---
title: "A Textbook Publisher That Beat Big Tech in AI Education — How?"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-07-14T11:30:52+00:00"
canonical: "https://ffcap.cn/en/research/src-20260714-02html"
source: "https://uniqueresearch.substack.com/p/src-20260714-02html"
language: "en"
---

# A Textbook Publisher That Beat Big Tech in AI Education — How?

[![](https://substackcdn.com/image/fetch/$s_!X-D_!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2b613643-4bfd-4ea4-9d54-b97873d4d461_2048x1152.png)](https://substackcdn.com/image/fetch/$s_!X-D_!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2b613643-4bfd-4ea4-9d54-b97873d4d461_2048x1152.png)

_Original · Unique Research · 2026-07-14_

_Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the Century Tianhong / XiaoHong Zhujiao narrative, the themed sections, and the complete 10-question Q&A with Zhang Minsong. All named companies, products, and figures are preserved. Company and interviewee statements are source attributions, not independently verified findings._

AI Industry Observation

A textbook publisher that beat big tech in AI education — how?

"The real battlefield of education AI was never technology, but content and scenario."

508 million in annual revenue, 1 billion copies in cumulative circulation, 30 years of supplementary-materials accumulation — this is the foundation of a "traditional publishing company."

Deployed in hundreds of education bureaus and schools nationwide; the entire teaching-research platform of Beijing's Tongzhou District was built by it; a China Education Expo innovation award, selected for the AI education technology TOP 30 — this is its AI product's report card.

The education AI sector is crowded with the "regular army" — iFlytek (50,000 schools), ByteDance, Tencent, TAL. But the one that ran the path from "single-school pilot" to "regional full coverage" is a company that started by selling supplementary exercise books — Century Tianhong (世纪天鸿).

On what basis?

I talked with people at Century Tianhong for an afternoon and found the key isn't who has stronger technology, but who understands better "what schools actually need." The real battlefield of education AI was never technology, but content and scenario — this counterintuitive conclusion may be the best entry point for understanding this company.

"Is This Thing Helping Teachers Cheat?" — The Skepticism of 2023

In 2023, when XiaoHong Zhujiao (小鸿助教) first launched, it met a strangely split situation.

Top teachers from Beijing No. 4 High School and National Institute of Education Sciences Experimental School sought the team out, chasing them to exchange ideas and offer suggestions. These are China's most elite teachers, with a keener nose for new technology than anyone. In their eyes, AI isn't a monster but a new tool that makes teaching more efficient.

But go further down, and the situation is completely different.

Go to some ordinary schools and talk; their first reaction is: what is this? The second is more direct — "is this helping teachers cheat?"

This reaction sounds funny now, but the logic was easy to understand then. XiaoHong Zhujiao can auto-generate lesson plans and courseware; teachers associate it with those "handwriting-less lesson-plan wonders" on e-commerce platforms — in plain terms, tools to help teachers slack off. In some traditional educators' view, writing lesson plans by machine is no different from plagiarism.

So when it was truly pushed to schools as an AI product, it faced not a technology problem but a trust problem. Schools didn't believe it helped education; they worried it would weaken the professionalism of teaching.

Some may ask: how did it cross this hurdle later?

The turning point came in late 2024. DeepSeek went viral, AI spread nationally, and dense national education policy guidance; schools' attitude suddenly shifted from "is this thing reliable?" to "can this meet my needs?" Schools that once questioned began proactively contacting Century Tianhong.

"Honestly, what really pushed education AI to spread wasn't a technology breakthrough, but that society's understanding of AI finally loosened."

This shift also exposes a deeper problem: most schools don't actually know how to use AI, where to use it, or where its boundaries are. A single tool can't solve complex teaching problems; teachers take it back and try it twice, then it can't land, can't be used. The result is "easy to enter schools, hard to grow into teaching." This phrase precisely captures the universal pain point of today's education AI industry.

Century Tianhong's people see this clearly. How to help schools pinpoint AI products' usage scenarios and boundaries, and truly integrate them into the whole teaching process — that's the core idea behind their product landing. It's exactly this idea that kept them from following the road many education-AI companies take: "holding a hammer looking for nails."

The Tongzhou Model: From Contact to Landing in Half a Year

This March, Beijing's Tongzhou District launched the "Smart +" integrated teaching, learning, research, and evaluation platform.

This platform was built by Century Tianhong's XiaoHong Zhujiao, covering the entire Tongzhou District, managing everything from lesson prep to assessment. From contact to landing took only about half a year.

Tongzhou wasn't chosen at random. As Beijing's city sub-center, Tongzhou is taking high-quality district-wide teaching research as its main line, driving systematic transformation across three levels: "training system, training courses, training evaluation." Century Tianhong's product happened to hit the bullseye of this need.

How was it done concretely? It unfolded around four directions: assisting teaching, assisting learning, assisting research, and assisting evaluation.

For teachers, it can auto-generate lesson plans, pair subject tools, and reduce the burden of mechanical lesson prep. On the student side, the "homework supermarket" smartly assigns practice, helping students explore and truly learn. At the teaching-research level, exam-setting cases and teaching resources are integrated, so school-based research truly collaborates. On assessment, smart grading plus one-click paper assembly make learning status clear at a glance.

These four directions don't work separately but are strung into one line. The platform can string together teaching, homework, and grade data; teaching researchers look at real data, not gut calls.

In plain terms, it isn't selling a tool to a school, but managing the entire region's teaching, research, and evaluation. From serving a single teacher to covering the full teaching-learning-research-evaluation scenario, this positioning shift is itself telling.

But Tongzhou is only one benchmark of regional landing. What really caught my eye for data change was three months of routine use in a county-level district in Shandong.

For three consecutive months, covering multiple grade levels from elementary to junior high and multiple subjects, with a broad sample base fitting routine teaching scenarios. What was the result? Grading time shortened by 40%, and the class-average error rate fell 18%. Teacher feedback clustered on one point: mechanical grading, learning-status statistics, and course-design production work dropped sharply.

And because the product is based on Century Tianhong's thirty years of supplementary-content accumulation, it adapts to the new curriculum standards and local teaching reality, with content accuracy far better than generic AI tools — this is widely recognized by frontline teachers.

From single-school pilot to regional full coverage — XiaoHong Zhujiao has formally crossed this threshold.

"Education + AI" and "AI + Education" Are Two Different Things

Honestly, talking to this point, one feeling grew stronger in me.

The education AI sector actually has two completely different kinds of companies, though few separate them.

One is "AI + education" — technology-driven. Core strength is large models and compute, pursuing technical advancement, multimodal ability, and general generalization. iFlytek, ByteDance, and Tencent all walk this road. The logic is clear: my technology is strongest; seeping down into education scenarios is just a matter of time.

The other is "education + AI" — content-driven. Core strength is accumulated education content and scenario understanding, focusing on content adaptability, scenario landing, and teaching practicality. Century Tianhong is the representative of this school.

The two start from completely different points, which determines completely different ways of working.

"AI + education" companies think: what technology do I have, how do I stuff it into schools? "Education + AI" companies think: what do schools truly need, and how can AI help?

"The core competitiveness of education AI is truly kneading AI technology, education content, and teaching scenarios together. Not many in the industry can do this well right now."

You may ask: is content really that important?

An example. XiaoHong Zhujiao's lesson-plan design module introduced two national key projects from Beijing Normal University, led by Shi Lei, a recipient of national teaching achievement awards, with 150 special-grade and senior teachers participating. The final output was the country's first core-literacy-oriented smart lesson-plan design feature; the product also appeared at the 7th China Education Expo, exhibited in the Beijing Normal University section.

This isn't something you can pile compute into. It's using AI to redo thirty years of content accumulation and teaching understanding.

Why does Century Tianhong think it can pull this off?

Honestly, scale supplementary-materials companies have three moats big tech can't replicate in the short term.

The first is the content moat. Systematic, standardized content takes ten to decades to accumulate. Century Tianhong has published over 1 billion books in 30 years, covering all grade levels and subjects; this accumulated content isn't caught up by buying a few datasets.

The second is the channel moat. Years of school-entry partnerships and regional service networks are a supplementary publisher's core assets. Big tech has technology, but the "last mile" into schools relies on long-built relationships and trust.

The third is the compliance moat. Content review, content safety, policy adaptation — these are hard thresholds in education, not something you can just decide to do. Education's seriousness determines an extremely low fault tolerance; products must be scientific and authoritative.

The AI product is built on these three moats, with deep reuse of content, channels, and scenarios. In Century Tianhong's own words: "It's not a new scenario, but a new form and carrier to achieve the same goal as the book business: serving education and helping teachers teach well."

One more interesting detail. Beyond its own accumulation, Century Tianhong also complements its content ecosystem through investment. Its portfolio company Bishen Composition (笔神作文) — a leading company focused on K-12 Chinese composition grading — has 22 million users, over 300,000 monthly submissions in its composition community, and has accumulated 5 million quality samples. This corpus provides unique data support for AI model training while forming synergy with Century Tianhong's own content resources.

Content and data can reinforce each other — that's what others can't copy short-term.

Teachers Go From Grading Papers to Reading Data

When it came to concrete use, I found teachers' roles shifting.

In grading, error statistics, and material organization, teachers' role is subtly shifting.

Previously, judging learning status relied entirely on personal experience — a class of 40 students, who mastered it and who didn't, the teacher had a rough sense, but with limited precision. After using AI tools, learning-status data sits right in front of them; teachers' core work shifts from "grading and tallying" to "reading data and designing teaching."

From transaction executors, slowly toward learning-status decision-makers.

This shift looks small but carries weight. After AI comes in, everything in the classroom is changing, and teachers' roles naturally adjust too.

Century Tianhong also observed two very different kinds of schools in its school-entry process.

One treats AI as a fad. Leaders want an AI project; it heats up for a while and dissipates; teachers do what they did. The other treats AI as infrastructure, weaving it into the complete chain of curriculum, classroom, academics, and governance.

What's the difference? When a product truly integrates into daily teaching flow, teachers shift from "passive users" to "co-builders of requirements," proactively suggesting improvements to the platform. Only then does AI truly "grow" into teaching.

But there are pitfalls here too.

Machine grading is far faster than humans; no dispute. But subjective-question grading has dimension and granularity problems — the machine gives 5 points, the teacher may feel it's 6. More crucially, some schools reported unsatisfactory results; the core reason is that after getting the data they didn't know how to use it.

Century Tianhong's team's understanding is: data itself isn't valuable; what's valuable is whether you can read the meaning behind it. This is also why their product doesn't just provide tools but does deep service around the whole teaching-learning-research-evaluation process. The tool is only the entrance; service is the purpose.

Honestly, some things in education never change.

Machines can't replace a teacher's influence on students. AI isn't here to replace teachers, but those pure-labor grading and tallying tasks are indeed being redefined.

A 50,000-Service Package and Three Business Models

Back to the business side. Century Tianhong currently has three approaches to its AI product.

Single-school service packages, with average order value controlled under 50,000 RMB. This price is friendly to education budgets and won't make schools hesitate long. For budget-limited small and medium schools, it's a low-threshold entry point.

Regional platforms, targeting district and county education bureaus. Beijing's Tongzhou is the model for this approach — from contact to landing in half a year, from a single school to the whole district. Regional deals are larger in amount and coverage, and create demonstration effects.

API calls, targeting smart-campus vendors and publishers. This line follows an ecosystem route, opening AI capability for others to build applications on.

Century Tianhong's people didn't disclose specific AI-revenue figures, but from the phrasing "shifting from initial trial to stable scaled rollout," commercial validation is already underway.

Interestingly, Century Tianhong sees the integration of AI and books very clearly — not two lines doing their own thing, but running through three dimensions: production, product, and business format.

On the production side, AI is already embedded in the whole book R&D flow — intelligent manuscript assembly, knowledge-point tagging, intelligent proofreading, auto-generation of supporting resources — directly shortening R&D cycles. On the product side, paper books plus AI features make paper-digital fusion, adapting to schools' demand for digitalization. On the business-format side, using content and channel advantages to push smart-teaching solutions to education bureaus and schools — this is the second growth curve.

In plain terms, AI is a direction the book industry can't bypass, but the premise is you have the foundation of content, channel, and compliance. Miss one and it's all talk.

The supplementary-materials business is hard now — policy tightening, fewer children, and AI disrupting the scene. The old playbook doesn't work. In this situation, though school demand remains, the school-entry market is heavily homogenized. AI happens to open a window: through the XiaoHong Zhujiao entry point, penetrating the complete teaching-learning-research-evaluation chain and building differentiated core competitiveness within the existing market.

When Schools No Longer Ask "What Is AI?"

In 2023, going to schools to talk AI, you heard "is this thing helping teachers cheat?" In 2025, going to schools to talk AI, you hear "can it meet my requirements and achieve my effect?"

This shift itself says a lot.

Education informatization has moved from the incremental-build phase to the installed-base operate phase. School demand remains, but the school-entry market is homogenized. AI happens to open a window: through the XiaoHong Zhujiao entry point, expanding from serving a single teacher to covering the full teaching-learning-research-evaluation scenario, from selling tools to doing service.

In the end, the starting point differs, so the road differs.

Big tech does education AI from technology toward scenarios. Century Tianhong does education AI from scenarios toward technology.

Who goes farther? Hard to say now. But at least in this round of "from pilot to scale" race, the company that walked out of the book stack is temporarily ahead.

Thirty years ago, Century Tianhong started making supplementary exercise books. Now it's using AI to redo that accumulation.

"The carrier changes, the form changes. But in the end, it's the same thing: helping teachers teach well, making teaching a little easier. When schools no longer ask 'what is AI,' Century Tianhong's era has arrived."

Interview Highlights Q&A

Q1. It's been over half a year since that November 2025 interview; XiaoHong Zhujiao has gone from "commercial pilot" to "substantive breakthrough." What was the most critical change in between?

Zhang Minsong: Over this half year, XiaoHong Zhujiao completed the core leap from "single-point effect, small-scale validation" to "quantifiable scenario ROI and regional scaled landing."

If I had to name the most critical change in the breakthrough stage, I think it's mainly three things: standardized product system, systematized landing model, and normalized service delivery. In other words, we're also moving from tool thinking to service thinking.

To date, XiaoHong Zhujiao has achieved scaled deployment in hundreds of education bureaus and schools nationwide, covering ordinary public schools, district-level platforms, and commercial validation has shifted from initial trial to stable scaled rollout.

In the past, many AI products that entered schools hit a common problem: easy to enter schools, hard to grow into teaching. Schools are curious at first and willing to try, but after taking it back they don't know how to use it, where to use it, or where the boundaries are. Teachers may try it twice, and finally it can't land or be used.

This in turn validates a judgment: AI can't just be a simple auxiliary tool; it must integrate into the whole research process and become a core element in the teaching system.

So what we truly need to solve isn't just selling the AI product into schools, but helping schools define the product's scenarios and boundaries, so it deeply integrates into the whole teaching process.

Q2. From "pilot" to "breakthrough," was there a specific moment or project marking XiaoHong Zhujiao's entry into a new stage?

Zhang Minsong: If I had to name a milestone, the regional landing in Beijing's Tongzhou District was a very important turning point.

This March, Beijing's Tongzhou District launched the "Smart +" integrated teaching-learning-research-evaluation platform, built on XiaoHong Zhujiao to set up a district-level teaching-research agent, providing digital foundation support for the whole district's education process.

Behind this wasn't simply giving schools a tool, but building an integrated teaching-learning-research-evaluation platform at the regional level. We can consolidate the full chain of teaching data — teaching, homework feedback, academic performance — providing quantifiable, traceable evidence for district and school-based research, helping researchers do more precise teaching diagnosis and targeted training design.

This project also shows XiaoHong Zhujiao formally moved from the "single-school pilot" stage into the new stage of "scaled school entry + regional full coverage."

Q3. How long did the Tongzhou project take from contact to landing? Why could it move so fast?

Zhang Minsong: About half a year from contact to landing.

As the model of Beijing's city sub-center, Tongzhou is itself driving systematic transformation across the three levels of "training system, training courses, training evaluation" along the main line of high-quality district-wide research. And our product capability happened to fit the regional education-upgrade demand.

XiaoHong Zhujiao provides deep enablement around four directions: assisting teaching, assisting learning, assisting research, and assisting evaluation.

At the teaching-assistance level, for teachers' lesson prep, it intelligently generates lesson plans with subject tools to reduce mechanical burden. At the learning-assistance level, relying on the homework supermarket, it smartly assigns practice to help students explore and learn deeply. At the research-assistance level, it integrates exam-setting cases and shares teaching resources to push school-based research toward deep collaboration. At the evaluation-assistance level, through smart grading, one-click paper-assembly reports, and learning-status diagnosis, it achieves evaluation-promoted teaching.

The key to regional projects isn't giving every teacher an AI tool, but giving regional research a sustainably running data and service foundation.

Q4. A county-level district in Shandong's usage data shows "grading time shortened 40%, class-average error rate down 18%." Where did this data come from?

Zhang Minsong: This data comes from three consecutive months of routine teaching use, covering the district's pilot schools across multiple elementary and junior-high grade levels and subjects. The sample base is fairly broad and fits real teaching scenarios; the data is traceable.

From teacher feedback, the whole thing is highly consistent with our expectations. Most teachers recognize the burden-reducing and efficiency-raising value of AI grading and learning analysis, with widespread feedback that mechanical grading, learning-status statistics, and course-design production work clearly decreased.

Also, XiaoHong Zhujiao isn't a generic AI tool. Behind it is Century Tianhong's thirty years of supplementary-content accumulation, adapting to the new curriculum standards and fitting local teaching conditions. So in content accuracy and teaching adaptability, it differs from generic AI tools.

Q5. Has there ever been a case where AI grading actually made teachers busier? How do you see this "efficiency paradox"?

Zhang Minsong: From the grading scenario, we haven't clearly hit this yet. Machine grading is far faster than human grading, not on the same order.

But in use, the problems teachers feedback most are mainly two: one is grading dimension and granularity, the other is subjective-question judgment. For example, a subjective question where the machine gives 5 and the teacher thinks it should be 6 — this objectively exists and is explainable and acceptable.

Every product has limits; the user itself is our product manager. Feedback during use is our product-iteration direction.

Of course, some schools did report less-than-ideal results. But we found the core problem often isn't whether AI is useful, but that schools don't know how to use the data after getting it.

Only by decoding the truth behind the data can you see the real problems in teaching. AI grading is only the first step; what matters more afterward is how teachers understand this data and turn it into teaching improvement.

Q6. Bishen Composition has over 22 million cumulative users and 300,000 monthly submissions. What's its relationship with XiaoHong Zhujiao?

Zhang Minsong: Bishen Composition is an independent company we invested in. From its birth, it focused on a very vertical scenario — K-12 Chinese composition grading — and is also a leading company in the composition category.

Though hit somewhat by large models, it's broadly stable.

Bishen Composition and XiaoHong Zhujiao collaborate closely. Whether from product capability or business scenario, both have room to complement. Bishen Composition covers the student side, composition grading, and Chinese-learning scenarios; XiaoHong Zhujiao more enters from the teacher side and in-school teaching scenarios.

Also, the Bishen Composition community has over 300,000 monthly submissions and has accumulated 5 million quality samples. This content on one side provides data support for AI model training, and on the other can synergize with Century Tianhong's own content resources.

Q7. XiaoHong Zhujiao collaborates with Beijing Normal University and became a national-key-project research result. What practical role do these endorsements play in commercialization?

Zhang Minsong: Education is a serious industry, so in our product iteration we've been seeking collaboration with authoritative experts. We need underlying educational-theory support, aiming to make the product more scientific, not simply a generation tool.

For example, XiaoHong Zhujiao's lesson-plan design module introduced two national key projects. This project is led by Shi Lei, a National Teaching Achievement Award recipient and researcher at Beijing Normal University, with 150 special-grade and senior teachers participating, using XiaoHong Zhujiao as the practical carrier of the project, ultimately forming the country's first core-literacy-oriented smart lesson-plan design feature.

With this education-innovation result, XiaoHong Zhujiao was also invited to the 7th China Education Expo, exhibited in the Beijing Normal University section.

These landable, verifiable results and industry recognition are themselves a kind of endorsement. It naturally solves a key question in commercialization: why should schools and teachers trust you?

Q8. Century Tianhong's traditional supplementary book business is the revenue mainstay, and the AI business is seen as the second growth curve. How do you balance the "supplementary-materials cash cow" and "AI investment"?

Zhang Minsong: The supplementary-materials industry is at an inflection point where policy regulation, demographic shifts, and technological change are interwoven, and industry rules are changing. Enterprise transformation is inevitable.

As education digitalization deepens, school demand is upgrading. The whole industry is seeking to upgrade from single books to a comprehensive education service provider of "content + technology + service."

AI empowerment is an irreversible direction for the book industry. Our attitude toward AI is: embrace is inevitable, but openness and rationality coexist.

Inside Century Tianhong, AI-book integration mainly shows in three dimensions.

First is production-side integration: embedding AI in the whole book R&D flow for intelligent manuscript assembly, knowledge-point tagging, intelligent proofreading, and auto-generation of supporting resources, raising human efficiency and shortening product R&D cycles.

Second is product-side integration: building paper-digital fusion products, using paper books as the carrier and AI features to raise product added value, adapting to schools' demand for digital quality improvement.

Third is business-format integration: relying on accumulated content and school-entry channel advantages to push smart-teaching solutions covering the whole teaching process to education bureaus and schools, opening the second growth curve.

So the AI product isn't built from scratch, but on top of Century Tianhong's existing content, channels, and scenarios. It isn't a brand-new scenario, but a new form and carrier to complete the same goal as the book business: serving education and supporting teaching.

Q9. What's XiaoHong Zhujiao's current business model? Per school license, per teacher account, or by usage?

Zhang Minsong: Currently we mainly run three ways.

The first is the single-school service package. We provide it to schools as a standard-version service, fitting public schools, private schools, and training institutions, priced by account and service, with average order value controlled under 50,000 RMB, plus value-added packages.

The second is the regional platform, mainly targeting district and county education bureaus and city-level education cloud platforms, priced by service content.

The third is API calls, targeting smart-campus vendors, publishers, and third-party tools, outputting underlying capability.

These three ways essentially correspond to different clients and scenarios: single schools need usable, lightweight, landable; regional platforms need full-process data and research support; ecosystem partners need underlying capability output.

Q10. The education AI sector is crowded now. iFlytek, TAL, ByteDance, Tencent all do it. As a supplementary-materials publishing company, why would clients choose you?

Zhang Minsong: Honestly, everyone's DNA differs, so strengths and focus differ.

In education AI, there are roughly two kinds of companies.

One is "AI + education," technology-driven. Core strengths are large-model R&D and compute, pursuing technical advancement, multimodal ability, and general generalization, hoping to seep down into education scenarios through underlying technology breakthroughs. Technology is their core longboard.

The other is "education + AI," content-driven. Core strengths are accumulated education content and scenario understanding, focusing more on content adaptability, scenario landing, and teaching practicality, upgrading existing content and service systems through AI. Content is their core longboard.

Right now, players who truly fuse technology and content efficiently are still relatively scarce. The core competitiveness of education AI isn't only AI technology, nor content alone, but the deep fusion of AI technology, education content, and teaching scenarios.

Century Tianhong's starting point is "using AI to upgrade content services." Different starting points determine our depth of understanding across content, scenario, and landing.

We first chose to enter from the teaching-assistance scenario, which was also determined by our company's business composition. Teachers are our first users; how to serve teachers well is something Century Tianhong has always thought about and wanted to do well.

In the future, XiaoHong Zhujiao won't just be a teacher tool, but will extend from teaching assistance to assisting learning, fostering, research, and evaluation, from a single-teacher service product to a campus smart-teaching operations partner covering the full teaching-learning-research-evaluation scenario.

Education informatization has moved from the incremental-build phase to the installed-base operate phase. Schools don't lack systems; what they really lack is products that embed into the teaching closed loop and continuously produce service value. AI happens to give us a new window.

---

Original publication: https://uniqueresearch.substack.com/p/src-20260714-02html
On-site reading page: https://ffcap.cn/en/research/src-20260714-02html