The numbers Brief Episode map Timeline Takeaways Charts Quotes Clips Glossary Go next Transcript

Episode 172 · Impact · 63 min

The rigour goes inside the timetable

While billions burned on direct-to-consumer video, SpeEdLabs spent three and a half years building for the person B2C tried to delete: the tier-three schoolteacher. Sixty to seventy per cent of its app is written for teachers, its books are designed for students who read a hundred words a minute against a metro child's three-fifty, and the whole high-rigour JEE and NEET programme costs ₹3,000–5,000 a year, billed inside the school fee.

Education & skilling Product strategy Consumer India

Vivek Varshney

Founder, SpeEdLabs · with Vishal Krishna

1:02:40

Said in this episode

▶ 34:48

800 → 2,000

Schools on the platform, and the target

Nine schools signed in the first school year, 2022; more than 800 today, growing about 2.5x a year, with a stated aim of crossing 2,000 within two years.

▶ 33:49

5–6x

LTV to CAC in the school business

Varshney's figure for the current model, where 180 to 200 students per school pull acquisition cost per student down and retention does the rest.

▶ 41:21

₹3,000–5,000

Price per student, per full year

Billed through the school and roughly 10% on top of the fee; he says he deliberately avoided designing a ₹8,000–20,000 product.

▶ 18:55

100 vs 350–400

Words per minute, tier-three vs metro student

Varshney's own metric for why the books use simple English, bullet points, flow charts and mind maps.

▶ 37:50

3.6% → 40%

Share of students vs share of fee spend

The premium and super-premium layer as he describes it; his student-count for the same layer moved between roughly one and two crore inside the same passage, so treat the headcount loosely.

▶ 56:40

$75 bn

India's K12 spend by parents

He expects it to roughly double in five to six years, a 12–14% CAGR, with edtech growing faster than the overall market.

The brief

The argument in sixty seconds

Varshney's claim is that Indian edtech did not fail. The direct-to-consumer version of it did, and for a reason that has nothing to do with technology. Roughly ninety per cent of students watching video never made notes, so revision meant re-watching, learning outcomes were inferior, and steep acquisition cost against high churn finished the arithmetic. His answer was to inject the entire high-rigour test-prep system into the school timetable rather than around it, which turns out to require five things, only one of which is software: books written in simple English for tier-three and tier-four students whose reading speed tops out near a hundred words a minute; a platform where sixty to seventy per cent of the functionality serves teachers who were never themselves trained for competitive exams; national-level assessments so that ranking first in a class of forty means something; a field team that sits through a full forty-minute class every alternate month; and dashboards for owners, principals and parents. Priced at ₹3,000–5,000 per student per year, about ten per cent on top of the fee, and sold through trained distribution partners into schools charging ₹15,000–40,000 annually, the model only pays at density: 180 to 200 students per school, regional teams instead of Kashmir-to-Kerala travel, and an LTV-to-CAC of five to six. Nine schools in 2022 became more than 800; the stated target is 2,000. The stakes are the 1.2 crore students who have moved from government to private schooling in five years, and a $75 billion K12 spend Varshney expects to double.

Worth your time if you are

Edtech founders who watched the B2C model burn

School owners weighing a test-prep partner

Parents deciding when a child should start competitive prep

Operators building distribution into tier-three India

Investors sizing India's K12 spend

Episode map

Where the conversation travels

Every block is a chapter, coloured by what it's about. Click any of it to jump straight to that minute on YouTube.

Education & skillingProduct strategyConsumer IndiaUnit economicsSales, GTM & growthAI & machine learning

01Cold open: from IIT to investment banking 0:00 Vishal frames edtech as a category that refuses to die, and Varshney traces the route in: IIT undergrad, an IIM Lucknow finance MBA, eleven years across Morgan Stanley, Deutsche and ICICI Venture doing more than $500 million of deals, and the decision at 35 to walk out of it and stick with the consequences. 02Education is the only currency 4:35 The outcome that matters is not 95% in the tenth board but a career, and because premium seats are scarce and aspirants are not, test prep is simply the reality of India, which is why he began building an adaptive JEE and NEET platform in 2016 to keep every student practising inside their own zone. 03From test cricket to T20 7:30 Varshney sat a nine-hour JEE of fifteen subjective questions per three-hour session; today's paper is three hours and seventy-five questions, so with 45% on the advanced paper enough for IIT Bombay, the skill under test has shifted from depth to speed and question selection. 04Three years of product, a free first customer 10:28 He designed for everyone rather than a segment, stopped writing code in 2016 and failed to get a process patent in India, spent three and a half years building before the first clients arrived in 2019. They were coaching institutes, the first of them non-paying and one of them still paying eight years on, and then COVID forced the market online. 05Why B2C learning outcomes broke 14:39 Acquisition cost was steep and churn high, but the real defect was pedagogic: about ninety per cent of students watching video never made notes, so revision meant re-watching, nobody supervised, and there was no human connect. He settled from day one that the local physical teacher would not be designed out. 06Pillar one: books for a hundred words a minute 16:39 The founding principle is to absorb the high-rigour curriculum into school hours without extra cost or a second admission, and the first pillar is books in simple English with bullet points, flow charts and mind maps, because a tier-three student reads no more than 100 words a minute against a metro child's 350 to 400. 07Pillar two: an app aimed at teachers 19:54 The learning app is the intuitive half; the counter-intuitive half is a teacher app carrying lesson plans, lecture plans, slides and pre- and post-class videos for teachers in small towns who were never trained for test prep and have no playbook. That is 60 to 70% of functionality, by his count. 08Ranks, classroom visits and dashboards 22:09 National-level exams give a small-town rank meaning in percentile terms with a stated path to the next five points; a field team sits through full forty-minute classes every alternate month as accountability; and owners, principals and parents get dashboards and monthly implementation reports. 09The ₹1,500 pitch to a school owner 24:56 For the first two years nobody saw why a full-stack solution should cost 1.5 to 2x the books alone; the argument that worked was arithmetic: the teacher already teaches 300 hours a year, the student spends 500, the parent already pays, so an extra ₹1,000–1,500 is a cheap bet, droppable next year. 10Start at grade six, not after tenth 29:00 More than 70% of students chase board excellence and only switch gears to competition after the tenth board; the other 30% start early, and Varshney, who lost a year to the JEE because nobody told him it existed, argues middle school is the moment aspiration and self-control arrive. 11Density, not reach, fixes the economics 31:38 Thin spread across the country is what makes the early years unprofitable; at 800 schools you can cut regional teams that travel less and serve more schools per kilometre, and with 180 to 200 students per school the LTV-to-CAC lands at five to six, with distribution partners trained hard against misselling. 12Three mini Indias 35:05 Olympiads as milestones and the middle class's Saraswati-first instinct give way to the market map: about 12 of 25 crore students in government schools, a 3.6% premium sliver carrying roughly 40% of fee spend, an affordable-private middle split in two, and 1.2 crore students migrating out of government schools in five years. 13₹3–5k a year, and the loan nobody defaults on 41:10 He deliberately avoided the ₹8,000–20,000 bracket and priced at ₹3,000–5,000 per student per year, roughly 10% on the school fee and billed by the school, then argues that parents borrowing for fees are the least likely borrowers to default because nobody wants their child pulled out. 14More machine learning than AI 44:50 Serving more than 10,000 teachers needs a playbook beyond individual brilliance, and Varshney is blunt that the platform runs on algorithms and machine learning rather than pure-play AI. AI is used internally for content creation, tagging and test generation, with a vernacular layer built but not yet live. 15STEM Learn and the NEP's AI mandate 49:30 A second brand, stemlearn.ai, takes robotics, coding and AI courses to grades three to twelve, a mandatory line rather than an optional one under the NEP's AI requirement from academic year 2026. Varshney argues robotics is really a vehicle for creativity and critical thinking, not electronics. 16CUET as UPI, and a $75 billion market 52:17 Three decades of a binary engineering-or-medical choice are ending: CUET is becoming a single unified admission platform he compares to the SAT, IPMAT seats have gone from 150 to more than 2,500 and could reach 25,000, all inside a $75 billion K12 spend he expects to double at 12–14% CAGR, before a closing recommendation of Endurance.

Timeline

The arc, briefly

A decade of one edtech company that never sold to the consumer, dated from the conversation.

2016The adaptive engine starts. Varshney had spent eleven years across Morgan Stanley, Deutsche and ICICI Venture on more than $500 million of deals in India, South East Asia and China, and decided at 35 that he did not want to do it for the rest of his life. He began building an adaptive JEE and NEET platform, and stopped writing code the same year. He tried for a process patent in India and did not get one.

2019The first customers are coaching institutes. Three and a half years of product came before them, and the first one paid nothing. The oldest paying customer still on the books is about eight years old.

2020COVID sends the market online, and the go-to-market is wrong. Everybody moved online as a forced choice, but SpeEdLabs chased a direct approach with a handful of students per client. Varshney says flatly that he messed it up, and that knowing the numbers did not save him.

2022Nine schools. The school business proper begins, hybrid rather than pure tech, because most clients sit in tier-three and tier-four towns where students get nothing like the exposure of a child in Bangalore or Mumbai and need local mentoring.

2022-24Two years of arguing about ₹1,500. Full stack cost a school 1.5 to 2 times what books alone did, and nobody saw why. The pitch that worked was arithmetic: the teacher already teaches about 300 hours a year, the student spends around 500 hours of their life, the parent is already paying, so another thousand or fifteen hundred rupees is a cheap bet you can drop next year.

2026More than 800 schools, and density fixes the economics. At 180 to 200 students per school the LTV to CAC lands at five to six, regional teams have replaced travel from Kashmir to Kerala, and the count is growing about 2.5 times a year. Around it, 1.2 crore students have moved from government schools to private ones in five years, and the NEP makes AI compulsory from academic year 2026, which is the opening the second brand, stemlearn.ai, is built for. Vishal last met him at about 50 schools.

2028The target is 2,000 schools. He wants to cross it within two years now that the playbook is ready, inside a $75 billion K12 spend by Indian parents that he expects to roughly double in five or six years at a 12 to 14 per cent CAGR.

Takeaways

Ideas to carry out of this hour

B2C edtech broke on notes, not on technology

The defect Varshney names is pedagogic rather than commercial: about ninety per cent of students watching video never made notes, so every revision meant re-watching the whole thing, and the learning process stayed incomplete. Nobody supervised, nobody knew the student, and there was no human connect. The steep customer acquisition cost and high churn that killed premium-priced B2C were the second problem, not the first, which is why he decided from day one that the local physical teacher would never be designed out.

▶ 14:53 ▶ 15:39 ▶ 16:10

Sixty to seventy per cent of the app is built for the teacher

In a tier-three or tier-four town the teacher has often never sat a competitive exam, has no training ground and no playbook for delivering high-rigour content on a fixed timetable. So SpeEdLabs ships lesson plans, lecture plans, slides and pre- and post-class videos into a teacher app, and by Varshney's count the majority of the product's functionality points at teachers rather than students. That inversion is what makes the school a channel instead of an obstacle.

▶ 14:10 ▶ 20:54 ▶ 21:41

Books are written for a hundred words a minute

A tier-three or tier-four student reads no more than about 100 words per minute; a metro student averages 350 to 400. Varshney, who was Hindi medium until class twelve in Mathura, sat the JEE in Hindi and first met English education at IIT Kanpur, writes the books in simple English with bullet points, flow charts and mind maps. The point is not simplification for its own sake: every minute spent decoding English is a minute not spent practising physics.

▶ 18:25 ▶ 18:55 ▶ 19:26

The exam turned from test cricket into T20

When Varshney sat the JEE it was a nine-hour paper, three hours each for physics, chemistry and maths, with about fifteen subjective questions per session and time to iterate. Today it is three hours and seventy-five questions, so the skill under test moved from depth to speed and to choosing what not to attempt. Because roughly 45% on the advanced paper can be enough for IIT Bombay, strategy now beats completeness, which is precisely the gap an adaptive practice engine is built to close.

▶ 8:15 ▶ 8:42 ▶ 9:56

Density, not reach, is what fixes the unit economics

The early years lose money because a small school count spreads a field team from Kashmir to Kerala. At 800 schools you can carve regional teams that cover smaller territories, serve more schools per kilometre and build relationships in the local language, and with 180 to 200 students per school, acquisition cost per student collapses. Varshney puts the current business at an LTV-to-CAC of five to six, helped by stickiness: once teachers and students change consumption behaviour, they do not go back.

▶ 32:23 ▶ 32:51 ▶ 33:49

Three mini Indias, and only the middle one is sellable

Of roughly 25 crore school students, about 12 crore sit in government schools where a private company can do little that is meaningful. At the other end, a sliver Varshney sizes at 3.6% of students accounts for around 40% of all fee spend, the lakh-rupee-a-year layer. SpeEdLabs deliberately targets the affordable-private band charging ₹15,000 to ₹40,000 a year, where affordability, awareness and principal-level acceptance all just about coexist, and where 1.2 crore migrating students are arriving.

▶ 37:05 ▶ 37:50 ▶ 39:21

The pitch to a school owner is arithmetic, not aspiration

Full stack costs a school 1.5 to 2x what books alone did, and for the first two years nobody saw why they should pay it. The argument that worked prices the increment against spending already underway: the teacher teaches roughly 300 hours a year, the student spends around 500 hours of their life, and the parent is already paying, so ₹1,000 to ₹1,500 more is a small bet on the outcome everybody claims to want, droppable next year if it fails. At 800 schools he now leads with reference points and referrals instead.

▶ 26:55 ▶ 27:40 ▶ 28:25

AI is an internal tool first and a student product later

Varshney is unusually direct that his platform runs on algorithms and machine learning rather than pure-play AI, and that in tier-three, tier-four and semi-rural schools an AI tutor will take longer to matter than it will in the top tier. Internally, AI already does content creation, content tagging and test generation, and a vernacular voice layer is being built but is not live. Asked whether AI will be the most important thing in education in two or three years, he says yes, just not this year and not for his users first.

▶ 46:16 ▶ 49:03 ▶ 49:17

The numbers, drawn

What the episode measures

Every figure below was said on air, with timestamps included and caveats kept.

Reading speed decides how much practice is left

words per minute

As stated in conversation: a tier-three or tier-four student reads 'not more than 100' words per minute against a metro average of 350 to 400 (midpoint shown). Varshney's stated reason for writing the books in simple English.▶ 18:55

Where India's 25 crore school students sit

crore students

Government schools48%

Private schools52%

As stated in conversation: roughly 25 crore school students in all, about 12 crore in government schools and 12 to 13 crore outside; he separately noted that 1.2 to 1.3 crore students have moved government-to-private in five years.▶ 37:05

IPMAT seats, and where he thinks they go

undergraduate seats

Seat counts as stated in conversation: one IIM and 150 seats six years ago, more than 2,500 now, and his own projection of 25,000. The accompanying caption claim that '40 IIMs' now offer the programme exceeds the number of IIMs and is likely a transcription error, so only the seat figures are plotted.▶ 54:22

Five things, only one of them software

what it takes to put competitive prep inside a school day

Varshney's own list of what a school needs before high-rigour prep survives a normal timetable. The field visit and the national ranking are there because coming first in a class of forty tells a small-town student nothing about where they stand.▶ 16:39

Worth keeping

Lines that stay

For a country like India, education is the only currency to come out of poverty.

▶ 5:20

From test cricket, we have moved to T20. So the skills required are very different.

▶ 8:28

Students watching video — 90% of them didn't make notes. And if you don't make notes, your learning process is incomplete, it's broken.

▶ 15:39

This was very clear from day one of the company: whatever I do, I cannot and I do not want to do away with the physical local teacher.

▶ 16:10

People ask me how the app is distributed between teachers and students. Almost 60% or 70% of our functionalities are for teachers.

▶ 21:41

Clips that travel

Short on time? Start here

Parents and students staring down JEE or NEET

From test cricket to T20

How a nine-hour subjective paper became a three-hour, 75-question sprint, and why 45% on the advanced paper can be enough for IIT Bombay.

7:30 → 10:28 · 3 min ▶ Watch clip

Edtech founders and the investors who funded the last cycle

Why B2C edtech broke

The notes argument: ninety per cent of video learners never wrote anything down, and the CAC-and-churn arithmetic that followed.

14:39 → 16:39 · 2 min ▶ Watch clip

Operators building product for tier-three India

The five pillars, in his own words

Books at a hundred words a minute, a teacher-first app, national ranks for a class of forty, a field team inside the classroom, and dashboards that make it all legible.

16:39 → 24:56 · 8 min ▶ Watch clip

Founders selling low-ticket products into fragmented markets

Density, not reach, fixes the economics

Why the first years lose money, and how 800 schools convert cross-country travel into regional density and a 5–6x LTV-to-CAC.

31:38 → 35:05 · 3 min ▶ Watch clip

Investors sizing India's education spend

Three mini Indias

25 crore students, 12 crore in government schools, 3.6% carrying 40% of fee spend, and the 1.2 crore migrating to private schooling.

35:05 → 41:10 · 6 min ▶ Watch clip

Glossary

The jargon, unpacked

JEE

The Joint Entrance Examination for India's engineering colleges, now a three-hour, seventy-five-question paper where roughly 45% can be enough for a seat at IIT Bombay.

NEET

India's single national entrance exam for medical seats; with the JEE, one of the two binary outcomes Varshney built the original adaptive platform for.

B2B2C

Selling to an institution that delivers to the end consumer: SpeEdLabs sells to the school, the school delivers to the student, and the parent ultimately pays.

LTV / CAC

Lifetime value of a customer divided by the cost of acquiring one; SpeEdLabs' school business runs at five to six, against B2C economics Varshney calls unworkable at premium pricing.

Adaptive learning

Practice that tunes question difficulty to the learner's current level so they stay in their proximal zone, neither wasting time on questions that are too easy nor losing confidence on ones that are too hard.

CUET

The Common University Entrance Test, a single unified admission exam across central, state and private universities; Varshney expects it to become India's SAT, with subject papers stacked on top of aptitude.

IPMAT

The entrance test for the IIMs' five-year integrated undergraduate management programmes. There were 150 seats at one IIM six years ago and more than 2,500 today.

NEP

India's National Education Policy, which Varshney says makes AI a mandated part of the school curriculum from academic year 2026, which is the opening for his STEM Learn brand.

Connections

If this resonated, go here next

EP 52 · Consumer

Full transcript

The whole conversation, searchable

Auto-generated captions, lightly cleaned. Click a timestamp to open that moment on YouTube.