How much does digitizing old manuals cost — and how to cap the bill

Digitization (OCR) is the only variable expense when you turn your documentation into an AI assistant. It's also the one where you can get surprised. Here is how the cost really forms, and the safeguards to insist on.

Does everything really need OCR?

A computer-generated PDF (exported from Word, from desktop-publishing software) already contains its text — free extraction. Only scanned documents — photographed paper — need visual reading. On a typical industrial SME corpus, that's often 10 to 30% of the pages. First safeguard: demand an inventory that distinguishes the two before any quote. Paying for OCR of a clean PDF means paying for nothing.

Why can a PDF "with text" be illegible?

A nasty case: PDFs run through an old OCR fifteen years ago. They seem to have text, but it's riddled with errors ("c0nnector", "torque 45 Nm" turned into "torqu3 45 Nrn"). A good inventory measures the quality of the text layer and recommends re-digitization only when it's truly bad — with the corresponding quote, not a surprise.

What forms the price?

• The number of scanned pages — the dominant factor in the quote.
• Images and diagrams — describing a hydraulic diagram costs more than a page of text; a preliminary sort (useful images vs logos and decorations) avoids paying for noise.
• Reprocessing — the hidden line item. If the vendor re-processes the same pages every time the corpus is updated, the bill repeats. Pages already digitized must be remembered and never re-billed.

Which safeguards to demand from any vendor?

1. Exact quote before any spending — not an estimate: the number of pages is known in advance, and so is the price.
2. Explicit GO — no page processed without your written approval, and ideally a technical lock (not just a commercial promise) that prevents spending without that GO.
3. Capped budget — a hard limit in the contract: if processing reaches the cap, it stops and you decide.

The question that sums it all up

"If I re-upload the same corpus tomorrow, how much does it cost?" The right answer is zero — everything already read is remembered. If the vendor hesitates, you've just found their business model.