The Edtech Recovery Playbook: AI for the Coaching Industry

I haven't written much about this before. Pre AI K12 companies like BYJU'S didn't die from bad technology. It died from CAC > LTV. Here is how an Indian coaching business should actually point AI — at the three line items that killed the category, not at the demo that wins a pitch.

━━━━━━━━━━━━━━━━━━━━

BYJU'S did not collapse because its technology was weak. It collapsed because it spent ₹2 to make ₹1 — a sales-led machine that bought students faster than it could retain or graduate them, and called the gap "growth." Every coaching operator in India watched it happen, and most have drawn exactly the wrong lesson: that edtech itself was the mistake. It wasn't. The mistake was pointing capital at acquisition instead of unit economics.

So when an AI vendor walks into a NEET or JEE coaching business in FY27 with a slick "personalised learning" demo, the operator's first question should not be "is this impressive?" It should be "which of my three broken line items does this fix?" Those line items — customer acquisition cost, cost-to-serve per student, and retention-and-outcomes — are what killed the category. AI is only worth deploying where it moves one of them. This is the playbook for doing that, in order.

The Reframe: AI Is A Margin Tool, Not A Marketing Tool

━━━━━━━━━━━━━━━━━━━━

The early winner is already visible. PhysicsWallah built a profitable business at a fraction of BYJU'S burn by inverting the model — content-led acquisition, frugal cost-to-serve, hybrid offline centres for trust and conversion. It did not out-spend anyone. It out-structured them. Allen and Aakash, the offline incumbents, survived precisely because they never had the option to torch cash on Google and Meta.

That is the lens for every AI decision below. If a use case makes the demo better but leaves CAC, cost-to-serve, and outcomes untouched, it is theatre. Skip it.

Step 1 — Fix CAC Before You Touch Pedagogy

━━━━━━━━━━━━━━━━━━━━

The line item that killed BYJU'S was acquisition. By 2022 it was reportedly spending a large share of revenue on sales and marketing — a performance-marketing addiction that only works while funding is cheap. The moment capital tightened, the model imploded.

Point AI at the organic funnel first. Use it to industrialise content-led acquisition the way PW did with free YouTube lectures: AI-assisted production of doubt-solving shorts, exam-pattern explainers, and previous-year-question walkthroughs in volume, in the languages your aspirants actually search in. The goal is to make your blended CAC fall every quarter because owned content compounds, while paid spend does not. If your AI budget goes to writing better Meta ad copy, you are using AI to dig the same hole faster.

Step 2 — Make AI Doubt-Solving The Wedge

━━━━━━━━━━━━━━━━━━━━

Doubt-solving is the highest-frequency, highest-cost student interaction in Indian coaching, and the one most correlated with retention. A student who gets a doubt resolved at 11 PM the night before a test stays. One who waits two days for a human mentor churns.

This is the single best place to deploy AI, because it hits two line items at once: it slashes cost-to-serve (a human doubt-mentor handling 40 queries a day is replaced at the margin by an AI handling thousands) and it lifts retention. Build it vernacular-first — a Hindi and regional-language doubt solver is worth more in a Kota-feeder town than an English one in a metro. Keep humans in the loop for the hard 10% and for escalation, so quality holds. This is your wedge: narrow, measurable, defensible.

Step 3 — Use AI For Teacher Leverage, Not Teacher Replacement

━━━━━━━━━━━━━━━━━━━━

The scarce asset in Indian coaching is a great teacher, not technology. Star faculty are why students pick Allen over a no-name local institute, and why teacher poaching is a real line item. The growth-at-all-costs edtechs tried to make the platform the star. It doesn't work — Indian parents trust a named teacher with a track record, not a brand.

So use AI to scale the teacher you already have, not to replace them. Auto-generate practice sets calibrated to each student's weak chapters. Triage which doubts actually need the star versus which an AI handles. Grade descriptive answers at draft quality so the teacher edits instead of marks from scratch. The arithmetic: if AI lets one great faculty member effectively serve 1.5× the students at constant quality, you have just cut your most expensive cost-to-serve line without cutting the thing students pay for.

Step 4 — Instrument Outcomes, Then Price On Them

━━━━━━━━━━━━━━━━━━━━

The deepest wound BYJU'S left is a trust deficit. Parents in 2026 have read the headlines and assume edtech over-promises. The antidote is not a louder promise — it is a demonstrated outcome.

Use AI to predict at-risk students weeks before a mock-test crash, not after. Behavioural signals — attendance, doubt frequency, practice completion, sleep-time study spikes — let a model flag the student drifting toward a drop-out or a rank collapse, in time for a mentor to intervene. Then prove it: a parent dashboard that shows percentile movement against the cohort. Once you can demonstrate outcome, you can price on it. The per-outcome logic from Pricing AI for Indian Buyers applies directly here — a fee partly tied to rank improvement is a far easier sell to a burned parent than a flat annual package, and it forces your own model honest.

Step 5 — Go Hybrid-Smart, And Let AI Pick The Geography

━━━━━━━━━━━━━━━━━━━━

Pure-online died. Pure-offline doesn't scale past a few cities. The PW answer — hybrid Vidyapeeth centres bolted onto a strong online layer — is the model the market has voted for, because the offline centre solves the two things online never could: parental trust and conversion.

AI's job here is the layer logic and the site selection. Run the online doubt-solving, practice, and analytics at near-zero marginal cost; reserve scarce offline capacity for the high-trust moments — counselling, doubt-clearing intensives, parent meetings. Use AI on your enquiry and search-demand data to decide which tier-2 town gets the next centre, instead of betting on a founder's hometown. Capital discipline is the whole lesson of the last cycle; geography is where coaching businesses still waste it.

Step 6 — Treat Vernacular As The Growth Frontier, Not A Feature

━━━━━━━━━━━━━━━━━━━━

The English-first edtechs left the biggest TAM in Indian education on the table: the tier-2 and tier-3 NEET, JEE, and UPSC aspirant who studies in Hindi, Marathi, Tamil, or Telugu. Producing genuinely good content in five languages was uneconomic when every asset needed a separate human team. AI changes that cost curve.

AI-assisted localisation — not raw translation, but idiom-aware, example-localised adaptation — lets one strong content engine serve many language markets at a fraction of the old cost. This is where a sharp operator can build a moat the incumbents are structurally slow to copy. It also compounds with Step 1: vernacular organic content is where CAC is lowest, because the competition for that search demand is thinnest.

Step 7 — Do Not Rebuild The BYJU'S Sales Machine With AI

━━━━━━━━━━━━━━━━━━━━

This is the guardrail, and it matters most because it is the most tempting. The fastest way to show a board a revenue spike is to point AI at telesales — auto-dialling, AI call agents, aggressive nurture sequences pressuring parents toward a loan-funded enrolment. It will work for two quarters. It is also the exact relapse that killed the category, now automated and cheaper to scale, which makes it more dangerous, not less.

Keep AI on the product and outcome side of the house. Let acquisition stay content-led and trust-led. The discipline to NOT automate the thing that produces a short-term number is what separates the operators who survive this cycle from the ones who become next year's cautionary slide.

The Strategic Read

━━━━━━━━━━━━━━━━━━━━

The post-BYJU'S market does not reward the most advanced AI. It rewards the most disciplined deployment of it. The three line items are the scoreboard: is your CAC compounding downward, is your cost-to-serve per student falling at constant quality, and can you prove an outcome to a sceptical parent? Every AI rupee should be traceable to one of those three. The vendor demos that don't map to them are not your recovery — they are how you repeat the mistake with better graphics.

BYJU'S PROVED YOU CAN LOSE ₹22,000 CRORE OF VALUE WITH THE BEST TECHNOLOGY IN THE MARKET. THE RECOVERY PLAYBOOK ISN'T MORE TECHNOLOGY — IT'S POINTING IT, FINALLY, AT THE LINE ITEMS THAT PAY THE BILLS.

━━━━━━━━━━━━━━━━━━━━

#EdTech #AIinEducation #IndianStartups #SectorPlaybooks #FY27