, the content is quite important. I mentioned a bit before – our company has ~6 years of history, but our monetization started 30 months ago. What happened to the other 30 months? The key event is that we realize in the late 2014 and early 2015, how important systematic curriculum is. And that's when we started to build our in-house curriculum team to build systematic contents within our system. At this moment for us, the content, technology, and product all have their corresponding shares in the success of our product. Like other consumer products, you need to know how to grow the userbase. At this moment, we build the technology and the contents, and we use technology to make the content better. Technology is definitely important, but as a consumer company, it's very rare for that company to win consumers beyond just by having good technology.

Adam: Given that the AI technology foundation you’ve built, with growing database and improving algorithms. Have you thought about expanding to other education markets? So let's say beyond learning English into different types of subject categories.

Ben: That is a great question. Honestly, there are already a lot of other companies doing that. So I'm not sure if it's good or bad. After our IPO, a lot of the folks when they pitch their idea, they would say “we are the Liulishuo in whichever subject."

So from our end, I think language learning, especially English learning, will continue to be a huge market for the next one or two years, and it will remain our focus. There are other subjects like math, physics or programming that we keep in mind when building our in-house infrastructure, like our content system or our in-house technology components. We try to make them generic enough so that in the future they apply to other areas.

In the short term, the company will continue to be a language company. English language learning is a big enough market, and I think we still have a lot to do to make our product more attractive or more useful for the users.

Chinese Market and Consumer Behavior

Adam: It's interesting how the English learning market is so big in China. I feel like most folks globally don't realize that, but in China, everyone wants to learn English. Back in the day, if you were able to speak English, your salary would be two times higher instantly. Even these days, despite China's rapid growth, the most popular international language is still English. So there's going to be a large demand for that. Especially with the one-child policy, everyone in the family from immediate parents to grandparents is supporting that one child. Therefore, so much money ends up going to education and English learning is a large part of that.

Ben: China is definitely one of the top spenders in language learning.

Adam: Let’s dig deeper on Chinese consumers' willingness to pay when it comes to subscription services in particular. I believe China is known for pirating content or software, but this seems to be changing. When I look at the universe of consumer products, there are products like Dedao (得到) or Ximalaya (喜马拉雅FM) etc. Chinese consumers are paying for content and paying for courses online. People in the US don't really pay for them, because podcasts are essentially all free; it's good for PR marketing and for general knowledge sharing. Could you tell us a little bit more about consumer subscription behavior in China?

Ben: I think consumer behavior in China changed a lot for the past several years. One key factor here is mobile payment. Over the past several years, mobile payments via the phone with WeChat and Alipay have been made available to many people via the "red pocket" feature, which allows the older generation to have money in their WeChat wallet or Alipay wallet and share money via the red packets. Even if they don't like to link their credit card or Debit card with their WeChat or Alipay account, they are still able to have money in the wallet through the red pocket. This basically opens up a new way that more people are able to pay for something on a mobile phone.

The macro trend plays a very important role as well. The government is very serious at cracking down a lot of the pirated content. They are promoting intellectual property aggressively by promoting original contents of text, audio, video, movies, and etc. Also in general, the population is getting richer with more dispensable income. So this macro trend and also the popularity of mobile payment play very important roles.

You mentioned the user groups specifically for podcasts. We see a lot of people living in big cities in China. So the time on traffic and especially taking subways and the commute makes the consumer behavior a bit different. They have much time to kill, and want to spend on those podcasts. I think that changed the dynamics a bit. There are still a lot of free podcast and free audio contents as well, even on Ximalaya. However, people are starting to be willing to pay for premium content. I think they would listen to free content as well, but they would pay for ad-free premium content, especially when they are only around 20 or 50 RMB per month. So why not?

Growth Strategy

Adam: Those are good points. It’s also interesting about paid content in China being high-quality. Kind of to your point about Liulishuo. It’s not just about AI, but also about understanding the user needs and having educators on a team. Dedao (得到) or Ximalaya (喜马拉雅FM), for example, have really big teams that help edit content that create classes. These classes are not just recordings of conversations between multiple people that are unplanned and unedited; they are very high quality. And there's a commission or revenue share that goes to Dedao (得到) obviously for that effort.

Let's talk a little bit more about your growth strategy. So you mention you're focusing on English learning as you see a huge market. You're considering a few other educational markets in China. What about expanding to other markets abroad? Are there any countries that you are looking at? How do you think about that?

Ben: This is a great question. I think for the product offering we are having right now in China, the main offering is general English ability product. I think it has a much larger audience globally as well. People from all over the world have very similar needs to improve their English, as it's now the language for international collaboration, international trade or international research. Everyone who wants to do more international trades and wants to make their life better wants to be able to benefit more from this globalization trend. We're looking into other developing countries similar to China five or ten years ago like countries in Southeast Asia, South Asia, particular India or Latin America as well. So for us, to better understand those markets and to pick similar markets to start is going to be our focus for the global expansion.

Adam: Perhaps we can cover some more of those details in a future conversation. I have a few more quick questions. We're talking about building an excellent English learning app for China and potentially for a global market. How about a Chinese learning app for the US market?

Ben: I think my answer is more of a personal one. My opinion doesn't represent that of the company. Because internally, we have a lot of debates and a lot of discussions on this topic from time to time. My personal opinion on this is that when you are a startup and talking to people in the states, especially in both coasts like west coast and east coast, you have an idea that people in the States really want to learn Chinese. My personal opinion is that it only represents the elites in both coasts. Because China is such an important player right now in the world in terms of trade and the business and everything. So it's very natural for them to want to learn Chinese to understand China better. However, I think for the majority of the population in the states, Chinese is not the language they want to learn. We also have data from other language app players who are doing pretty well in the states, and I think Spanish is probably the language for a lot of the Americans to learn.

Moreover, I also had a friend who started Chinese learning in the States like five or six years or even longer before. He told me terrible stories that they misjudged the whole user needs and consumer behavior. So my personal opinion right now is that it’s still too early. There’s a niche market in the states right now, but it's definitely not a huge market yet.

Adam: That makes sense. If it's these are elites who are trying to learn or have their kids learn it, then it's a pretty small population and they have high ability to pay. There's probably enough Chinese speaking teachers to fulfill their demand in person.

Ben: Yeah, so the value proposition with our approach to them may not be that high. Yeah, that's correct.

Impact of IPO

Adam: One last question for you, Ben. So you guys had your IPO at the end of 2018. I just wanted to know like how things have changed at Liulishuo since?

Ben: Honestly, it doesn't change much after IPO. First of all, it’s only almost four months after IPO. I think it's still an early stage for us as a public company, so the management team and even the whole company still don’t feel that much like a public company. When you just become a public company, it's not like you have been a public company for like 3, 5, or 10 years. We still work very hard, because I think IPO is just one of the milestones for us. It means that the company needs to run in a more mature way to face the public. So I think that all the employees understand that especially the middle and high level management teams. I understand that we still have a lot of work to do. Also, because we are still at the early stage and honestly, I think we haven't gotten the pressure as a public company. You probably understand what I mean. So for us, at this moment, it hasn't changed much. I think honestly we're going to feel more and more pressure as a public company. The good side is that as a public company, you get all of the benefits of the public company, but on the other hand, you get the pressures from the public investors. So I think it's going to take some time for us to feel that pressure. I think it's too early.

Adam: Got it. Well, it sounds like you guys still have the entrepreneurial fire and fight to you. It's exciting to hear that. Look forward to seeing what you guys build out further at Liulishuo, and we’ll be in touch.