What successful AI collaboration looks like in publishing

Express Publishing started nearly 40 years ago with two titles. Today, they produce ELT materials for 110+ countries, built on a reputation for quality, localisation, and pedagogical rigour that few competitors can match.

Every product is designed with the CEFR in mind. Every market gets locally adapted materials. Every book goes through focus groups and pilot stages. This is a publisher that does not cut corners.

ECEFR correlation is the process of mapping every exercise in a coursebook against the CEFR descriptors. For Express Publishing, it serves as both internal quality validation and external proof for partners and ministries.

The old process: one experienced editor, sitting with the book and the full CEFR descriptor set, going page by page, exercise by exercise. For a single book, this took 30 to 40 working days. Then, 5 to 7 more days of QA review.

The work required experienced editors with deep familiarity with both the material and the CEFR framework. Assigning them to correlation meant pulling them off new title development. And demand was growing: partners and ministries increasingly expected CEFR reports as standard, but Express Publishing could only produce them reactively, on request.

The bottleneck was slowing sales, limiting scalability, and draining the editorial team.

Ludenso’s Correlation Engine was built for one job: mapping educational content against curriculum frameworks with editorial-grade precision. For a single book, the engine makes over 30,000 individual micro-decisions - the equivalent of making 30,000 individual prompts in a general-purpose AI, far beyond what any editor could replicate manually.

The system operates on a human-in-the-loop model. It generates the mapping. Editors retain full control over the final output. No shortcuts, no blind trust.

The mapping phase dropped from 30-40 days to approximately 2 hours. QA review shortened from 5-7 days to ~3 days, because the starting quality is so much higher that there is simply less to correct. Total cycle: ~5 days.

Speed gains are expected. What surprised everyone was the quality improvement.

When the collaboration started, the engine’s output was above the level of a typical editor doing manual mapping, but below the standard that the experienced and senior management, Maria and Vasso, set for themselves. Through months of close iteration, the system surpassed the human baseline.

The engine achieves 98% descriptor accuracy. It does not skip descriptors out of fatigue. It does not make the errors that creep in during week three of tedious cross-referencing.

‍The key insight: generic AI applied to correlations produces output that needs more review, not less. The time saved in generation gets consumed in QA. Ludenso’s purpose-built engine inverts this: higher quality means less rework, not just shifted work.

The quality leap came from an unusually close collaboration. Vasso Dimopoulou worked hand in hand with Benjamin Kjær, who’s a Product Director of the correlations product at Ludenso: The duo ended up sending challenging descriptors to stress-test the system, reviewing edge cases, calibrating thresholds, and exchanging emails on weekends.

Benjamin spent over 250 hours in December alone refining the engine. The team set a 10-out-of-10 matching threshold: 100% match or it doesn’t get included. No compromises.

Proactive, not reactive. Express now publishes CEFR correlation reports on its website next to the sample button. Teachers and partners can access them anytime, no request needed.

‍Sales enablement. Correlation reports serve as proof of pedagogical quality for institutional buyers, school chains, and country partners.

Scalability unlocked. Express is ready for the next phase: correlating more titles, running the tool in-house, and exploring ministerial curriculum mapping beyond CEFR.

The editors are still there. They still own the final quality. They still make the creative decisions that define Express’s materials.

What changed is the nature of their work. Weeks of repetitive cross-referencing became days of high-value review. Reactive reports became proactive proof. A production bottleneck became a competitive advantage.

The Correlation Engine didn’t just save time. It elevated the floor. Better inputs, better outputs. Higher quality, less rework. Faster cycles, more opportunities captured.

For publishers facing growing demands for curriculum alignment, shrinking timelines, and increasing competition for institutional partnerships, the question is no longer whether to adopt purpose-built correlation tools. It’s how quickly you can start.