What slows media centres down: five findings from public research

In brief

Our research into German media centres and conversion services reveals a recurring pattern: bottlenecks do not primarily arise from extracting text. They arise where subject structure, multiple target media, quality review and limited time meet.

This matters to IncluLearn.AI because these are precisely the points where a semantic-first approach may help, provided that it reduces preparatory work without displacing subject-matter review.

Finding 1: Manual conversion can require substantial effort per page

Landesmedienzentrum Baden-Württemberg describes its text service as a labour-intensive process and reports 20 minutes to one hour per textbook page, depending on subject and material.

This is a strong indication that a PDF with recognised text is not yet accessible learning material. Reading order, headings, formulae, figures and subject relationships still need meaningful structure.

For IncluLearn.AI, this leads to a concrete research question: How much of this structural and semantic preparation can be machine-assisted without weakening subject review?

Finding 2: Lead times reveal real capacity constraints

The media centre at Johann-Peter-Schäfer-Schule in Friedberg specifies four to eight weeks of lead time. It also reports additional constraints during examination periods, strongly increased demand with stable staffing, and textbooks that may be converted section by section.

Mediablis Bayern recommends registration for centrally set examinations at least three months before the first examination date.

These figures are institution-specific and must not be read as averages across all media centres. They do show that throughput, prioritisation and predictable preprocessing are practically relevant issues.

Finding 3: Accessible media production is a subject transformation

The organisations we reviewed produce a wide range of target media: Braille, accessible e-books, large print, tactile graphics, swell paper, thermoform models, 3D models and in some cases audio or DAISY formats.

Mediablis also publishes binding conversion standards for Braille, e-books, mathematics, physics and chemistry. Blista describes developing models and teaching materials in cooperation with subject teachers and improving them based on learner feedback.

This makes the difference to generic image description clear: the task is to capture subject meaning so that suitable and consistent representations can be produced from it.

Finding 4: Reuse is already part of the workflow

LBZB Hannover describes checking its own holdings and then searching across the German-speaking region before a new conversion is started. If another media centre already has suitable data, this work can be reused.

Sehkon likewise records existing and planned accessible conversions. TU Dortmund reports more than 14,000 records and explicitly describes avoiding expensive duplicated production as one of its purposes.

For IncluLearn.AI, this is an important signal: a future system should not only support better new conversions, but also recognise, reference and — where legally and technically possible — reuse existing work.

Finding 5: Examinations form a distinct use case

For examinations, speed is not enough. Adaptations must remain subject-equivalent, avoid disclosing answers, and be produced through controlled processes.

Mediablis refers to binding standards and long lead times for centrally set examinations. Friedberg describes examination periods as a prioritised production phase.

For IncluLearn.AI, this means examinations are not simply another upload workflow. They combine confidentiality, auditability, controlled approval and particularly strict human-in-the-loop review.

What we cannot conclude yet

Public research shows real processes at specific organisations. It does not yet establish which bottlenecks are most important across Germany or which forms of automation media centres would actually accept.

We therefore treat these findings as evidence for the next questions, not as a finished requirements list. These points now need to be tested with media centres, conversion services and other partners.