A 12k won "24-hour card": the Korean TOEIC speaking AI trainer 토스트, and how it earns a second slice on a path someone else already proved

A small Korean team of 9 people launched a TOEIC speaking AI scoring app in mid-August. It sells unlimited non-renewing day passes, and within two months it squeezed into the top 29 of Korea's Education top-grossing chart. It didn't invent anything: it copied a model a rival had already reached break-even on a year earlier, priced 30–40% lower, and added a team that knows how to run paid acquisition. Below: who it sells to, how the money comes in, who it's competing with, and where this playbook can be moved.

"₩60k a month... hmm"

2026 Oct 7, someone on the Korean App Store left a review for 토스트 (TOAST) titled "The app itself is good". The body is one sentence: "The app itself is good, but I didn't know I'd have to pay. ₩60K a month… hmm…" They gave it 5 stars.

That review is basically the whole business: it's too expensive, but they still gave a good rating, and they probably paid anyway. The reason is another number: a TOEIC Speaking test costs ₩84,000 to register in Korea (Hackers blog, 2026). Test takers are mostly job seekers, graduates and working people who need a score. The reviews keep saying things like "my company needs my TOEIC Speaking score urgently", "I need it to graduate", "the exam is one week away". Compared with "taking it again", a ₩60K four-week card doesn't sound expensive.

What it is

A TOEIC Speaking practice app listed only in the Korean store, made by Apeiro Corp. (아페로, Seongnam, about 9 people, raised a seed round in 2023 December). The flow is:

1. a free 10-minute diagnostic test that estimates scores per section and shows weak spots;
2. enter a target level (IM3 / IH / AL) and exam date, and it schedules daily practice automatically;
3. 200 answer templates for each question type, plus real-time AI correction of pronunciation and intonation while you shadow;
4. 10 mock tests in the same format as the real exam (11 questions, timed), with AI scoring and an explanation of "why points were deducted".

The store description is blunt: "Latest speech recognition + LLM scoring — not just a score, but why you lost points." Until recently this kind of scoring was only affordable for companies like Riiid with tens of thousands of real answer sheets. Now a small team can ship it in a month or two.

How it charges: no subscription, it sells the days before the exam

TOAST doesn't do auto-renewing monthly subscriptions. It sells day-based membership cards (Korean App Store in-app purchase list, 11-06):

The diagnostic exam is free, but you hit a paywall right after your first mock test. A 28-star review from Aug 2 said: "The diagnostic felt great, so I figured the app was worth it. Then the first mock test ended the free access, and everything else was paid. I saw the price and deleted it." The paywall comes early, and for people cramming one or two weeks before the exam, a 24-hour pass is an impulse buy.

Revenue isn't public. What we can see: 11-06 it ranks #29 in Korea's education Top Grossing chart, and the same company's vocabulary app "26초" ranks #30. 26초 has 6,944 ratings, TOAST has only 439. The user count is an order of magnitude smaller but the chart rank is about the same, which suggests each user pays a lot (inference).

Why now

• Someone already proved it makes money. TOSME (토스미) launched in 2024 September, backed by the offline cram school 티처K. It follows the same "free diagnostic → AI mock test → sell 24-hour / 3-day / 7-day unlimited cards" model. Press reports say it reached 2k users within 30 months, 8k learners after 50 months, and broke even (머니투데이 2025-02; 전자신문 2025-09). TOAST's in-app purchase structure is almost identical: its 24-hour card is 12,000 versus TOSME's 19,600, about 30–40% cheaper.
• The team already knows how to acquire users. Apeiro's 26초 got 500k users a year after launch and was profitable for 6 consecutive months (Platum, 2026-05-20). Their job postings say their playbook is "Meta ad creatives to drive installs + partnering with Instagram influencers". People who study TOEIC vocabulary and people who take the TOEIC Speaking test are the same audience.
• It caught the hiring season. Launched in Aug 18, right before Korea's 9–October second-half public recruitment, when the most people need language scores.
• The official side hasn't come in. ETS launched an official AI practice app, "TOEIC Pal", globally in 2025 December, but it can't be found on the Korean App Store (TOEIC in Korea is administered by YBM). The local app doesn't have to go head to head with the official one.

Where users come from

The publicly visible channels are all small:

• The Instagram campus account "대학생활 모음집" co-posted with the official account @tos_trainer in Sep 27 a "TOEIC Speaking in two weeks" post, with 14 likes. The official account itself has 265 followers and 5 posts.
• Some alt accounts on YouTube posted a few Shorts tagging it, each with single-digit likes.
• Store search: the title is packed with keywords "토익스피킹 AI 채점 트레이너, 모의고사". Searching "토익스피킹 AI 앱" on Google, its Google Play page is the first result.
• Meta ads: following the team's own playbook, they're probably running paid ads (inference), but no Korean ads for it turned up in the ad library.
• Review farming: on August 20–27 there was a run of 5-star long reviews with the same tone ("used it for about a week, only praise," "self-discipline tool for homebodies"). Later written reviews were mostly 2–3 stars, yet the overall score is 4.9 (442 ratings, 11-06). The Google Play listing shows "1천+" downloads but 213 reviews. The app is nudging users to rate, which is almost certain; whether the first batch of reviews was bought can't be confirmed.

No media has covered it, and nobody has mentioned it on X in nearly 60 days. Its rating count rises steadily by 3–11 a day (10-28 at 385 to 11-06 at 439), with no spikes from a chart push. It looks more like steady, ongoing spend than a single piece of content going viral (inference).

What users are complaining about

The complaints in the written reviews are concentrated, and this is where a later entrant could do better:

• Scoring is too rigid. "I used the expressions taught in every YouTube course and it still marked me wrong. For example Part 5: Most of all (×) → First of all (○)… After five mock tests my confidence dropped, so I stopped using it." (10-10,2 stars, paid)
• It often fails to score. "I bought the membership and it worked well, but it keeps failing to read my answer for scoring, and it just ends after I speak." (9-18) "It lags badly, why does it keep saying it can't score?" (9-14)
• Questions repeat. "Part 4 has all kinds of scenario tables, and this only has the workshop schedule." (9-13) "The mock test question types cycle through the same few, and the vocabulary repeats too." (9-05)

Who's competing

The category isn't empty. TOAST itself is a follower that came almost two years late, winning on price and the team's existing know-how, not on a product generation gap.

If it were you

Verdict: another TOEIC Speaking app in Korea is too crowded. But the playbook of "urgent need before the exam + instant AI scoring + selling day-based cards" is a clear opportunity if you move it to other speaking exams and other markets.

Who it fits: small teams that can make Meta / TikTok ad creatives and understand the exam culture of a particular country. Ideally you already have users for the same exam (from vocabulary or question-bank apps).

Not for whom: people who can build a product but can't run ads. Nobody in this category spreads it organically. There are 0 posts about it on X and the official IG account has 265 followers, so growth comes almost entirely from paid acquisition and store search.

How to do it:

• Pick a speaking exam where the score decides your future, the test is expensive to retake, and there's no good official or local app: OPIc, Japan's Eiken speaking test / TOEIC S&W, IELTS speaking in a specific country, or HSK speaking (HSKK). Check the local App Store first to see whether there's a version like TOSME already on the grossing chart, proving people pay.
• Copy the pricing structure: a free 10-minute diagnostic that estimates scores → a plan scheduled to the exam date → non-renewing 24-hour / 3-day / 1-week / 4-week cards. Users compare it to the cost of retaking the test, so you don't have to sell them on subscriptions.
• Treat competitors' bad reviews as a feature list: be lenient about common acceptable phrasings, make scoring reliable, and prepare enough varied questions for each question type.
• Put the paywall after the first mock test score, not before.

How to acquire users: run Meta / Instagram ads with "X days until the exam" creative, plus partner posts with campus-life and exam-prep accounts. Pack the store title and description with exam names and search terms like "AI 채점" and "모의고사". After launch, prompt for ratings inside the app and build up the rating count early.

When to get in: follow the exam calendar. In Korea that's the second-half hiring season (9–October) and the first half (3–April). In another market, launch one to two months before that exam's registration peak; if you miss it, wait for the next cycle.

What would mean you got it wrong: no similar app in the local App Store grossing chart is making money for the exam you picked; or after a month of paid acquisition, the diagnostic-to-paid conversion can't cover install costs; or users' main complaint becomes "scoring isn't accurate" and your model can't fix it. In this category, if the scores aren't accurate, nobody buys a second card.