Assignment - ( Building a scalable quality tuition

offering product)

Vision: Building an affordable and scalable digital alternative to current brick

and mortar tuition classes where students from anywhere can experience the top
notch quality classes by top quality teachers , using technology to bring the best
qualities of traditional tuitions plus the quality of enhanced learning via innovative
content + support ecosystem to drive career outcomes using synchronous
learning for the Indian Ecosystem for the K-12 segment ( especially in-times of
the pandemic - COVID 19)

Why build a tuition offering for the Indian Ecosystem? The

construct

1. The Indian K-12 segment size ( students from grade 1 to grade 12 ) is around
270 million based on a 2019 estimate
2. Out of this segment 62% of families with students fall under the aspirer segment (
which means they realise the value of quality education but income bracket is
between 150k to 250k per year - a low income segment)

3. And 85+ all of the total segment still resides in tier 2 cities in India
4. From a spend perspective - India spent around 39 billion dollars in K-12 private
spending on education where the aspirer and mid income segment comprised of
60% and more of the spend
5. Of the 271 MN enrolled schoolchildren in India, approximately 90 MN use
supplementary education in the form of offline private coaching. The majority of
them—55%–come from low income households . The cumulative expenditure on
supplementary education is estimated to be $16 BN, of which low households
account for $5.8 BN, i.e. 36%. In terms of market value, this is second only to the
mid-income group spending.

6. In sum, households with children enrolled in grades 1–12 currently use 41% of
their education expenditure on supplementary education. By 2022, this share will
increase to 48% due to the growing need for out-of-school coaching, driven by
increasing aspirations, competitiveness, and demand for quality education.
7. All the above points confirm the need of supplementary education
particularly in the mid to low income segment as the demand is growing
rapidly
 8. The share of supplementary education will not only increase in % but also in
volume to around 30 billion dollars from the current 16 billion dollars by 2022 -
which is 100% growth predicted over the next 2 years
9. Thus outcome based learning is the segment for India , thats why
supplementary education becomes really important like exams such as IIT-JEE ,
NEET

What is the core need , such that the demand for supplementary
education is growing so fast ?

1. India suffers many impediments to learning outcomes. In the context of grades

1–12 education, these include teacher shortages, poor teacher training,
disparities in teacher-student ratios across geographies, disparities in access to
 quality education due to varying costs of education, and absence of pedagogy
that focuses on teaching at the right level
a. Lack of personal attention and poor teaching pedagogy;
b. Poor teacher quality / teaching quality
c. Value relative to cost of education
d. Lack of international standards, particularly among mid-income aspirer
segments

Overview of challenges in the supplementary education system ( reference of

data in the end)
In order to solve these challenges edtech solves the unmet opportunities of access to
quality education as infra spending by govt is limited

The Ed- Tech readiness framework ? - why edtech is suited to accelerate

innovation in the ecosystem ?

Ed-tech as a ecosystem solution is the best way to cater to the un-met needs of
the consumer market in supplementary education because

1. Digital adoption among families (and individuals) is accelerating

Currently, about 580 MN Indians have access to the internet, of which 350 MN are
active users. By 2022, this number is expected to surge to half a billion.

2. A Large EdTech-Addressable Population

India’s K12 EdTech-addressable population, as of 2019, is 150 MN (see figure below)..

The internet is, and will be, more of a utility than luxury for this demographic, which uses
mobile data far more than other age groups.
Increasing smartphone penetration and cheap data will continue to push existing and
future EdTech players to innovate for children from mobile-first, lower-income families,
i.e. aspirers.

3. Awareness about EdTech’s value in terms of content and delivery of

content is high (and growing)

Hybrid study models incorporating videos, online course materials, and study groups
blurred the line between brick-and-mortar and online education .Now, Massive Open
Online Courses (MOOCs), virtual classrooms, distance education, and online job
searches are becoming exceedingly familiar to India’s digital natives

Also the high engagement due to personalised delivery of content , access to content
via MOOC’s etc , grade to student level models are multitudes more powerful than a
single teach - orientation based value for quality education

Among internet users, there has been a dramatic increase, to 80%, in the awareness
about EdTech as a supplemental education alternative to offline tuitions

Understanding the basic needs, drivers, and learning outcomes of various segments is
fundamental in pinpointing a target customer’s willingness to pay

4. Vertical strengthening of fundamental needs of student parent ecosystem

has resulted in multiple innovation models of content education delivery
which has made the user base identify value and pay for it - driving
towards desired career outcomes
Outcomes such as qualifications of exams such as JEE , NEET , CAT etc

The multiple models are

1. Live Class - 1 to 1 coaching ---> Better teaching quality , personalised attention

→ difficult to scale
a.
2. Live Class - 1 to n ( n < 25) coaching model ---> Better teaching quality , slightly
lower attention ---> difficult to scale ( lack of quality teachers , internet depth )

a. Hypothesis to try solution 1 & 2 :- The hypothesis was that India currently
spends 40-50% of its spend on education by the top 20% segment
 b. Building a personalized attention model will drive higher ARPU like in
KOTA where the ARPU is 2 to 3 x higher than online course
c. Post analysis after MVP , we realised that it's very difficult to scale as
quality teacher supply in India is limited and most of the volume is in public
schools where teacher quality is Inadequate . This leads to a smaller
addressable market and lower scale opportunity
3. Recorded class , live support → 1 to n ( n > 100) → Better quality ,slightly less
personal attention ---> easy to scale

The Product Solution & its scope

The business hypothesis :

India is a budding nation in terms of internet penetration of active users . Still there are
more than 330 million indians on 2G networks and hence it becomes the need of the
hour to build a scalable solution for the mid to low income alongside the high income
groups

Based on the need the core issues are :-

1. Great quality teaching access at lower network speeds too

2. Doubt resolution and engagement structure
3. Innovative models of learning which exist in offline networks but not at scale so
hence its needed
a. What i mean by innovative models is that - in an offline tuition center - the
engagement is physical and hence all the three elements of learning are
easy to replicate with a good quality teacher
i. Personal connect is easy since generally a tuition center has 2-3
teachers which are assigned to a class
ii. Doubt resolution is instantaneous and both pre and post class
iii. Student to teacher , student to student , teacher to parent and
parent to student interactions are seamless and easy to access
b. But at scale this takes for a toss as brick and mortar institutions are highly
capital intensive and specific to area like Kota , Hyderabad which only the
high income segment can access
i. Bansal classes after starting in Kota in 90’s has been able to train
around a million students in 25 years
 ii. Also there is no free student learning structure so the families which
cant afford are left out of the loop to move to a desired career
outcome
iii. On the same premise an online ed-tech player like BYJU’s has
been able to scale beyond 7 million paid students in 5 years with 50
million free users deriving value
iv. Since the barrier to entry is very less in ed-tech and quality of
content is superior it makes all the more sense for a edtech solution
for supplementary education

For the 3 core hypothesis , a hybrid edtech solution then comes to mind

1. Recorded lectures which can be accessed at any point of time by the great
quality teachers such as in Kota etc
2. Live doubt structure as a chat or a teaching assistant + student model as a
product solution as video
3. Student to student interactions / student to teacher interaction / teacher to
teacher interaction

The Product Vision

“ To build a scalable platform for providing the best tuition class experience such
that the product has the same fundamental experience of a tuition class + the
advantages of ed-tech”

In order to build a product , these were the sample flow structure which was built for the
tuition offering

The product was a combination of :-

1. Recorded video player on which content is showcased

2. Real time messaging (RTM) module - on which interaction between
teacher-student is done
3. Interactive class - content interactions such as poll questions , interactive
questions
 1. Content delivery through recorded lectures by top teachers . This will be
showcased as a video screen playing the lecture
2. Also delivery of interactive content such as polls , interactive questions ,
whiteboard interactions launched by TA

Since its mostly recorded content to it does not require bandwidth of 100 mb plus as in
a live class so solves the purpose
Poll questions as interactive content

3. Chat as a method to incite engagement between students , for doubt resolution

and personal engagement like emojis, emoticons , expressions :)
 4. A batch system based on board or grade to ensure size of class does not dilute
the proposition of engagement and personal connect i.e batches are of the size
of 1 to 200 max - which should be reduced with the advent of time

The core business model

The business model is based on 3 aspects

 1. Core content which is driven by teacher - which is recorded the content has value
and also the problem of scale and internet infra does not occur
2. The TA / teaching assistant as a class facilitator which does doubt solving and
engagement on chat by doing polls , sharing content , interacting with students
and doing a video based engagement with them
a. This also is easy to scale as it does not need a healthy teaching skill and
would need basic skill of a graduate and a good income of 7-8 lacs which
is okay to afford on the ROI
b. As the LTV is 6+ years ( a student generally starts from 6th grade in
tuitions and ticket size is also high( min 500 dollars) so thus its a profitable
venture
3. The support ecosystem of self paced content - for pre post class behavior
4. Presence on all platforms which are wide like web , app , tablet etc

1 + 2 + 3 +4 = a compelling tuition offering

Stakeholder Buy in

The stakeholder buy in was majorly on what kind of product we have to build as due to
COVID - 19 , the emergence of a supplementary education structure was already
popular so we experimented with multiple models to arrive at the most logical model
which can scale and also provide user value

Hypothesis 1 : Build a 1 to 1live model which has most personal attention

and quality

1. The process started with first doing a 1 to 1 model as the tuition sector was
rapidly growing which is the most personalized
2. We built a 1 to 1 product using some external vendors such as agora for chat

Outcomes

1. We realised as a product 1 to 1 was difficult to scale from a value standpoint

because of “supply of teachers” was a constraint
 2. We also realised that the live model demands commitment from the teacher as
well as student so the ticket size will be outside the mass market
3. Also internet depth i.e 4G is needed to conductive live classes which is still in the
early 30% of the size of the market so limits the total addressable opportunity

Shelf life of experiment - 6 months

Hypothesis 2 : Build a 1 to many live model which has personal attention

and quality and some sort of scale too

1. The bet was on the fact that 1 to 30 will scale and will be relatively easy to
do as teacher supply requirement dramatically gets reduced

Outcomes

1. The hypothesis failed quickly because we realised soon the teacher supply +
internet depth again became a challenge

Shelf life of experiment - 4 months

Hypothesis 3 : Build a 1 to many live support model + recorded content

which has personal attention and quality and some sort of scale too

1. We moved to third model because of the no dependency on teacher supply

as content was available as a business with us
2. Instead of teacher supply - you need a teaching assistant supply which is
easy to get and cheaper to keep
3. Keep the TA + pre post content requisites as a support structure

Outcomes

1. V1 scaled easily - with synchronous learning structure scaled to 2 million +

classes in 8 months

Business consensus for model V3

 1. Since we were already a direct sales model . This model gave us the
opportunity to set a free trial structure of the class - since the product gave
us the flexibility to set up the free class structure without impediments from
“ teacher supply” and hence from a business consensus this was an easy
sell to business

Tech consensus for model V3

1. Since we had already done 2 versions ( live 1 to 1 ) , ( live 1 to many) -

this was an easy implementation of tech perspective - since it was
pre-recorded played on Shaka player ( native to google) and then the chat
experience via re-modeling agoras chat api
2. We started off on our app which already had traction for other products to
drive traffic for this product

Content perspective for model V3

1. It was an easy sell because we had most of the content ready , and
recorded so it was easy to set it across
a. Grade
b. Board
c. Topic
2. We also leveraged other interactive content like games and interactive
layers to give an amazing pre / post class experience

Ops consensus for model V3

1. This was the most difficult of all because we had to set up an army
of TA for the scale we had
2. The TA’s had to be train initially and set up a engagement structure /
doubt resolution structure
3. Post the initial version , we solved most operation problems at scale
a. I.e like we faced the issue of volume of chats in the class to be
a hassle as it becomes difficult to monitor questions asked
b. So we built a doubt assignment structure for the TA’s and
hence we could scale the same :)
 4. TA’s help in engagement , assessment , performance review and
doubt resolution

Rough user flow how the user experiences the product and then becomes a
paid user based on experience

How did we measure the success of the product in terms of

target KPI’s

The core success metric of the product was the fact that the core value
proposition of
● Quality teaching
● Instant doubt resolution
● Pre / post class assessment structures
● Grading
● Performance review
Should be taken care by the core baseline metrics and then by leading metrics
for finding improvements in the baseline

Area of focus Target Metric Comments Metric

1. No of users No of registrations / This is done to Baseline for

 entering top Total traffic on web , measure the revenue
of funnel app and other demand affinity of
platforms the free user set
which can be
converted as a paid
user

2. Attendance No of free session This allows us to Baseline for

% joins/ Total identify the intent of revenue
registrations the free user and
allows the user to
experience the
product . This is a
MQL for us to
convert to a paid
user

3. Session Average session The higher this Baseline for

watchtime % time spent by the percentage higher retention/renewel /
student in a batch ( the engagement revenue also
total time spent by and qualification
students / sum( rate of the student .
total time available Because this is the
for students) true experience of
the class . ( 70%
plus watchtime
have a 3x more
times renewal rate
in paid users

4. Doubt No of doubts This gives the feel Leading metric for

resolution answered in class / of instant doubt watchtime %
percentage No of doubts asked resolution to the
in class user and also
instant value and a
personal connect

5. Average NPS of the session This gives the Baseline metric for
rating per average feedback revenue ( if this is
session of the users in the high retention /
session i.e with revenue is high)
core issues like
teaching quality
captured post the
user tries to leave
the session
6. Course The total course This includes the Baseline for
completion completed by the core offering retention/ renewal
rate student / total including live metrics
available course for sessions , pre
him requisites , post
requisites ,
assessments etc .
This gives a
semblance of
retention of the
user as he has paid
for the product .
Higher the course
completion rate ,
higher the chance
to retain the user

7. User Total session in a This is average Baseline for paid

retention month done by the user retention renewal for us
user / total sessions which should be
available in the growing or flat with
month for him time

8. Engagement Positive chat We have built Leading metric for

messages per user persuasion watch time %
elements like
emoticons , stickers
for having the
feeling of fun in the
class
These are the few core business and product KPIs

we measure for the product besides the product analytics behavior the
user like the app , web funnel

Implementation structure and hurdles

The core hypothesis was to create a product which takes care the three core
areas
1. Should be able to conduct classes at a concurrent scale of 20k+ users
initially
2. Should be able to provide a personalised class experience ( batch split to
be in such a way - large classes don't make the experience monotonous )
 3. The experience should be seamless across all our platforms web , free
app , paid app
4. Should be able to support the operational efficiency of the TA to support
the students

The core product flow was built in such a way

1. The core challenge was to test the concurrent users at scale such
that our production system should be able to scale seamlessly
a. In order to take care of that - we created a scalable system of batch
moderation where the backend takes care of breaking classes apart
in a ration of max 300 per class as the content is common
b. What is needed as a TA structure which should be available to join
any session in case traffic / registrations are high
c. I.e assume 3000 students book for a class , the backend 15 min
before the class starts splits the batch into 10 classes with similar
content
d. The TA team is notified on the ops dashboard in real time about 10
new class id’s created and hence they in read time all more TA’s to
all classes
e. We typically keep the balance of 1 to 30 ( TA - student )
2. The second biggest challenge was to create a trouble shoot module
for students in case they face issues
a. Because at scale if 1% students face an issue , it impacts a large set
of users
b. So we built a troubleshoot module for students and tech to de bugg
issues faster in the product due to scale for example the sample
screenshot given below i.e video buffering , any other issues
 3. The third most important element was to create a doubt resolution
flow for students to track doubts , get answers and feel happy in the
class
a. This was because at scale without a doubt resolution flow - this
becomes very difficult to cater to the questions of the student as on
average we saw a student asking 6-8 doubts per session (+ 40 plus
average messages on the chat ecosystem )

Experiment result - V3
 1. As we started the product offering to free users / paid users - we saw the
following metrics in terms of success

2. Revenue funnel

In the month of march 2020 - we did an experiment to test the hypothesis that
synchronous learning by launching BYJU’s free class as a product . Due to the
advent of covid and our SEO presence the product lead to 30k to 40k users
joining classes every day on our android app and web platform - this paved the
way for us to experiment on the one to mega model for synchronous learning

Free class was a structured learning course available to the entire large top of
funnel where classes were done a particular time every day and since most
schools were closed this lead to the learning gap in india and hence most users
flocked to the same giving us a user growth of over 100% M-O-M ( month on
month)
Based on the success of the same - we launched the supplementary product
called BYJU’S Classes in the month of august on web first and then on android

The model was as described

Top of funnel

1. SEO - we get around 70 million hits a month on our web platform

2. Android - 50 million plus installs - we get around 1.5 million installs on our
app organically
3. Ads on fb and google - which brought another 10 million installs on our
android app
4. We also leveraged IPL as a lever for brand ads since we were the official
sponsor for Indian cricket team
Middle funnel

1. The users which land on the platforms were giving visibility to book a free
class which keeps happening everyday at multiple time slots and since its
majorly recorded content + live support - its easy to drive content creation
at scale since same content can be used across
a. Different boards - CBSE , ICSE
b. At different time slots same video can be played + only TA’s need to
be assigned
c. A user then attends the class - which we remind him via multiple
marketing communication such as email , sms , whatsapp

Bottom funnel

1. Once the student attends the class - the lead is sent across to our CRM
where it gets assigned to a salesperson - who the reaches out via tele -
calling
2. Based on the skill and need of the student - the sales person modifies the
pitch to showcase value to the parent and drive multiple year enrollment
programs to drive LTV ( yearly value is around a 1000 dollars which goes
upto 2000 dollars if a tablet is sold with the subscription too)

a. Business outcomes
i. Free users
1. Free trial class offering resulted into a registration rate of
50k+ bookings a day with the split of 20% on web , 80% on
our app ecosystem
2. Joining rates in the range of 30% for free users which is 20%
higher than the industry standard in the live trial class model
3. Watch time % in the range of 50%+
4. Revenue funnel lead to sales conversion of 10% at a ticket
size of 800$ per annum for free users
5. Since the product offering was lean + of the assisted sales
model - this allowed us to scale at a ticket size of 500 $ to
1500$ in the mid to low income segment - we had a
directional conversion of 5-10% at a ticket size of avg 900 $
 which is really good as at this ticket size industry average is
3%
b. The funnel - the product at current scale everyday does around 30k
booked classes with attendance of around 8k students daily leading to
around 200k unique students attending our free classes in a month - since
the conversion is around 7 to 10 percent - learning to 14k to 20k paid
users on an average ticket size of 800$ approximating to around 10-12
million $ per month as revenue
c. From a unit economics standpoint - majorly traffic on our website comes
from SEO to organic traffic and we run paid ads on google / facebook
where the CAC of a customer comes around 300$ for a paid customer

i. Paid users
1. The course completion rate shot to 60% plus due to the
session offering and TA ecosystem from an early model of
40% in 6 months time
2. Watch time % on average was 70% plus for paid user
3. Average class rating 4.5 / 5
d. Product outcomes
i. For paid users the 6 month retention was around 66% and 12
month around 58%
ii. Due to optimisation of product funnels - we saw the join
percentage shoot up to 30% from 21% at start of the
experiment in August , majorly due to contextual notification
structure and a smooth product discovery experience
iii. Also the course completion rate was around 52% ( industry
average < 40%)

This in a nutshell has lead to establishment of a synchronous learning

model which is poised to scale

Next steps

The next steps for the V4 for the tuition offering product is
 1. How you to exactly replicate the offline tuition behavior on web platforms ?
2. How do you offer interactive content for students to identify the value of
tuition classes
3. How to build even more personalised touch in the classes and not only
drive tutor student but student student engagement and learning outcomes
too :)

Some ideas

1. Building a live TA + small class model i.e TA can stop the session in
between and ask for doubts by bringing class student on video and
leveraging features such as whiteboard and interactive content
2. Even before the session - students can join just like a tuition class about to
begin and clear doubts among themselves
3. Content to shelved as x min modules
4. This will break the session in subsessions and be close to the actual tuition
offering that is there

We are trying to implement this flow as a TA not as a teacher but as a facilitator

model of class - it solves 3 problems for the business model

1. TA by acting as a facilitator than a teacher - is easy to source for the

smooth functioning of the class
2. Personal attention becomes easy to reproduce for the students in
supplementary learning
3. Content stay as the core learning resource as it is keeping watching at
scale intact
Its the merging of the sync+async learning behavior as is a popular model
in the Chinese ecosystem

As the Edtech industry is poised to grow atleast 5.5x in user base in the
next 3 years this becomes a major model in supplementary learning

Technical implementation