China's AI Adaptive Learning Industry Whitepaper - March 2021 - EY
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
China’s AI Adaptive
Learning Industry
Whitepaper
March 2021
China’s AI Adaptive Learning Industry Whitepaper I 1
Content
Abstract 04
I OVERVIEW OF CHINA’S AI ADAPTIVE 05
LEARNING INDUSTRY
1.1 The Development of China’s AI Adaptive Learning
Industry
1.2 Market Size of China’s K-12 AI Adaptive Learning
Industry
1.3 Competitive Landscape of China’s K-12 Offline AI
Adaptive Learning Industry
1.4 Key Success Factors of China’s K-12 AI Adaptive
Learning
II OVERVIEW OF CHINA’S OMO AI ADAPTIVE 28
LEARNING
2.1 Background of OMO AI Adaptive Learning
2.2 Business Models for OMO AI Adaptive Learning
2.3 Exploration Direction of China’s OMO AI Adaptive
Learning
III CHINA’S AI ADAPTIVE LEARNING MARKET 40
OUTLOOK
3.1 Development trends of China’s AI Adaptive Learning
Industry
3.2 Future of China’s AI Adaptive Learning Industry
China’s AI Adaptive Learning Industry Whitepaper I 2
China’s AI Adaptive Learning Industry Whitepaper I 3
Abstract
With the breakthrough from computation This white paper will begin with the analysis of
power and algorithm capability, Artificial the industry status-quo, a definition of the
Intelligence (“AI”) has been evolving quickly in intelligent level, an estimation of the market
simulating human intelligence and penetrating size of China's K-12 AI Adaptive Learning
multiple aspects of our daily life. “AI Adaptive industry and a summary of key success factors.
Learning” is a major achievement after AI is Furthermore, Online-Merge-Offline (“OMO”)
applied to the education industry, which truly business model will be discussed with its
helps to implement the teaching philosophy of background, form and future direction. In the
"teaching in accordance with one's aptitude". end, we will raise our preliminary thoughts
about how AI adaptive learning will evolve
In recent years, China has been encouraging
from social environment and competition
the implementation of digital and intelligent
landscape perspectives.
transformation in the field of education. In
2017, the State Council issued the Key insights and findings of this white paper:
“Development Plan for the New Generation of
► AI Adaptive Learning refers to the
Artificial Intelligence”, which defined the
educational products in which artificial
pivotal position of AI in education. Shanghai is
intelligence can take charge of the teaching
one step ahead of the others by issuing the
activity to provide students personalized
“Shanghai Education Informatization 2.0
learning experiences.
Action Plan (2018-2022)" to clarify the future
direction of intelligent education, in which "AI ► We estimate the China K-12 AI Adaptive
Adaptive Learning" was mentioned for the first Learning market will reach around USD 40
time. billion (RMB 270 billion) by 2025. The low-
tier cities’ offline education channel will be
The market potential of AI Adaptive Learning is
the main sub-segment of penetration.
growing rapidly with diversified service
offerings. Meanwhile, due to the high technical ► The market competition landscape is highly
barriers and changing application scenarios, its concentrated for now, with more than half
market definition and user recognition remain of the market share in the hands of a few
unclear. Under such situations, EY-Parthenon top players. However, it is also expected
publishes this white paper with an analysis on that more tech companies, traditional
the industry status-quo and future trends. We education institutions, and online education
hope this white paper could help market platforms will announce their entry into the
participants and users get a deeper market in the future.
understanding of AI Adaptive Learning and
► The key success factors of AI Adaptive
move the industry forward together.
Learning enterprises are AI technology
optimization, content development, and
user data accumulation.
► OMO AI Adaptive Learning is a new
exploration direction that can promote deep
integration of multi-scene teaching and
solve the shortcomings of traditional online
and offline education.
► In the future, China's AI Adaptive Learning
industry will show four trends: More
traditional entrants, more comprehensive
subject coverage, more education stage
coverage and more approach oriented.
China’s AI Adaptive Learning Industry Whitepaper I 4
01
Overview of China’s
AI Adaptive
Learning Industry
China’s AI Adaptive Learning Industry Whitepaper I 5
1.1 The Development of China’s AI
Adaptive Learning Industry
In recent years, artificial intelligence (AI), big 1.1.1 Development Background
data analysis, cloud computing and other
The idea of “AI Adaptive Learning” is derived
emerging technologies gradually penetrated
from “Adaptive Education,” both of which
into all walks of life, driving the transformation
adhere to the principle of teaching students in
of our society from the “Internet era” to a new
accordance with their aptitudes. Adaptive
“intelligent era.” Meanwhile, AI-empowered
Education came into the light when
application has been put in practice in all
programmed learning theory and the teaching
aspects of the education industry and evolved
machine were introduced in the 1920s. With
from single-scene and single-point teaching
the development of information technology and
tools to educational products that integrate
the popularizing application of AI in the past
multi-scenes and cover all links, and eventually
years, AI Adaptive Learning had formed when
realizing a personalized and innovative
AI was adopted in the core links in teaching.
teaching experience.
Adaptive educational products have gone
Under this context, the concept of “AI Adaptive
through 4 phases: Traditional Teaching,
Learning” came into being. AI Adaptive
Efficiency Improvement, Decision-aid and AI
Learning refers to the educational products
Adaptive Learning, during which the level of
that are based on the use of intelligent
intelligence has been progressively improved,
technologies such as AI, big data analysis, and
and more teaching links have been covered.
IoT, combined with a large amount of user data,
to provide a unique and personalized learning
experience for students in accordance with
their aptitudes. By simulating the process of
one-on-one human teaching, AI Adaptive
Learning alleviates the shortcomings of
traditional education, such as the inefficiencies
in large classes and the high cost of one-on-one
teaching.
2000 2012 2016
High
Traditional Teaching Efficiency Improvement Decision-aiding AI Adaptive Learning
Stage Stage Stage Stage
Large class Dual-teacher class Adaptive test AI adaptive learning
Intelligent Level
Small class Online broadcast question database AI teacher
One-on-one class Intelligent Expression …
correction recognition
…
… AI teaching
assistant
…
Low
Peripheral Core
Covered Teaching Processes
Sources: Deck research, EY-Parthenon analysis
China’s AI Adaptive Learning Industry Whitepaper I 6
1.1.2 Intelligence Levels of Adaptive ► L4 - Advanced AI Tutoring (AA): AI is
Education adopted in the teaching link, and it leads the
According to the degree of intelligence, whole education process. Students receive
adaptive education can be ranked into six levels education from AI teachers on terminal
(L0-L5). Only when artificial intelligence enters devices, while real teachers are responsible
the core "teaching" link (L3-L5) can it be called for final checks and correction. These
“AI Adaptive Learning". products are typically named as AI Adaptive
Learning Platform or AI Teacher.
► L0 - Traditional Teaching (TT): Traditional Representative companies include IBM
real-person teaching in all links without Watson, BYJU’S, Realizeit abroad, and
automation tools. Squirrel Ai in China (some product lines
► L1 - Internet Teaching (IT): Human-led include real teachers).
teaching in all links, with the addition of ► L5 - Full AI Teaching (FA): It is the ultimate
digital systems such as online testing, form of AI Adaptive Learning. At this level,
remote platform, etc. to improve teaching AI can fully simulate the teaching process of
efficiency. Representative forms include excellent human teachers without
online live classes, MOOCs, etc. intervention. The current international
► L2 - AI Tools (AT): AI has been applied in representative is Korbit, an online AI
some non-teaching links to further improve Adaptive learning platform founded in 2018.
the efficiency, such as photo search, voice It was born out of the artificial intelligence
testing etc., but the teaching is still led by lab led by Yoshua Bengio, a Turing Award
real teachers. Representative products winner and renowned AI professor. This
include Zuoyebang, Yuanfudao etc. platform can automatically teach courses
including data science, machine learning and
► L3 - Partial AI Teaching (PA): AI was artificial intelligence. Squirrel Ai, a leading
adopted in the teaching process to provide enterprise in the industry, has also achieved
systematic analysis and recommendation to the L5 level with some product applications
assist teachers in decision-making, but the (such as AI Foundation Courses and free
teaching process was still dominated by courses partnered with Ding-Talk) with the
human teachers. Companies usually offer a wide application and fruitful data retrieved
complete set of teaching solution, typically from China K-12 education adoption.
named AI-assisted platform or AI assistant,
achieve full-process teaching data access It is worth noting that though, in theory, AI can
and diagnosis before-, during- and after- simulate or even surpass excellent teachers,
class. Representative companies include and be applied to all aspects of education, we
ALEKS, Knewton, Coursera, Khan Academy believe that under current AI’s technical level,
abroad, TAL and iFLYTEK in China. it cannot replace the role of real teachers in
supervision, communication, motivation and
personality training. Therefore, it might be an
ideal form of education to have AI responsible
for teaching and real teachers for cultivating.
Looking beyond, with the technological
breakthrough, it is not ruled out that there will
be a higher level of intelligence that can
independently complete all the work of
teaching and cultivating.
China’s AI Adaptive Learning Industry Whitepaper I 7
Teach Educate
Definition Diagnose Plan Teach Practice Feedback Cultivate Example
Low
Traditional
L0-TT Traditional Teaching
AI assists teaching
offline education
1-on-1, livestream,
L1-IT Internet Teaching
MOOC, recorded
Intelligent Level
Zuoyebang
L2-AT AI Tools
Yuanfudao
L3-PA Partial AI Teaching Knewton
AI leads teaching
Iflytek
IBM Watson
L4-AA Advanced AI Teaching Squirrel AI
Korbit
L5-FA Full AI Teaching Squirrel AI
AI Adaptive Learning High
Human teacher Internet tool AI
Sources: Deck research, Interviews, EY-Parthenon analysis
1.1.3 Overseas AI Adaptive Learning Market United States: Pioneer in AI Adaptive Learning
Overview with mature technical background, excellent
content, and clear IP rights. However, due to
With the rise of big data, AI and other
the multi-race, diverse student background,
technologies, AI Adaptive Learning has moved
academic performance is not the greatest
from theory to practice. In the early 2000s, the
indicator of education, it is difficult to realize
first AI adaptive learning product was born in
large-scale To-C promotion of AI Adaptive
the United States. Companies such as Knewton,
Learning products. At present, the AI adaptive
Realizeit have come into the public’s attention,
learning products in the United States are
which marked the start of AI Adaptive Learning
mainly focused on K-3 preschool education and
in commercial uses. The Indian market began
higher education. However, due to inability to
to emerge in 2010, with a few start-ups came
realize large-scale monetarization, a number of
to light. At present, due to the variance in
market players have been acquired by large
teaching content, technical level, and cultural
publishers. A representative of such is Knewton.
background in different countries, market
development diverges as well.
China’s AI Adaptive Learning Industry Whitepaper I 8
Europe: started later than the United States India: the last to enter the game, but due to
but has a similar level of the research base. India’s large population, the popularity of
Whereas, due to the complex language Internet, unified K-12 content and urgent
background in Europe, To-C business is also demand for good grades, the development of
hard to achieve on a large scale. Limited by the AI adaptive learning products in the Indian
narrow domestic market, many European market has achieved large-scale promotion,
market players have entered the U.S. market primarily towards K-12 students. At present,
by cooperating with American universities, India is the most similar AI Adaptive Learning
among which a representative is Realizeit, an market to China, among which the classic
AI Adaptive learning platform enabler- representative is Byju's.
Case Study of BYJU'S
1. Background Based on students’ proficiency and aptitudes,
BYJU’S produces 5-minute videos for each
As a leading company, BYJU’S was founded in
knowledge point and reorder them to provide
2011 aiming to provide students with an online
customized learning paths. In terms of business
learning application covering math, physics,
model, BYJU’S provides free courses at the
chemistry and biology as well as standardized
beginning, then charges afterwards. This way,
test such as JEE, NEET, CAT, IAS, GRE, and
it can cultivate users’ behavior and bring in
GMAT. BYJU’S main products are Early Learn
large amount of registered users.
App for G1-G3 students and The Learning App
for G4 – G12 students. 3. Accomplishments
2. Features To date, BYJU’S has accumulated 75 million
registered users, covering 1,701 cities, and is
As one of the most popular learning APP in
one of the largest online education company in
India, BYJU’S major target user is individual
India. It has extended its partnership worldwide.
students, dedicated to promoting students’
BYJU’S is currently valued at USD 10.8 billion,
autonomous learning in their spare time. The
with 19 financing projects completed,
product’s most prominent feature is its
accumulating to a total of USD 2,688 million, of
exquisite videos.
which are many well-known PE/VCs such as
Sequoia Capital, Tencent, etc.
$9M $25M $75M $65M $30M $40M $100M $2.24B
Series A Series B Series C Series D Series E Series F Series G Pre-IPO
Sources: Tianyancha.com, BYJU’S Website, Desk Research, Interviews, EY-Parthenon analysis
China’s AI Adaptive Learning Industry Whitepaper I 9
1.1.4 Major Application in China ► China’s K-12 education is closely tied to
Gaokao (the college-entrance exam), which
Due to the different educational cultures
leads to a higher willingness to pay for K-12
between China and the U.S., the areas of
education from Chinese parents. EY-
application are quite different even though
Parthenon estimates the size of China’s
China’s AI adaptive learning industry bought
overall education market has been growing
lots of experience from US in the early stage. In
steadily at a CAGR of 22% in recent years.
China, the major application area focuses on
Meanwhile, K-12 education outgrows at a
the K-12 education, mostly due to two reasons:
CAGR of 30% among other sub-segments.
► Compared with the other education Long-term speaking, although the overall
segments, K-12 education in China is more education market shrunk in 2020 due to the
exam-oriented and knowledge point driven. COVID-19 outbreak, we remain optimistic
As a result, students’ learning effects are that the market will soon recover, with K-12
quantifiable, making K-12 education the best education still being the fastest growing.
landing point for initial application. While
education in US focuses more on the
thinking process than pure knowledge input,
leaving little room for the application of AI
Adaptive Learning.
China's Education Market Size
Unit: Billion USD
Pre-school K-12 Vocational Higher Other
CAGR CAGR
17-19 20E-25E
445 22% 18%
375 18% 15% 21%
316 18%
257 267 17%
51%
15% 225 17% 30% 22%
209
191 17% 49%
172 16% 48%
16%
17% 48% 47%
46% 45% 11% 15%
42% 44% 13%
13% 13%
13% 14%
14%
15% 13% 15% 12% 18% 10%
15% 15% 14% 13%
16% 9% 16% 17% 16%
9% 9% 8% 8% 8% 7% 7% 7% 22% 15%
2017 2018 2019 2020E 2021E 2022E 2023E 2024E 2025E
Sources: Deck research, Interviews, EY-Parthenon analysis
Note: The figures are converted from RMB into USD according to the exchange rate of RMB against USD (647:100) on February 18, 2021
China’s AI Adaptive Learning Industry Whitepaper I 101.2 Market Size of China’s K-12 AI
Adaptive Learning Industry
As mentioned above, China's K-12 AI Adaptive Compared to the offline education market, the
Learning industry is undergoing a period of online market is a relatively new and small
rapid development. Many existing players and market, derived from the digital transformation
new entrants are sparing no effort to conduct of K-12 education in recent years. It is
quick entry and business monetization in this estimated that the size of China’s K-12 online
market. education market is expected to reach 63
billion by 2025 at a CAGR of 35%, with the
1.2.1 Market Size of China’s K-12 Education
following key drivers:
China’s K-12 education market can be divided
► Demand side: the problem of unequal
into online and offline channels. The offline
distribution of educational resources in
market is formed by traditional education
different regions has long existed. Online
institutions, offering face-to-face teaching. The
education products have effectively solved
online market is an emerging market generated
the pain point of the lack of high-quality
by moving part or full of the teaching activity
educational resources in remote and rural
online in recent years.
regions.
The offline education market is the main
► Supply side: the launch of many online
battlefield of K-12 education institutions
educational products has transformed the
traditionally. We estimate the offline education
traditional teaching scenarios, breaking
market has reached approximately USD 115
through the limitations of time and space in
billion in 2019, attributable to the following
offline education and bringing new
driving factors:
opportunities.
► Demand side: Chinese households’
We believe the offline market will still be the
disposable income continues to rise, and the
main battlefield of K-12 education in the long
consumption of educational products is also
run, which can be referred to from what
growing, driven by traditional channel
happened in the retail industry. The online
preference.
retail business has shown explosive growth in
► Supply side: Education institutions began to the past decades. However, according to the
focus on low-tier cities, the potential National Bureau of Statistics, its proportion in
demand for high-quality educational the total sales of consumer goods only
recourses in the sinking markets is further accounts for 25% of consumption made offline.
released. The main reason is that online retail cannot
make up for the same consumer experience
offline. Similarly, in the education sector,
customers will pay more attention to the
teaching experience when choosing the best
education product. Offline teaching tends to be
more effective by offering direct interaction,
supervision and motivation, thus offline
education will remain the first choice for
parents and students.
China’s AI Adaptive Learning Industry Whitepaper I 11China’s K-12 Education Market Size
Unit: Billion USD
224 CAGR CAGR
17-19 20E-25E
Online Education Offline Education
184 63 30% 22%
150 48
122 124 35 73% 35%
6
96 5 102 26
84 20
72 2 161
14 136
116 115
91 98
70 82
70 29% 18%
2017 2018 2019 2020E 2021E 2022E 2023E 2024E 2025E
Sources: Deck research, Interviews, EY-Parthenon analysis
Note: The figures are converted from RMB into USD according to the exchange rate of RMB against USD (647:100) on February 18, 2021
China’s AI Adaptive Learning Industry Whitepaper I 12China’s Offline K-12 Education Market Size
Unit: Billion USD
161 CAGR CAGR
17-19 20E-25E
1st&2nd Tier Cities Low-tier Cities 136
40
29% 18%
118 114
34
98
92 33 29 20% 16%
82
71 25
28 70
22 121
23 19
102
85 85
64 73
51 60
48 18%
33% 19%
2017 2018 2019 2020E 2021E 2022E 2023E 2024E 2025E
Sources: Deck research, Interviews, EY-Parthenon analysis
Note: The figures are converted from RMB into USD according to the exchange rate of RMB against USD (647:100) on February 18, 2021
Taking a closer look of the K-12 offline ► Supply side: EY-Parthenon calculates that
education market from the city levels, it is the current share of large education
found that more than 70% of the market is institutions in the lower-tier market is only
concentrated in the low-tier cities, reaching a around 2%. Facing market saturation in first-
scale of USD 1.5 billion and a CAGR higher than and second-tier cities, more and more of
that of first- and second-tier market with the these players begin to explore into the low-
following key drivers: tier cities to maintain their growth
momentum, which will bring a new round of
► Demand side: according to the National
upgrade.
Bureau of Statistics, there are 180 million
primary and secondary school students in Greater market potential of China K-12 offline
China, of which more than 70% come from education can be found in low-tier cities, which
low-tier cities. The huge student base lays implies that how to compete for these students
the foundation for the market prospects. and lay out an efficient route will be the key
strategic direction of market players in the
near future.
China’s AI Adaptive Learning Industry Whitepaper I 131.2.2 Market Size of China’s K-12 AI Adaptive ► Technology: the development of hardware
Learning and software had improved the application
effect of AI Adaptive Learning. In recent
EY-Parthenon calculates the market size of
years, the performance of computing chips
China’s K-12 AI Adaptive Learning industry
has greatly improved on the cost-
which has reached USD 4.6 billion in 2019 and
effectiveness, which provides the foundation
is expected to reach USD 40 billion by 2025
for wide application of AI Adaptive Learning.
at a CAGR of 62%. Our projection is based on
With the advancement of AI, deep learning
the following factors:
and algorithm model, AI Adaptive Learning
► Policy: National and local policies have had moved from a theoretical level to
continually defined the strategic positioning practical application. Besides, the high-
of AI Adaptive Learning. At the national quality labelled data gradually accumulated
level, the State Council issued the “New during the interaction between user and
Generation Artificial Intelligence application process, which further improves
Development Plan” in June 2017, proposing the algorithm accuracy and enables the
a new education system that includes product effect to be proved.
intelligent learning and interactive learning
► Supply: the intensified competition in the
and promoted the application of AI in the full
education industry has enhanced the cost-
process of teaching, management and
reducing and efficiency improving features
resources construction, bringing AI
of AI Adaptive Learning. In recent years,
education to a national strategic level for the
with the entry of more online players, the
first time. In April 2018, the Ministry of
traditional education industry has been
Education issued the “ Action Plan for
abundant with more novel product forms
Educational Informatization 2.0 ” to
and business models, leading to a more
vigorously promote intelligent education,
competitive market. AI Adaptive Learning
carry out the construction of intelligent
technology, aiming at replacing human
teaching environment, and promote the
teachers, can be a way to reducing cost and
application of artificial intelligence in
improving efficiency for institutions, and
education. On a local scale, in September
additionally, providing differentiated
2018, the Shanghai Education Commission
features in terms of customer acquisition
issued “Shanghai Education Informatization
and teaching.
2.0 Action Plan (2018-2022)”, which
suggested “to explore new education form
based on Knowledge Graph and AI adaptive
technology, and to create unified-standards
shanghai solution to crowdfund and
crowdsource academic resources. This plan
explicated the explorative direction of AI
Adaptive Learning.
China’s AI Adaptive Learning Industry Whitepaper I 14China’s AI Adaptive Learning Market Size
Unit: Billion USD Online Education Offline Education
Penetration of Intelligent adaptive education:
4% 4% 4% 5% 6% 8% 11% 15% 19%
42.1 CAGR CAGR
17-19 20E-25E
14%
31% 62%
27.6
13%
97% 55%
17.1 86%
10.6 14%
6.5 15% 87%
3.8 15%
4.5 86%
4.3 17%
2.6 3% 4% 85%
2% 97% 96% 85% 30% 63%
98% 83%
2017 2018 2019 2020E 2021E 2022E 2023E 2024E 2025E
Sources: Deck research, Interviews, EY-Parthenon analysis
Note: The figures are converted from RMB into USD according to the exchange rate of RMB against USD (647:100) on February 18, 2021
We foresee the penetration rate of AI Adaptive base of K-12 offline institutions in low-tier
Learning products in China’s K-12 education cities and their high demand for AI adaptive
market will increase from 4% to 19% in the learning solutions. On the other hand, the high
next five years with a faster penetration rate in customer acquisition cost in the online
the offline segment compared to online. By channels makes fewer players put the online
2025, the offline market is expected to channel as their major focus.
account for 86% of China’s K-12 AI Adaptive
Learning market, mainly because of the huge
China’s AI Adaptive Learning Industry Whitepaper I 151.2.3 Penetration Trends of AI Adaptive Based on China’s K-12 population, EY-
Learning in K-12 Education Parthenon first estimates the quantity of K-12
education institutions in China, considering the
In order to analyze the route of AI Adaptive
adoption rate, annual tuition fee and revenue
Learning’s penetration into the offline channel,
scale of K-12 education institutions in first- and
we pursued further analysis of the penetration
second-tier cities. It is estimated that there are
rate in the offline segment by different types of
around 1 million K-12 education institutions in
offline institutions.
China: 500,000 to 800,000 institutions with an
annual revenue less than USD 30 thousand,
followed by 50,000 to 100,000 small
institutions with an annual revenue between
USD 30 thousand - 3 million and 700 -1,400
medium-sized institutions with annual revenue
between USD 3 - 75 million. There are only 20
large-scale institutions with annual revenue
above USD 75 million in China, accounting for
about 0.1% of the market.
Method of Calculating the Number of K-12 Education Institutions
Core logic Segmentation factors
Number of K-12 education institutions
K-12 population size City level: 1st & 2nd tier/lower-tier
Penetration rate of K-12 education City level：1st & 2nd tier/lower-tier
institution Institution type：Large/Medium/Small/Micro
Per customer transaction of K-12 City level： 1st & 2nd tier/lower-tier
education institution Institution type ：Large/Medium/Small/Micro
Revenue scale of K-12 education
institution Institution type ：Large/Medium/Small/Micro
Sources: Deck research, Interviews, EY-Parthenon analysis
China’s AI Adaptive Learning Industry Whitepaper I 16Number and Revenue Scale of China’s K-12 Education Institutions
Number and portion Total Rev. scale and portion
Large institutions Unit: Billion USD
(Annual rev. > $78M) ~20 ~4.8（~7%）
Mid-sized institutions
(Annual rev. $3M - $78M) 700-1400 ~4.5（~6%）
~70
Small institutions ~22
(Annual rev $300k - $3M) 50-100K （~31%）
Micro-scale ~39
institutions 500-800K （~55%）
(Annual rev. < $300k)
Sources: Deck research, Interviews, EY-Parthenon analysis
Note:
1) The number of institutions is calculated by brand, that is, if there are multiple branches under the brand, it is calculated by one.
2) The figures are converted from RMB into USD according to the exchange rate of RMB against USD (647:100) on February 18, 2021
Penetration of AI adaptive learning in China’s K-12 Offline Education Institutions
Large Mid-sized Small Micro Penetration of AI adaptive learning products Unit：Billion USD
1ST&2nd tier Lower-tier 1ST&2nd tier Lower-tier
2020E 2025E
Sources: Deck research, Interviews, EY-Parthenon analysis
China’s AI Adaptive Learning Industry Whitepaper I 17Looking into the penetration of AI Adaptive 20%. As a result, traditional classroom tutoring
Learning products of different city tiers, EY- cannot meet the needs of every student, but it
Parthenon found that low-tier cities will be the is exactly what AI adaptive learning is capable
focus of penetration in the future due to the of.
following reasons:
Small and mid-sized education institutions are
► The lack of high-quality educational the main users of AI Adaptive Learning
resources in low-tier cities can be products currently, and we predict it will
effectively alleviated through the gradually penetrate mid and large-sized
introduction of AI Adaptive Learning institutions in the future:
products. The greatest dilemma faced by
► Small and mid-sized institutions are
education institutions in low-tier cities is the
generally difficult to recruit excellent
lack of excellent teachers, which could be
teachers and develop curriculum systems,
replaced by AI Adaptive Learning products
which makes them eager to cooperate with
to solve the pain points.
AI Adaptive Learning platforms.
► The education expenditure level in low-tier Traditionally, small and mid-sized institutions
cities is limited, and AI Adaptive Learning in low-tier cities are more likely to face the
products with higher cost-effectiveness are problem of teacher shortage and weak
more likely to be welcomed. The operating curriculum content. Through cooperating
costs of AI Adaptive Learning products are with upstream suppliers and apply adaptive
much lower than one-on-one or small-class content into their teaching, these
tutoring no matter it is conducted online or institutions can build their own competitive
offline, which makes AI Adaptive Learning advantages to realize brand upgrades.
products cheaper and more appealing to the
► Traditional large-size education institutions
low-tier demographic.
are more cautious in adopting AI adaptive
► As customers in low-tier cities need more learning, but it will be part of their business
time to accept new technologies, higher portfolio in the end. Though the giants of
potential will be inspired along with the traditional education institutions have
penetration in the future. At present, deployed large R&D resources into AI
customers from tier 1 and 2 cities have Adaptive Learning related products, most of
better access and willingness of adoption to them have not yet been mainstream
new technologies. However, with more business. There are 2 major reasons behind:
solutions implemented in lower-tier cities, AI 1) Large-sized institutions are mostly
Adaptive Learning businesses can bring located in first-tier and second-tier cities and
confidence to the customers there and it has good teacher resources; 2) Most large
increase their adoption rate in the long run. institutions are directly operated with heavy
asset, which is unlikely to have a rapid
► Capability gaps among students in lower-
expansion in short-term. However, when
tier cities are larger than higher-tier cities,
large-sized institutions are ready to
which gives AI adaptive learning a better
penetrate lower-tier cities, we believe they
chance to show its benefits in “customized
will adopt AI Adaptive Learning as a crucial
tutoring”. According to the Ministry of
part of their business to compete in new
Education, the proportion of junior high
markets.
students entering in senior high school is
about 70% in first-tier cities, such as Beijing
and Shanghai, while the proportion in low-
tier cities is about 50% with some less than
China’s AI Adaptive Learning Industry Whitepaper I 181.3 Competitive Landscape of
China’s K-12 Offline AI Adaptive
Learning Industry
Technology Suppliers Content Suppliers
Upstream
Suppliers AI adaptive
Voice Image System Lower Question Lecture Teaching
process process building terminal bank video content
algorithm suppliers
AI adaptive
Product
Development
learning platform enablers
（To-B）
AI adaptive education
institutions
(To-C)
Education
Institutions
Education institutions
Students Students
Core players in AI adaptive learning industry
Sources: Deck research, Interviews, EY-Parthenon analysis
1.3.1 Overview of China K-12 AI Adaptive ► Upstream suppliers: including AI adaptive
Learning Industry Chain algorithm suppliers, technology suppliers
and content suppliers. AI algorithm
China’s K-12 AI Adaptive Learning industry
providers typically have higher technical
chain can be divided into three parts: upstream
advantages, which enables offline education
suppliers, product development, and education
institutions directly. Technology provider
institutions:
includes voice-recognition, cloud service and
other application developers. Content
publisher provides question databases,
teaching videos and materials, etc. Choosing
from upstream suppliers could be highly
fragmented since the cooperation model will
be varying according to different capability
distribution among downstream players.
China’s AI Adaptive Learning Industry Whitepaper I 19► Platform enablers: providing content and The core players in the chain are adaptive AI
algorithm platforms. Services include algorithm suppliers, platform suppliers and self-
knowledge graph development, labelling of operated education institutions. These players
knowledge points, and user-data-based AI can rely on their own platform to help
algorithm development. traditional education institutions or establish
their own institutions to serve to the K-12
► Education institutions: small or mid-sized
market.
education institutions are the major
customer of platform enablers for now. They Meanwhile, most AI Adaptive Learning
get franchised or licensed from platform platforms are adopting B2B business model at
enablers to get curriculum content, front- the current stage to achieve profitability and
end and back-end applications, operating gain the first-mover advantage by enabling
training, marketing support, etc. We also education institutions with intelligence. Long-
observed another type of full-fledged term speaking, as the market expands and
education institutions that has their own matures, we can anticipate more players with
platform for their users. B2C business models will begin to emerge.
China’s AI Adaptive Learning Industry Whitepaper I 201.3.2 Overview of China K-12 AI Adaptive ► L4&L5 - Advanced AI Tutoring & Full AI
Learning Offerings Teaching: It leverages AI Adaptive Learning
system to identify weak link for each student
As shown in the figure above, there are two
via AI deep learning algorithm, knowledge
major offering categories of K-12 AI Adaptive
evaluation method and knowledge tracking
Learning in China:
theory before class to generate personalized
► L3 - Partial AI Teaching: It uses AI-based learning path. During classroom tutoring,
tests to conduct a personalized diagnosis each student will use handy devices (e.g., AI-
before classroom tutoring, and teachers can integrated machine, PC, Tablets, etc.) to
make lecture notes based on the diagnosis learn based on pre-defined learning path
report. During tutoring, the human teacher whilst human teacher will be only
will lead the teaching activity with support responsible for the supervision, emotional
from ad-hoc online personalized exercises to counseling of the students. After class, the
adjust content in real-time. After class, platform will provide personalized exercises
personalized homework with smart to students and intelligent correction
correction function will be generated based functions to human teachers. Some
on student’s behavior during class. institutions will provide human teachers to
Homework result will also be shared with the answer unsolved questions after-school as a
human teacher to provide later explanations value-added service to improve tutoring
to students. quality. Simultaneously, the system will also
provide the report to parents which contains
multi-dimensional data covering learning,
exercise and behavior to generate a
comprehensive deep view of their kids.
Diagnose Planning Teaching Practice Feedback
According to the Personalized
Real teachers-led,
standardized question Analysis of
Based on AI with intelligent
material and recommendation students' learning
L3 competence test,
systematic
adaptive
Intelligent situation based
a customized technology
Partial AI diagnosis is
feedback,
providing
correction on small amount
teachers make Teachers provide of data such as
Teaching formed
lecture notes
learning data to
explanation exercises and
help teachers
tuned to the basing on the tests
make decisions
class’s level class test result
Personalized
The system question Analysis of
AI-led, with real
L4&L5- Based on AI recommends
teacher
recommendation students' learning
Advanced AI competence test, customized Intelligent situation based
responsible for
a customized learning path for correction on multi-
Tutoring & Full diagnosis is each student and
supervision and
Based on each dimensional data
emotional
AI Teaching formed teachers can tune
communication
students, AI carry of the whole
on such basis out customized learning process
counseling
Sources: Deck research, Interviews, EY-Parthenon analysis
China’s AI Adaptive Learning Industry Whitepaper I 21Comparing the two main offerings of AI “AI” teach, which disrupts our prior
Adaptive Learning, L3 is an enhancement to acknowledgment to traditional education. The
traditional education by assisting human two different starting points lead to differences
teachers. L4&L5 are real game-changers to in depth, coverage and application scenarios of
replace the role of human teachers by letting AI technology.
L3- L4&L5-
Partial AI Teaching Advanced AI Tutoring & Full AI Teaching
Collect multi-dimensional data, covering
Collect fewer dimensional data in small the whole process of teaching
Technical amount
level Involves multi-link data analysis and
Relatively simple algorithm, uses more customized path recommendation, uses
recognition AI technology more strategy AI technology
Real teachers in core teaching links,
VS AI-led whole process teaching, with real
teachers responsible for support
Teaching assisted by intelligent adaptative tools
mode Mixed-class-mixed-subject method,
Traditional education institutions where students of different levels and
teaching method subjects can study in one classroom
Schools with sufficient educational Schools lacking educational resources:
resources: full-time schools in first- and small and mid-sized educational
Application
scenarios second-tier cities and traditional institutions in low-tier cities
educational institutions
Sources: Deck research, Interviews, EY-Parthenon analysis
China’s AI Adaptive Learning Industry Whitepaper I 221.3.3 Competitive landscape of China’s K-12
AI Adaptive Learning
Market share Number of players Company feature
> 2,000 partner institutions
Distributed country-wide, with large customer base and
Leading revenue scale
players ~60% 2-3 Strength in technology capability, subject coverage,
branding and other aspects, maintain a leading role in the
industry
< 2,000 partner institutions
Started late, mostly start-ups or new business incubated
Tail
players ~40% >50
by large enterprises
Has not achieved large-scale profit yet, some of which
have growth potential due to advantage in niche segment
Sources: Deck research, Interviews, EY-Parthenon analysis
We estimate there are around 50 – 70 K-12 AI Overall speaking, the current China’s AI
Adaptive Learning platform companies in China, Adaptive Learning market is highly
with a total of 10,000 franchise or direct- concentrated (CR2>50% in terms of institution
owned institutions. Based on their market quantity). However, we think the competitive
share and business scale, we further landscape still has the chance to be disrupted in
categorized them into leading players and future due to the following reasons:
others:
► Rapid market expansion: AI Adaptive
► Leading players: There are only 2-3 Learning has enormous market potential and
institutions with more than 2000 franchise its penetration in the K-12 education market
or direct-owned institutions currently, is currently low. The large potential for
accounting for 60% of the total market. market expansion has attracted players from
Their institutions are distributed across the different industries to enter, resulting in
country due to their competitive advantages many tail players.
of technical strength, strong subject
► High technical barriers: AI Adaptive
coverage and brand communication. Squirrel
Learning solutions require large investments
Ai, iFLYTEK are two leading players with
into R&D and data collection at the early
their market share still growing rapidly.
stage due to high capital and talent needs.
► Tail players: most players in this group have Tail players still have the chance to set up
dozens to hundreds of institutions for now their own moat if they have the right
due to getting into the market relatively late. capability and sufficient investments.
However, the players can still deepen in
Looking into the future, we expect more
their niche market by relying on regional
emerging players will enter the market due to
advantages. Backgrounds of these tail
the further release of demand in the low-tier
players vary from technology companies,
cities. Particularly, technology companies will
online education companies and traditional
leverage their strength in AI technology and
education companies. They think AI Adaptive
enter into the L4 or L5 AI adaptive learning
Learning could be helpful to enter the
market directly, which may bring more
market quickly, especially for lower-tier
challenges to the current leading players. Tail
cities. Their solutions are at the early stage
players without a good business model and
and far from scale.
enough investments will be difficult to survive
in the market. Meanwhile, current leading
players may further diverge, with few top
players evolving into full-fledged AI adaptive
learning providers while others have specific
strengths in dedicated areas.
China’s AI Adaptive Learning Industry Whitepaper I 231.4 Key Success Factors of China’s
K-12 AI Adaptive Learning
Key Success Factors of
AI Adaptive Learning
AI technology Content User data
optimization development accumulation
Although operating models vary among ► Strong and market-leading technical talents:
different AI Adaptive Learning industry players, In the international academic community,
the key success factors can be summarized in the development of frontier AI theory is
three common elements: AI technology changing, and the effect is becoming more
optimization, content development, and user and more intelligent all the time. Given that
data accumulation. AI is a field with high technology barrier,
global talents are always in need, and AI
1.4.1 Key Success Factor 1: AI Technology
Adaptive Learning companies should ensure
Optimization
the strength of their AI R&D team by
The AI algorithm used by the AI Adaptive introducing more talents with diverse
Learning system is referred as Strategic AI academic background and maintain a close
Algorithm which is different from Recognition relationship with academic institutions to
AI Algorithm used in face recognition, image stay ahead in both theoretical and practical
recognition, speech detection and other means.
applications and puts forward a stricter
► Multiple technical routes in AI algorithms to
requirement for algorithm structure and
maximize performance: In the course of
training. Furthermore, the effect of customized
years' research of AI Adaptive Learning,
learning diagnosis and learning path planning
many kinds of AI algorithm theories have
largely depends on the excellence of the AI
been formed. Each one is more suitable for
algorithm model. For that reason, AI Adaptive
solving a specific problem in the teaching
Learning players are investing heavily in
process. The practice has proved that there
developing AI algorithm and strive to achieve
is no one-size-fits-all AI algorithm model for
the optimization of AI algorithm technology
all application scenarios. Under the complex
with the following key considerations:
and random scenarios of different students,
it may be the only way to realize the
ultimate goal of adaptive education by
adopting various algorithm models and
theories.
China’s AI Adaptive Learning Industry Whitepaper I 241.4.2 Key Success Factor 2: Content to the algorithm of AI adaptive learning system
Development during curriculum development. In essence,
traditional curriculum content needs to be
Curriculum content development has always
optimized and reshaped, particularly in the
been closely linked with the competitiveness of
following aspects:
traditional K-12 education institutions. For AI
Adaptive Learning players, they should not
only know how to teach students with
knowledge points, but also how to adapt
teaching process
Refined knowledge Highly localized
map teaching material
Wider coverage of subjects
`
Low High
R & D level of course
content
China’s AI Adaptive Learning Industry Whitepaper I 25► Refined knowledge map to maximize ► Highly localized teaching material:
personalization: China’s K-12 education Presently, the industry players develop the
system is mostly exam-oriented. The teaching materials mainly based on several
measurement of student’s capability highly mainstream versions of textbooks from
depends on the mastery of relevant different provinces in China. However,
knowledge points. Students cannot get a considering of widely differentiated
higher score without the establishment of a textbooks and question types among
complete knowledge graph. Fine granularity different provinces in China, it will maximize
of subject knowledge points is essential for students’ learning efficiency if more
an AI Adaptive Learning solution to assess appropriate teaching materials can be
student’s capability accurately and provide developed to match with their daily-used
personalized learning content efficiently. If textbooks.
the content reserve is limited (not enough
► Wider coverage of subjects: Current China
materials) and the knowledge points are not
AI Adaptive Learning players are still
precise enough (a video corresponds to
fragmented in targeting primary and middle
multiple knowledge points and last more
school students, with more focus on science
than 5 minutes), there will be only a few
subjects. As a result, many parents need to
fixed learning paths or content to
use multiple platforms at the same time to
recommend for students, which is difficult to
cover their children’s whole learning period,
realize the effect of the real customized
which can be unfavorable to both customer
learning. Therefore, to improve the fineness
cultivation and customer loyalty. Companies
of knowledge points, some leading
with broader coverage of subjects will get
companies choose to design nano-level
higher customer satisfaction and credits.
knowledge graph through research and
develop related teaching content (such as
short videos, animation, etc.) on their own.
The whole R&D process involves a large
amount of investment and time input but can
set up a high barrier and enlarge the gap
with competitors once completed.
China’s AI Adaptive Learning Industry Whitepaper I 261.4.3 Key Success Factor 3: User Data From this point of view, platform that wants to
Accumulation build up the competitive advantage should
expand their coverage to the whole learning
User data is becoming a common barrier in the
process as soon as possible to start user data
digital era, which also works in the AI adaptive
accumulation. On the contrary, those question
learning area. The large amount of student and
databases and homework apps only have
teacher data generated through teaching
access to student’s behavioral data in one
activity could lead to a huge competitive
learning link, and the static student portrait is
advantage in product and operation
difficult for the system to generate real-time
optimization:
interaction.
► Achieving customized teaching requires the
► Proper mining and utilization of user data
storage of the entirety of learning process
generated during the application of AI
data, in which AI Adaptive Learning
Adaptive Learning can bring great
platform has a natural advantage. The data
commercial value to the company. In
set needed for AI algorithm training comes
current AI Adaptive Learning solutions, user
from the whole learning process of each
data such as personal data (name, age, etc.),
user, including feedback and interactive data
learning data (knowledge structure and
between learner and teacher. It helps to
progress, etc.) and behavior data (time to
generate the portrait of a student in real-
answer a question, etc.) are mainly used for
time through this continuous and multi-
learning status diagnosis and learning route
dimensional data accumulation, which helps
planning. Other interactive data (e.g.,
to predict their reaction and carry out
expression, ECG, etc.) plays a supplementary
personalized learning path planning. Data
role increasing learning quality. Companies
collection has encountered stricter policies
can fully leverage the value via data mining
and regulations nowadays, and the only way
to optimize their product or even monetize
for companies to get access to valuable data
the data in cross-industry to bring more
is through their own products.
value-add to itself.
► Name
► Age
Personal ► …
data
► Expres
sion
► ECG
► … ► Time to
answer
Interactiv Behavioral ► …
e data data
Learning
data
► Knowledge
point
structure
► Learning
progress
► …
China’s AI Adaptive Learning Industry Whitepaper I 2702
Overview of China’s
OMO AI Adaptive
Learning
Common
1 Constraints on space and 1 Poor learning experience
Limitation of Offline
Limitation of Online
limitation
time
Lack of high-
2 Capital-intensive operation
Education
Education
quality teacher is
with high marginal cost the bottleneck of
development
2 High customer acquisition
3 Small institutions lack and marketing cost
systematic development of
teaching material Insufficient
customization of 3 Customer’s willingness to
4 Small institutions lack brand class and high renew is low
construction cost of 1-on-1
teaching
Sources: Deck research, Interviews, EY-Parthenon analysis
China’s AI Adaptive Learning Industry Whitepaper I 282.1 Background of OMO AI Adaptive
Learning
2.1.1 Limitation of Online/Offline Education traditional offline education. However, offline
education has maintained its prevalence
Emerging from the wave of “Internet+”,
throughout the decades since online education,
innovative online education modes, such as
although made up some weaknesses of offline,
MOOC and live streaming, have brought
also has its own limit. In essence, both online
disruptive changes to the traditional education
and offline education have their own limitation
industry through a break to the limit of time
in terms of business model and user experience.
and space. It was once assumed that the
capital-light online business model would
certainly overtake, or even replace most of the
Common
1 Constraints on space and 1 Poor learning experience
Limitation of Online
limitation
Limitation of Offline
time
Lack of high-
2 Capital-intensive operation
Education
Education
quality teacher is
with high marginal cost the bottleneck of
development
2 High customer acquisition
3 Small institutions lack and marketing cost
systematic development of
teaching material Insufficient
customization of 3 Customer’s willingness to
4 Small institutions lack brand class and high renew is low
construction cost of 1-on-1
teaching
Sources: Deck research, Interviews, EY-Parthenon analysis
China’s AI Adaptive Learning Industry Whitepaper I 29► Limitation of offline education: ► Small-sized institutions lack brand
building: Although the number of small
► Constraint on time and space:
and medium-sized institutions in China's
Convenience is one of the main factors
K-12 education market is large, they are
considered by parents and students since
too scattered due to the lack of branding,
offline education requires students’
with less attraction to consumers and
physical presence. The demand for short
therefore lower revenue. Non-brand small
distance and suitable class time leads to a
players are at a significant disadvantage
smaller target customer group since only
in acquiring customers due to
a nearby population can be reached,
parents/students’ preference to brand
creating a ceiling for customer acquisition.
popularity and reputation. However,
► Capital-intensive operation with a high brand building takes a long time, and it’s
marginal cost: Offline education challenging to quickly establish trust in a
institutions usually adopt a capital- large consumer base. Small education
intensive operating model, with a large institutions are in urgent need of
investment in teachers, venues, enhancing their brand influence within a
equipment, etc., making it more short time.
challenging for scale expansion with slim
profit margin and high marginal cost.
Cost Structure of K-12 Offline Education Institutions
► Small-sized institutions lack systematic
development of teaching material:
100%
According to EY-Parthenon analysis,
more than 90% of offline education ~20%
institutions are small and medium-sized
players. These institutions usually lack ~35%
systematic content development
methodology, with unstable education ~15%
quality depends entirely on teachers’ own
~15%
experience, making it difficult to
guarantee students’ learning effect. ~15%
Revenue Profit Teacher Rent Marketing Other cost
Salary
Sources: Deck research, Interviews, EY-Parthenon analysis
“ A good teaching content development system should be highly standardized and localized.
Only giant educational institutions can realize this. Small and micro training institutions do
not have such capability. As a result, in small institutions, teaching quality highly relies on
teachers themselves rather than institutions’ R&D capability. Once the good teacher leaves,
students are easily lost as well.
——Expert from Traditional Offline Institution
China’s AI Adaptive Learning Industry Whitepaper I 30► Limitation of Online Education: However, due to fierce competition and
dissatisfying user experiences, what huge
► Poor learning experience: Student-
marketing investment generates is not a
teacher interaction is limited as students
proportionate growth in customer base
take classes online, which may negatively
but a surge in customer acquisition cost.
affect their attention, learning efficiency
Research has shown that the current
and class atmosphere. In particular,
average customer acquisition cost of
children in K-12 group have relatively
online education is around 500-1,200
weak self-control and need to form good
USD per capita, far exceeding the around
learning habits under the teacher’s
80-160 USD per capita for offline. The
supervision, which is challenging for
extreme high cost has a significant
online education. A survey during the
negative impact on online institutions’
covid-19 pandemic shows that the main
profit even when they have achieved
drawbacks of online education are related
positive business growth. Most online
to learning experience, among which
education companies’ financial reports
66.8% of users believed that the
show that they are actually in a state of
interaction is limited online; 50.2%
loss.
mentioned that teachers and students are
not quite adapted to the new mode, ► Customer’s willingness to renew is low:
45.5% considered online education with User loyalty is usually low in online
insufficient supervision, and 37.8% noted education due to the dissatisfied learning
that their online experience is not that experience and refund issues related to
user-friendly. the crash of some startups. According to
an industry insider, the online conversion
► High customer acquisition cost: The
rate of long-term and regular-priced
development of offline education has
customer is only 10% - 30%, with a
slowed down due to the pandemic.
renewal rate of 50% - 80%. How to
Investors’ attention, therefore, shifts to
improve user loyalty is another challenge
online, with several players gaining
for online players to sustain their
colossal finance. Capital-abundant players
business.
advertised heavily and launched price-off
promotions to achieve rapid growth.
Customer Acquisition Cost (USD per capita)
Survey on the Shortcomings of Online Education
Insufficient Teachers and
66.8% interaction
50.2% students are
not adapted
Inadequate Experience is
45.5% supervision 37.8% not user-
friendly
80-160 500-1,200
Sources: Beijing Normal University’s New Media Research center,
Guangming Daily education research center
Offline education Online education
Note: customer acquisition cost = (total marketing expense + sales
expense)/ number of new customers
China’s AI Adaptive Learning Industry Whitepaper I 31You can also read
