Skip to content

🎯 Business model decision of record: B2B-first — sell seats and engagements to companies, not courses to students #83

Open
Open
🎯 Business model decision of record: B2B-first — sell seats and engagements to companies, not courses to students#83
Labels
enterpriseEnterprise & universitylaunch-windowTier 1: during beta, before public launchmarketingMarketing surfacesv1V1 (6-12 months)

Description

@teetangh
teetangh
opened on Aug 27, 2026

Status: Decision of record. Supersedes the positioning in temp/feedback/01 and temp/feedback/03–05 (2026-08-26) where they conflict.
Date: 2026-08-27 · Decided by: founder · Review: at the 90-day gate (2026-11-25)

TL;DR — the decision

We are not building a K-12 / NCERT / JEE product. We are not building a horizontal Udemy. We are changing who we sell to.

elluminar stops trying to sell courses to individual students and starts selling seats and engagements to companies and institutions, delivered on the platform that already exists, with the mentor-reviewed rubric + verifiable credential as the thing that justifies a premium price — not as a product sold on its own.

Nothing already built gets thrown away. The catalog, quizzes, assignments, cohorts and projects stay as delivery scaffolding. What gets dropped is the ambition to be a self-serve marketplace and the assumption that individual consumers are the buyer.

Price floor is ₹50,000 per unit of sale. Not ₹4,000.

Why this issue exists

Current state, stated plainly:
Age ~8 weeks (first commit 2026-07-03)
Authors 1
Hand-written app code ~33,400 LOC TS/TSX
Prisma models 110 (≈34 have zero app code touching them)
Pages 77
Open issues ~40
Paying customers 0
Live payments No (Razorpay test mode; #38 open)
Real catalog content None — 100% seeded fiction (#37 open)
Product analytics None (#12 open)
Landing page social proof Hardcoded arrays — invented mentors, testimonials, stats

The founder is a solopreneur running this as a side project alongside a full-time job. He is not the teacher and not the reviewer — teachers and mentors are supply that must be recruited; students are demand that must be acquired.

This issue answers: pivot to school curriculum? keep the current model? organise what exists? or scrap it?

Research basis: a full codebase audit plus ~190 web searches across four independent research sweeps (Indian K-12/test-prep; global course marketplaces & cohorts; solo/small-operator economics; audited competitor financials). Every load-bearing claim below carries a source. Gaps are flagged as gaps, not filled with guesses.

PART 1 — What we are NOT building, and why

1.1 ❌ K-12 / NCERT solutions / PYQ papers / JEE Advanced prep

Five independent reasons, any one of which would be sufficient.

(a) The core value proposition is already free, and better.
India has 100M weekly active ChatGPT users — the second-largest market on earth — with 80% of usage under age 30 and 18–24-year-olds alone sending nearly half of all Indian messages. OpenAI launched a sub-$5 "ChatGPT Go" tier and then made it free for a year in India. (TechCrunch, 15 Feb 2026)

Our prospective customer already has an unlimited, multilingual, multimodal NCERT/JEE tutor on their phone at ₹0, funded by a company that does not need to monetise them.

(b) The distribution channel is being demolished while we would be taxiing onto it.

  • AI Overviews now trigger on ~83% of education queries; when one appears, organic CTR falls 61%. (upGrowth, 25 Jun 2026)
  • Chegg — the fully-worked case study — Q1 2026 revenue −48%, Academic Services −57%, subscribers −31%, stock $0.96 (−99% from peak), 56% of workforce cut, fighting NYSE delisting and suing Google. (Forbes, 29 Oct 2025)
  • India-specific proof: Info Edge's Shiksha posted Q4 FY26 billings −13% YoY, its first decline in six quarters, explicitly blamed on AI-driven search. (Medianama, May 2026)

Free-content-SEO → paid-conversion is a 2018 business model in active liquidation.

(c) The unit economics have never closed, for anyone.

  • PhysicsWallah online ARPU: ₹3,918 (management: "less than ₹4,000") — net of 18% GST ≈ ₹3,320
  • Indian edtech CAC: ₹1,500–4,000
  • B2C edtech annual retention: ~40% (vs ~85% B2B)

LTV:CAC is at or below 1:1. This is why BYJU'S died, why Vedantu spends ₹1.96 to earn ₹1 (FY25: ₹227 Cr revenue, ₹210 Cr pre-tax loss, ₹40 Cr cash left), and why Unacademy sold at −94% ($3.5B peak → ~₹1,955–2,055 Cr / ~$218–230M).

(d) The market leader nets approximately zero, and #2 and #3 are shrinking.

Player FY26/FY25 Revenue Result
PhysicsWallah FY26 ₹3,900 Cr (+35%) net loss ₹24 Cr
Allen Career Institute FY25 ₹3,067 Cr (−5%) PAT ₹41 Cr (−70%)
Aakash Educational FY25 ₹2,032 Cr (−16.6%) loss before tax up 4.8×
Vedantu FY25 ₹227 Cr loss ₹210 Cr

Doubtnut — a top-3 free doubt-solving property with enormous traffic — sold to Allen for ~$10M, down from a $150M offer in 2020. That is the terminal value of NCERT-solutions traffic.

(e) PW has already shipped exactly this product.
"NCERT Pitara" generates MCQ/fill-in-the-blank questions directly from NCERT textbooks; AI Guru has resolved 100M+ questions and evaluated 2M+ answer sheets. 1.5M users in two months. Backed by a formal Microsoft Research partnership and a listed balance sheet. Their stated thesis: whoever holds the most behavioural + Q&A data wins the AI-tutor race. (Microsoft Research; Business Today, 1 Jun 2026)

We would start at zero on both inputs.

(f) And it discards 100% of what we built. Parents buy marks, not portfolios. Mentor-reviewed capstones, rubric scoring, verifiable credentials, seat licensing, double-entry ledgers — none of it transfers to a Class 10 student.

Regulatory note: MoE coaching guidelines (Jan 2024) prohibit enrolling students under 16; Rajasthan made this statutory in 2025. CCPA's 2024 misleading-ad guidelines are actively enforced (45 notices, ₹61.6 lakh in fines in 2024; Vision IAS fined ₹11 lakh Dec 2025) and now require written post-selection consent before using a selected candidate's name or photo. Private coaching and online learning carry 18% GST — GST 2.0 (Sep 2025) explicitly declined to extend relief.

➡️ Decision: NO. Not now, not as a side bet, not as an SEO funnel.

1.2 ❌ A horizontal "better Udemy" marketplace

(a) Udemy no longer exists as an independent company.
Coursera acquired Udemy in an all-stock deal that closed 11 May 2026; UDMY delisted from Nasdaq at 0.800 COUR per share, ~$2.5B implied. (Coursera IR; Axios, 11 May 2026)

Read it correctly: two sub-scale public consumer-education companies merged defensively into one enterprise L&D vendor, targeting $115M in cost synergies, then announced layoffs. The marketplace model wasn't disrupted — it was absorbed and de-prioritised. There is no angry incumbent to take share from, only a shrinking pond.

  • Udemy FY25 revenue $789.8M (flat); Q4 2025 −3% YoY, first decline since IPO
  • Consumer segment FY25 −9%; Q4 −14% (Udemy 10-K FY2025)
  • Coursera Q2 2026: consumer −5% normalized, enterprise NRR down to 91%, net loss $80.4M

(b) "Instructor pricing freedom" is a real grievance and a false wedge.
The pain is genuine and documented — instructors received 32¢ of every learner dollar in 2025, down from 35¢; subscription share was cut three years running (25%→20%→17.5%→15% from Jan 2026) after a public promise not to; sitewide 80–90% discounts have no instructor opt-out.

But the arithmetic kills it:

  • 75% of Udemy instructors earn under $1,000/year. They cannot pay us anything.
  • ~1% earn >$50,000/year — and they already have their own funnels and use the 97% coupon channel.
  • The instructors angry enough to switch are angry precisely because Udemy's marketplace demand is the only demand they have. Take them off Udemy and they sell zero.

The complaint is about revenue share. The actual problem is demand dependence. Better terms without demand solves the wrong half. And "set your own price / keep 100%" is already the default marketing copy of Thinkific, Teachable, Podia, Kajabi, LearnWorlds, FreshLearn, LearnHouse, Ruzuku and Igniter.

(c) Nobody has solved demand aggregation at any scale we could reach.
Maven — a16z-backed, $30M raised, five years, an ex-Udemy co-founder — takes 10%, and only ~20% of instructor sales come from Maven's own marketplace. The other 80% the instructor brings. Maven's own published figure: average Year 1 expert earnings = $10,000. No round since 2021; no revenue, GMV or headcount disclosed since 2023. (maven.com/teach; Gagan Biyani, Sept 2023)

(d) The India version was attempted with ~$159M and abandoned.
Classplus built exactly "Shopify for Indian tutors" (₹13,000–50,000/yr SaaS + 15% commission). Post-COVID, tutors reverted to WhatsApp and YouTube. SaaS traction collapsed. Classplus is now a test-prep company (Testbook, which contributed 64% of consolidated FY24 revenue), plus an offline BTech college and an education-lending stake. No round since Series D, Mar 2022. (Inc42, Jul 2025)

(e) The India instructor math is brutal. A ₹1,299 course on a 37% marketplace share nets the instructor ~₹480. To reach ₹1L/month at our 20% commission we'd need volumes that do not exist without years of free-content funnel-building.

➡️ Decision: NO self-serve horizontal marketplace. Supply becomes a small curated bench we recruit by hand.

1.3 ❌ Live cohorts as the differentiator

(a) The practical reason, which is decisive on its own: the founder has a full-time job. Live classes happen at fixed times regardless of what else is happening. This is the one delivery format a side-project operator structurally cannot run.

(b) Live cohorts have linear cost and zero operating leverage. 30 students costs 30× the hours of 1. Kraftshala is the cautionary case at the right price point: ₹1,64,000 + GST, 920+ hours of live instruction, a 60% placement-linked refund — and FY25 revenue ₹10.1 Cr (−19.9%) with a ₹1.8 Cr loss. (Inc42 Kraftshala financials)

(c) The 2021 CBC hype did not survive. On Deck cut 25% of staff (May 2022) then a further third three months later. The famous "90–96% cohort completion" figure traces to Maven's own self-reported numbers — no independent study verifies it.

(d) It does not exist in our codebase anyway.

Capability Reality
getLiveSessionEmbedData() (src/lib/fermion/live.ts:33) zero callers
A page a learner can open to join a class does not exist
createVideoUpload() / markUploadedAndProcess() (src/lib/fermion/video.ts) zero callers
LiveAttendance table never written
Recordings none
DRM none — player is a bare <video> with controlsList="nodownload"
Code labs (src/lib/fermion/labs.ts) all three functions zero callers
Fermion API key not held; vendor acquired by Testpress, signup paused mid-transition

Live cohorts today are a scheduling + notification feature, not a delivery feature. See #52, #53, #59, #66.

➡️ Decision: NO. Cohorts remain in the schema as a scheduling primitive for B2B batches, not as a consumer product.

PART 2 — What we ARE building

elluminar is a B2B training and verified-assessment vendor.
Companies and institutions buy seats or engagements. A small curated bench of expert instructors and mentors delivers. The platform runs enrolment, content, assessment, rubric-scored human review, credentialing and reporting. Every learner finishes with a publicly verifiable, rubric-scored record of what they can actually do — which is what lets us charge the top of the market rate rather than the bottom.

What changes

From To
Buyer individual student company L&D / GCC / college / coaching chain
Unit of sale a ₹1,499–8,499 course a ₹50,000+ engagement or a seat block
Supply open self-serve marketplace ~5–10 hand-recruited instructors/mentors
Acquisition Google/Meta/Insta ads LinkedIn + WhatsApp outreach, founder-led
Delivery live video we don't have async: content + assignment + rubric review
Credential role the product the pricing lever inside a B2B sale
Vendor dependency Fermion (blocked) none required

What is reused, not rebuilt

All of this is already implemented, wired and unit-tested:

  • Commerce: cart → Razorpay → signed idempotent webhook → fan-out fulfilment → double-entry ledger (LedgerAccount/LedgerEntry, signed, append-only, idempotency-keyed) → gapless GST invoice numbering (row-locked nextInvoiceNumber) → refunds/clawbacks
  • Enterprise: Program, ProgramCohort, OrgLicense, LicenseSeat, LicenseConsumption, credit-pool drawdown with SELECT … FOR UPDATE, roster CSV import, ReportExport, OIDC/SAML SSO with domain verification
  • The review loop: Project (SPRINT/CAPSTONE/FLAGSHIP) → Milestone → submission with file upload → MentorAssignment → rubric-scored ProjectReview → four enforced gates (commit e4e17c5) → Credential → /verify/[code] + printable certificate + /p/[slug] portfolio → mentor ledger payout at 55%
  • Multi-tenancy & RBAC: Tenant with TenantType CREATOR/ENTERPRISE/UNIVERSITY/HIRING_PARTNER, commission snapshotting at order time, BetterAuth organizations
  • Assessment: Quiz/QuizAttempt auto-grading, Assignment/AssignmentSubmission with a grading queue

This is already an institutional product. It was only ever pointed at the wrong buyer.

PART 3 — The market evidence for B2B

India corporate training is large, growing, and structurally fragmented

Metric Value
Market size 2025 USD 6,180M
Projected 2031 USD 12,910M
CAGR 13.1%
Paid learner seats (2025) 39.6M → 72.4M by 2031
Average spend per seat USD 156 ≈ ₹13,600/seat/year
Total providers ~640
Share held by local players ~70%
Largest revenue pool Technical & Digital Skills

Source: Ken Research, India Corporate Training Market, Jul 2026

640 providers, 70% local, no dominant player, largest segment is exactly our subject matter. Compare to assessment-only (7.04% CAGR, ~$806M, where HackerEarth is shrinking 32%/yr) — corporate training is nearly 2× the growth rate in a market with no incumbent to displace.

Budget exists and is rising

GCCs are the single best target segment

That is, verbatim, demand for rubric-based skill verification. GCCs are cost centres with allocated L&D budget, they hire mid-senior engineers, and they buy from vendors.

Published rate cards we can price against

Format Rate
Virtual delivery ₹25,000–1,25,000/day
Expert-led session ₹30,000–1,50,000/day
Small customised group ₹10,000–50,000/day
Large enterprise initiative ₹80,000–5,00,000+
Org-wide annual, per employee ₹15,000–50,000
E-learning ₹800–1,400/user/month
Worked example 50 managers × 3 days @ ~₹30,000 = ₹15,00,000 per engagement

Sources: KnowledgeHut, 27 May 2026 · Augusta Talent, Apr 2024. Both vendor-published; treat as directional and validate on the first three sales calls.

Colleges: real budget, concentrated in the top ~200

  • Published rate card: ₹3,000/student for a 20-hour workshop at college premises → 200 students = ₹6,00,000 (Perfect Soft Skills)
  • Campus placement platforms: Superset ₹50,000–2,00,000/month per institution (₹6–24L/yr), explicitly targeting "top 200 colleges with dedicated placement budgets" (PlacementPilot, May 2026)
  • Retail assessment pricing: AMCAT ₹1,298 all-in, CoCubes ₹1,199 + GST
  • Exit comp: Accenture acquired TalentSprint for ₹245 Cr, 22 Apr 2025 (NSE disclosure)

⚠️ Colleges are slow-paying, seasonal and tender-driven — good revenue, poor fit for limited hours. Corporate first, colleges second.

PART 4 — Unit economics and the ₹50,000 price floor

What it takes to reach ₹1L/month

Path Anchor price Volume for ₹1L/mo Volume for ₹5L/mo
Corporate/GCC engagement ₹50,000–1,25,000/day 2 days/month 5–8 days/month
Enterprise seat licensing ₹13,600/seat/yr 90 seats 440 seats
High-ticket cohort/program ₹1,64,000 0.7 seats/month 3 seats/month
College engagement ₹3,000/student 34 students/month one 350-student college/quarter
Mid-ticket self-paced ₹12,900 8 sales/month 39 sales/month
Marketplace course ₹480 net/sale 208 sales/month 1,042 sales/month ❌

Two billable days a month clears the floor. That is the entire argument.

The ₹50,000 price floor is non-negotiable

Every Indian operator making real money sits at or above it: Kraftshala ₹1.64L · ChaiCode ₹1.8–2.1L · Masters' Union ₹29.38L · Interview Kickstart (price hidden behind a sales call). The ₹4,000–13,000 tier works only behind a years-old free-content funnel — Codebasics needed 23,000 learners at ₹12,900, ByteByteGo needed 1M subscribers. We do not have years of evenings.

🚫 Paid advertising is switched off

At CAC ₹1,500–4,000 against a ₹5,000 product on which we keep 20%, small Google/Meta/Instagram budgets are a pure leak. Paid acquisition becomes viable at exactly two thresholds: AOV ≥ ₹40,000, or retargeting warm traffic. Neither is true today.

Approved channels: LinkedIn outreach, WhatsApp, warm intros, founder network. Cost ₹0, matches available hours, targets institutions.

Also priced in

  • 18% GST on every rupee of private training revenue (SAC 9992). GST 2.0 declined relief.
  • Mentor review SLA is a real COGS line, not free.
  • Refund reserve and dispute handling eat the take rate.

PART 5 — What this decision un-blocks

We no longer need a video/live/labs vendor to reach first revenue.

This retires the single largest structural risk in the codebase: a stated differentiator depending entirely on Fermion, a vendor that was acquired by Testpress mid-transition, whose signup is paused, which we have no API key for, and which our own vendor register grades B− / roadmap C (temp/elluminar-prompts/00-vendor-risk/DECISION.md).

Async delivery — content + assignment + rubric-scored human review + credential — needs none of it.

Directly de-blocked: #52, #53, #59, #60, #61, #62, #66.

PART 6 — Cold-start plan

A marketplace needs three sides at once. We are not going to solve that. We are going to hand-build one narrow supply bench and sell to institutions that bring their own demand.

Supply — target ~5, maximum 10. Practising senior engineers who can review async against a rubric, paid 55% via the existing ledger. Recruited by direct LinkedIn message. No self-serve creator onboarding push. Existing flows (applyAsCreator, /mentor apply → /admin/mentors vetting) already support this.

Demand — target 3 logos. One conversation with a company or college delivers 20–300 learners at once. That is the only acquisition motion that fits ₹0 budget and evenings.

Pick one vertical and go deep. Ranked by evidence of budget, not interest:

  1. GCC engineering upskilling — strongest evidence (2,100+ centres, stated talent problem, allocated budget)
  2. AI/ML for working engineers — 76.6% of orgs prioritising it; crowded, so verticalise by role or industry
  3. Semiconductor / VLSI specialisations — NITI Aayog (May 2026) explicitly identifies the gap as non-generic specialisations (process integration, advanced packaging, reliability). Government-backed money, too small for the giants. ⚠️ Course-fee data unavailable — price by calling ChipEdge / VLSI Guru / Cranes Varsity directly.
  4. Tier-2/3 college placement prep — real gap (incumbents priced at ₹2L/month exclude them), but slow-paying

Sequence: 5 mentors → 1 pilot logo (free or discounted, in exchange for a public case study and a named reference) → 2 paid → raise price.

PART 7 — 90-day gate and kill criteria

Sized for side-project hours. Gate date: 2026-11-25.

PASS if ANY of these is true

  • ≥1 paid corporate/GCC engagement invoiced and collected (≥₹50,000)
  • ≥1 institution (college or coaching chain) on a paid seat licence
  • ≥3 paying B2B logos of any size with <8% refund rate

Standing tripwires — any one triggers an immediate stop-and-fix

  • Any rupee spent on Google/Meta/Instagram ads before AOV ≥ ₹40,000 → stop, revert to outreach
  • Mentor review SLA breach >20% for two consecutive weeks → pause all selling, fix supply
  • Refund rate >12% → freeze acquisition, re-price
  • New feature work >25% of founder hours while zero gate items are met → scope quarantine. (This tripwire exists because 8 weeks produced 110 models, 77 pages and zero customer conversations.)
  • Two consecutive months with <5 qualified B2B conversations → the motion isn't real; re-target

FAIL all three

Stop feature work entirely. Two-week diagnosis: interview everyone who said no. Then either narrow to whichever segment showed warmth, or make the deliberate call to treat elluminar as a portfolio asset and redirect energy. An honest stop beats a slow fade.

Bull triggers — any two justify real money and hours

  • A company asks to pay before the pilot ends
  • Recruiters or candidates organically share /verify links (check referrers)
  • An employer asks about API access to verification
  • A second engagement from the same logo without discounting

Precondition for the gate to function at all: #12 (PostHog analytics) must ship. Without measurement, none of these numbers can be produced, and that itself is the finding.

PART 8 — Backlog triage

⚠️ Recorded here only. No issue is closed or relabelled until this is separately approved.

🅿️ PARK — vendor-dependent, no longer on the critical path

#52 #53 #59 #60 #61 #62 #66 — Fermion video/DRM, live sessions, Stream chat, sandbox/judge, vendor gate.
Rationale: async B2B delivery needs none of these. Revisit only if a paying customer demands live delivery in writing.

⬆️ PROMOTE — on the critical path to first B2B revenue

Issue Why it matters now
#36 Legal & compliance / entity Cannot invoice a company without a registered entity and a real grievance contact. Hard blocker.
#38 Payments & money (Razorpay live) Cannot collect. Hard blocker.
#46 B2B tax invoices for enterprise contracts Companies will not pay without a compliant GST invoice.
#12 PostHog analytics Without it the 90-day gate cannot be evaluated.
#37 Real content & brand Fabricated mentors/testimonials/stats on the landing page are a CCPA liability the moment a real buyer arrives.
#63 Enterprise SSO E2E First real enterprise buyer will ask.
#48 Revoke credit-pool instances on seat revoke Correctness bug in the seat-licensing path we are about to sell.
#35 Security & auth hardening Enterprise security review will surface it.

⬇️ DEMOTE to post-mvp — consumer-marketplace and growth mechanics

#1 #2 #3 #4 #5 #7 #9 #13 #15 #16 #19 #20 #21 #22 #25 #26 #27 #29
(Dodo Payments MoR, project tiers, AI defense engine, AI tutor wallet, AI review, staged free funnel, EMI financing, custom-domain storefronts, pods/XP/leaderboards, search & upsells, mock-interview marketplace, proctoring/plagiarism suite, PPP multi-currency, gifting, ISA/success-fee, digital products, scholarship engine, mobile app.)

⚠️ Note on #25 (income-share / success fee): the evidence is now decisive — every Indian ISA operator abandoned the model. Masai's pay-after-placement "crashed in the hiring freeze"; every survivor moved to prepaid. Do not revive this.

➡️ LEAVE OPEN, unchanged

#39 #40 #41 #42 #45 #47 #49 #50 #51 #58 #64 #65 #68

🗑️ Housekeeping

Remove shipped scaffolding: /sentry-example-page (237 lines) and /api/sentry-example-api. Note that showAllSurfaces() (src/lib/deploy-context.ts) unlocks every dashboard for any signed-in user on Netlify preview/branch deploys — intentional, but it means preview demos overstate access control; do not demo a prospect on a branch deploy.

PART 9 — Corrections to prior assumptions

9.1 "The product is the credential" — needs amending

temp/feedback/01 argued the entire product should be verified project proof. The strongest available evidence says employers say but do not do:

Harvard Business School + Burning Glass Institute, "Skills-Based Hiring: The Long Road from Pronouncements to Practice" (Feb 2024) studied 11,300 roles at large firms before and after degree-requirement removal. Net effect: ~97,000 workers out of 77 million annual hires ≈ 0.14%. Fewer than 1 in 700. (Burning Glass Institute)

Every "87% of employers value micro-credentials" statistic traces to credentialing vendors' own surveys. Selling verification standalone means fighting Mercer|Mettl (₹210 Cr) and Info Edge's DoSelect (₹57.5 Cr, +32.7%) in a 7%-CAGR market where HackerEarth is shrinking 32%/yr — with an enterprise procurement and security-review sales cycle, from a side project.

Amended position: the credential is a pricing lever, not a wedge. "We train your 300 engineers and give you a rubric-scored, independently verifiable record of who can actually do the work" is what moves a day rate from ₹25,000 to ₹1,25,000. That is a 5× on the same evening. Sell it bundled. Never alone.

⚠️ ONEST / ONDC-for-education is not a distribution channel to plan around — every substantive source dates from 2023, with no 2025–26 participant or transaction data. NCrF / Academic Bank of Credits is real infrastructure but has no demonstrated employer pull yet.

9.2 "Competitors like upGrad, Edureka, Masai, GrowthSchool are successful" — mostly false

Of every named company, exactly one has audited positive PAT.

Company Latest audited Revenue Bottom line
upGrad FY26 ₹2,070 Cr (+7%) net loss ₹130 Cr
Unacademy FY25 ₹826 Cr (−16%) loss ₹436 Cr → sold all-stock
Simplilearn FY25 ₹556 Cr (−26%) loss ₹43 Cr
Scaler FY25 ₹363 Cr (−5.5%) loss ₹2.3 Cr; ₹13 Cr cash vs ₹365 Cr expenses
Masai FY25 ₹63.8 Cr audited loss ₹31.9 Cr
Kraftshala FY25 ₹10.1 Cr (−19.9%) loss ₹1.8 Cr
Great Learning FY25 ₹1,039 Cr (+4.7%) +₹40 Cr operating; PAT undisclosed; in creditor hands
Eruditus FY24 ₹3,733 Cr adj. EBITDA −₹69 Cr; never published a PAT line; FY25 unfiled 17 months on
Jaro Education FY26 ₹273.9 Cr (+8.6%) PAT ₹52.9 Cr (19%), ROCE ~40%

Three factual corrections:

  • Edureka was not acquired by Great Learning. Veranda Learning bought it for ₹245 Cr (Sept 2021), then divested it in Oct 2025 into a JV with SNVA EduTech — an entity with ₹3.33 Cr of FY25 revenue — booking a ₹133 Cr disposal gain. That is a distressed disposal.
  • GrowthSchool raised $5M in Jan 2022 and never raised again. Zero financial press in 30 months; Tracxn still lists it seed-stage. No revenue, PAT or take-rate figure has ever been published.
  • "Masai School of Business" does not exist (masaischool.com/school-of-business → 404). It conflates Masai School with Masters' Union (₹29.38 lakh fee; financials never published by any Indian business publication — online estimates span 30× and are worthless).

⚠️ Watch for the collections-vs-revenue trick. Masai's widely-quoted "₹100 Cr" is collections; audited revenue from operations is ₹63.8 Cr. upGrad circulated an 11-month provisional PAT of +₹38.8 Cr during a fundraise; the audited full year was a ₹130 Cr loss. Vedantu announced "profitable Q4" on a collections metric while filing a ₹210 Cr annual loss. Do not benchmark against company-claimed numbers.

Also worth noting: upGrad's May 2026 round at a $1.73B valuation was 83% funded by the founder personally (₹300 Cr of ₹361 Cr). That is not external validation.

9.3 The pattern: profitable Indian edtech shares one trait — the buyer has no choice

Model Example Result
Education lending Avanse FY24 ₹1,726 Cr revenue, ₹332.5 Cr PAT (19%)
Commission on someone else's degree Jaro FY26 ₹273.9 Cr, ₹52.9 Cr PAT — no content, no faculty, no credit risk
One high-stakes exam, captive demand DailyRounds/Marrow FY24 ₹568 Cr revenue, ₹320 Cr PAT (56%), ₹0.40 spent per ₹1 earned
Offline coaching + real estate + brand Allen FY25 profitable, but PAT −70% and revenue declining
University-partnered B2B at scale Great Learning FY25 thin operating profit after a brutal reset

Loss-making at every single scale observed: ISA / pay-after-placement, cohort marketplaces, self-paced video libraries, K-12 tutoring, generalist upskilling.

That list was our roadmap. Not because it was built badly — the engineering is genuinely strong — but because it was aimed at buyers with infinite free alternatives.

Indian edtech funding 2025: $118M–$249M total across 22–31 deals (methodology-dependent), an 8-year low, ~87% off the 2021 peak. Assume no institutional capital. Plan to be default-alive.

9.4 David vs Goliath?

No — and that is the point. The Goliaths are losing money (PW FY24 loss ₹1,131 Cr; upGrad cumulative losses over ₹2,000 Cr on $766M raised) and consolidating defensively (Coursera+Udemy at $2.5B combined, then layoffs).

They structurally ignore:

  1. Anything with fewer than ~50,000 addressable buyers/year — a ₹2–5 Cr business is a rounding error to them and a career for us
  2. Anything requiring genuine expert judgement per learner — rubric-reviewed projects do not scale to 100,000 students, which is exactly why the moat holds
  3. B2B/GCC work with custom scope — they sell catalogue seats; GCCs want specific-stack, specific-team engagements. 640 fragmented providers holding 70% share is this fact in numbers.
  4. Regulated/professional niches — Academically Global does ₹30 Cr bootstrapped in international healthcare licensure while PW loses ₹1,131 Cr
  5. Verification bundled with training — Mettl/DoSelect do the assessment half; nobody sells trusted proof-of-capability inside the training transaction

We do not need to beat them. We need to be unattractively small to them and indispensable to thirty buyers.

Sources

Codebase: full audit of Practitionist/elluminar_web @ b6543d6, 2026-08-27.
Prior review: temp/feedback/00–05 (2026-08-26, gitignored) — directionally right on narrowing; §9.1 above amends its credential framing.
Market research: ~190 searches, 2026-08-27. Primary sources prioritised — SEC filings, company IR, MCA-derived filings via Entrackr/Inc42, Screener.in, NSE disclosures, HBS/Burning Glass, Ken Research, TechCrunch/Axios/Business Standard.

⚠️ Declared data gaps — do not treat as known

  • GrowthSchool audited financials — none published, ever
  • Masters' Union revenue and PAT — never reported; online estimates span 30×
  • Eruditus FY25/FY26 — unfiled 17 months past year-end; has never disclosed PAT
  • Edureka standalone revenue/PAT — never segment-reported
  • Contracted institutional rates colleges pay training vendors — only retail prices found
  • VLSI/semiconductor course fees in India — must be obtained by phone
  • Mettl acquisition price — genuinely undisclosed; ignore the ~₹300 Cr figure circulating
  • Free NCERT-solution traffic → paid conversion rate — no company anywhere discloses this; the silence is itself the finding
  • Corporate training rate cards are vendor-published; validate on the first three sales calls
  • All third-party "market size forecasts" from research vendors should be treated as directional at best

The uncomfortable summary: the engineering is not the problem and never was. Eight weeks produced 110 models, 77 pages, 40 issues and zero customer conversations. Every "should we pivot to X" question is a way of staying in the editor. The next artifact should be a signed pilot, not a merged PR.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enterpriseEnterprise & universitylaunch-windowTier 1: during beta, before public launchmarketingMarketing surfacesv1V1 (6-12 months)

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions