ClassIn Founder Maps a Route Through China's Tutoring Downturn

Junbo Song, founder and CEO of ClassIn operator EEO, used a 29 October closed-door gathering of tutoring operators in Zhengzhou to set out where he thinks growth can come from in a shrinking market. The transcript, published by Jiemodui on 3 November, names four pressures arriving at once: a falling birth pool, more cautious household spending, continuing uncertainty over the policy boundary for off-campus tutoring, and AI changing how parents judge the value of what children learn. His conclusion is that institutions must re-prove the reason families pay, rather than simply switching product category or age group.

One data point he cites shows that hardware is no refuge. According to RUNTO, sell-through of learning tablets in China fell 14.2% year on year to 2.287 million units in the first half of 2026, with the second-quarter decline widening to 26.2%; the research firm forecasts a 16.7% drop for the full year to about 5.268 million units, against 6.321 million in 2025. Revenue fell more slowly than volume — down 10.2% to RMB 7.72 billion in the first half — as average selling prices rose and the four largest vendors took roughly 80% of the market.

For demographic strategy, Song points to two contrasting neighbours. South Korea responded to low birth rates with greater intensity: denser courses and longer study hours, a path he argues Chinese providers should not copy. Japan’s operators raised prices gently while widening the service into after-school reading, self-study, activities and meals, becoming what he calls “half-schools”. If household education spending holds a roughly stable share of income, he argues, spending per child can rise — but only where the extra fee buys visibly better service. He extends the same logic to physical space, noting study-session cafés in Beijing and Shanghai, snack corners and vending machines in Korean cram schools, and bookstore-style study rooms.

On AI, the argument is about feedback rather than content generation. Song identifies three uses now practical in ClassIn: marking Chinese and English essays and homework through its AgentIn tools; AI lesson analysis that reviews classroom recordings against curriculum standards and lesson plans so principals and teachers can see what to change in the next class; and AI-graded responses on the platform’s in-class “small blackboard”, which feeds post-class learning reports even in large classes. Together they aim to make the kind of individual attention that was previously too costly to sustain routine.

He cautions that tablets alone do not create an AI-ready classroom. Capturing enough learning data requires redesigning when students answer and discuss, preparing curriculum standards, lesson plans and materials — often more work up front than before. Longer term he expects AI to generate new course formats: a composition class could extend into a novel and then an AI-assisted short film, and English lessons into shareable scenario-based work. Inside EEO, he says a product manager without a programming background used AI to build a live-streaming tool, prompting a shift to a “platform plus projects” structure; stronger tools, he notes, do not automatically multiply business results.

Two things, he concludes, become more valuable as tools improve: professional judgement about why a student is stuck, and the human attention a teacher gives a child. His advice is to put AI into one real workflow, reserve a pilot budget for new courses and let younger staff experiment alongside experienced teachers. It is a vendor’s view, but the RUNTO figures and the Japan-Korea comparison make it a concrete read of how China’s tutoring sector is repositioning around AI.