2. Market Reality Check
Figures with a source were checked on 29 Sept 2026. Sources and confidence levels are in research/10. Anything marked (estimate) is analyst judgment. Currency: ₹88 = $1 (assumed).
2.1 Who exactly is the buyer?
| Segment | Frequency of need | Willingness to pay | Pain (1–10) | Verdict |
|---|---|---|---|---|
| Hindi- or Bengali-medium job seekers aged 18–28 (graduates, diploma holders, Tier 2–3 towns) preparing for interviews, BPO, retail, hospitality or office roles | Daily while job-hunting | Proven but low. SpeakX: ~200k payers at ₹299/mo, 70% Hindi speakers (Oct 2025). Josh Skills: 1 lakh payers at ₹99–899 (2020). | 8 (a job depends on it) | Core buyer. Has a deadline, pays per month, and responds to short-video content. |
| Early-career staff in customer-facing or office roles who "freeze" in English (calls, meetings, emails read aloud) | Weekly | Medium. Some self-pay; employers sometimes pay. | 7 | Second core. Same product, and also reachable through employers and institutes. |
| Institutes and employers: skilling centres, hotel-management and retail-training institutes, BPO recruiters and trainers | Per batch or cohort | Per-seat budgets exist (e.g. EngVarta quotes ₹2,700 per employee for 25 short sessions; Versant tests cost ₹1,100 each) | 7 | The value multiplier from month 9. 100 seats is worth ~80 individual payers. |
| Casual learners ("I want to chat better in English") | Occasional | Low. Duolingo is free. | 3 | A free warm-up audience, not a paid segment. |
| Bengali-speaking learners in Bangladesh | Daily for students and job seekers | Very low per user. 10 Minute School sells a spoken-English course for Tk 450 one-time (~192k enrolments). | 7 | Later, by demand. The price level is a problem. |
| Diaspora parents wanting children to learn Hindi or Bengali (heritage) | Weekly | High, in USD (tutors $3–40 a lesson; 50 Bengali lessons for $299) | 5 | An alternative version (see 01). Children bring COPPA and DPDP duties, the audience is small, and demand is unproven. |
| Domestic migrants learning the local language (e.g. Bengali in Kolkata, Kannada in Bengaluru) | Daily in the first months | Low. Kannada WhatsApp courses cost ₹200/mo (~2,800 students); classes cost ₹450/hour. | 5 | Real but small, and scattered across many language pairs. Not first. |
| Foreigners in India learning Hindi | Occasional | Medium | 4 | Too small (hundreds of thousands, spread out). |
Pain score for the product as described: 5/10. High for job seekers facing interviews, low for casual learners. The design should serve the first group and let the second enter through free scenes.
Why the problem statement needs a fix: "even at level 2, the lesson starts from 0" is not true of the leaders. Duolingo has an adaptive placement test and "jump here" tests, and Babbel and Busuu have level checks. What is true:
- placement measures reading and listening, not speaking;
- most AI-built courses stop at A1–A2;
- there's little practice for the specific, high-stakes situations a learner actually faces.
The pitch should be "placed by how you speak, practising the situations you're actually heading into".
2.2 Competitors and incumbents
Duolingo: the free alternative your users already have
| Fact | Detail |
|---|---|
| Scale | 58.7M DAU, 140.6M MAU and 12.7M paid subscribers (Q2 2026). FY2026 revenue guidance is about $1.21B. |
| Strategy | It is putting user growth ahead of revenue, deliberately giving up $50M+ of bookings in 2026, and moving AI features into cheaper tiers. It targets 100M DAU by 2028. |
| Scenes | Adventures (2024): game-like scenes where you walk around and complete tasks such as ordering coffee or getting through passport control. Speaking Adventures (2026): scenario tasks completed by speaking, available to free users. |
| AI conversation | Video Call with the character Lily is moving into Super, which has about 10x as many subscribers as Max. The cost per call fell from about $0.30 to under $0.01 using open models. It remembers facts about the learner. |
| India | One of Duolingo's top 5 markets. It teaches English from Hindi and from Bengali (the Bangla course has 6M+ learners, per press). The CEO called Max's ~$70/year India price "too expensive". |
| Gaps | No Bengali course to learn. Placement is not by speaking. Courses are mostly A1–A2. There are no work-specific scenes (interview, customer call). |
Implication: anything that looks like "Duolingo with a city map" loses. You win only on specificity: work situations, speaking-first placement, explanations in the learner's language, and visible speaking progress.
AI speaking apps (global)
| Player | Status | Notes |
|---|---|---|
| Speak | $1B valuation (Dec 2024). $100M+ annualised revenue (Nov 2025). | Strongest in Korea and Japan. $80–200/year. |
| Praktika | $35.5M Series A (2024). ~$20M revenue at the time. | AI avatars. Global. |
| Loora | $22M Series B (16 Sep 2026). Claims "tens of millions" of ARR. | An English-only AI tutor with 100+ B2B customers. |
| ELSA Speak | ~$60M raised. 50M+ downloads. India a named focus. | Pronunciation-first. |
| Babbel (Babbel Speak), Busuu, Memrise (Podchats) | All added AI voice conversation in 2025–26 | AI conversation is now a standard feature. |
| Langua, Gliglish, Talkpal | Small. Langua ~$3.5M ARR (bootstrapped). Gliglish ~€8k MRR (solo founder). | These show that a focused small player can earn $1k–$300k MRR. |
| Preply, Cambly | Human-tutor marketplaces (Preply valued at $1.2B, Jan 2026) | The expensive alternative. Cambly in India costs ₹1,199–8,499/mo. |
Spoken-English apps built for India: your direct competitors
| Player | Scale | Model | Why a user would or wouldn't switch |
|---|---|---|---|
| SpeakX | ~200k payers, ~$7.5M ARR, EBITDA-positive, $16M raised (Oct 2025) | AI speaking practice, ₹299/mo via UPI Autopay. 70% Hindi speakers. Bengali, Telugu, Tamil and Marathi announced. | The closest competitor. Your edge has to be work scenes, speaking placement, Bengali-first quality, and institute seats. |
| Stimuler | $3.75M from Lightspeed (Apr 2025). Claims 4M+ users. | AI conversation, ₹199/mo or ₹999/yr, 5 free conversations a day | Generous free tier. Tests-oriented (IELTS). |
| Josh Talks / Josh Skills | ~₹20 Cr company revenue (FY25) | Vernacular courses, ₹99–899 | Course-based, not practice-based. |
| EnglishYaari | Claims ₹5 Cr ARR (paid press release, Dec 2025) | Hindi-belt spoken English | Shows local demand. Low confidence. |
| Seekho, Kuku FM (Guru) | Seekho FY25 revenue ₹142 Cr (ad spend ₹134 Cr). Kuku ₹242 Cr. | Broad "edutainment" subscriptions that include English | Growth is bought with ad spend close to revenue. Don't copy the playbook. |
| PlanetSpark | FY25 revenue ~₹81 Cr | Live 1:1 classes for children | A different, sales-heavy business. Avoid. |
| Hello English, Enguru | Hello English appears inactive. Enguru was acquired by Open English (2022). | Early gamified English apps | Early free, gamified English apps didn't last as stand-alone businesses. |
Scene-based and game language products
| Product | Outcome |
|---|---|
| Wagotabi (Japanese RPG, 2025) | ~$350k gross (estimate). The best of the genre. |
| Influent, Lingotopia, Noun Town VR | Small Kickstarter or indie results. Lingotopia has 34 reviews. |
| Mondly VR/AR | Now part of Pearson. No separate figures. |
Pattern: the scene or world mechanic has only been monetised at scale inside Duolingo, as a feature. Stand-alone "language games" are small.
General assistants
ChatGPT voice, Gemini Live and similar tools will role-play an interview for free. They don't place learners, don't track progress, don't use a fixed rubric, and don't explain in Bengali unless asked. Structure, progress and feedback are the edge over them, not the voice itself.
2.3 Vitamin or painkiller?
For casual learners, it's a vitamin. Skip a day and nothing happens.
For a job seeker with an interview in two weeks, it's a painkiller they already pay for (SpeakX's payers, Josh Skills' payers, coaching classes). The test: if the app disappeared, would they lose something they can't get elsewhere? Only if it holds their speaking level, their progress recordings, their scene history and their weekly score. Visible speaking progress is what turns the vitamin into a painkiller.
2.4 Market size: venture-scale, lifestyle, or a feature?
- The published market sizes are unreliable. Global "language-learning app" estimates range from $7B to $25B+, depending on definition. App-store revenue for the category was about $1.5B in 2025, most of it Duolingo's (Business of Apps; low confidence).
- As a business for you, it's lifestyle-scale. $10k MRR needs about 4,000 payers at ₹199 (or fewer with institute seats). SpeakX alone has about 200k. Market size isn't the constraint; distribution cost, renewal and voice cost per learner are.
- "AI conversation practice" is now a feature of every major app. Work-situation practice for Indian-language-medium learners, with measurable speaking progress, is the gap. It's real but narrow, and SpeakX is moving towards it.
- Venture-scale would mean competing with SpeakX and Seekho for the mass market on ad spend. That isn't enterable at this stage.
2.5 Geography
- India first.
- Hindi-medium learners are the larger group, and more contested.
- Bengali-medium learners are less served (West Bengal plus Bengali speakers elsewhere).
- Payments via UPI Autopay (53% of recurring payments by Jan 2025) and Razorpay.
- Bangladesh later, only by demand. It has a large Bengali-speaking market, but local price levels (Tk 450 one-time courses) make a subscription hard.
- Diaspora heritage learning: not planned unless inbound demand appears. It involves children, is small, and is unproven.
- US, UK or EU users: not targeted. A US launch would bring voice-biometric law (BIPA, Washington, Texas) and companion-chatbot law into play without much revenue in return.