What is a good OCR accuracy rate for business documents?

For clean, printed text, professional OCR typically achieves 98–99% character accuracy or higher. But the figure that matters is field-level accuracy on your key index values, such as reference numbers and names. Always ask a provider which measure they are quoting, and insist on human quality control for records you rely on legally or financially.

Can poor OCR affect GDPR compliance in the UK?

Yes. Under UK GDPR and the Data Protection Act 2018 you must locate and produce a person's records when they make a Subject Access Request, usually within one calendar month. If poor OCR means those records can't be found by search, you risk an incomplete response and potential ICO enforcement, with fines of up to £17.5 million or 4% of global annual turnover for serious failures.

Why can't I find documents in my scanned archive even though the scans look fine?

A scanned page can look perfect on screen while its underlying text layer is inaccurate or missing. Search relies on that text layer, not the image, so if OCR misread a name or reference the document will never appear in results. Testing real searches against a sample before scanning your full archive catches this early.

What scan resolution is needed for reliable OCR?

300 dpi is the practical minimum for dependable OCR on standard business documents. Scanning at lower resolutions to save storage space significantly increases recognition errors. Poor originals such as faxes, faded receipts or handwriting may need higher resolution, image pre-processing or specialist recognition to produce usable text.

Should I shred original documents after scanning?

Not until you have verified the digital copies are accurate, complete and searchable. Confirm OCR quality and run test retrievals first, then use a secure, auditable shredding service with a certificate of destruction. For records with strict statutory retention, check whether a signed original must be kept before destroying the paper.

Can Poor OCR Accuracy Make Your Scanned Files Useless?

Yes — poor OCR accuracy can quietly turn an expensive digitisation project into a filing cabinet you can’t search. Optical Character Recognition is what converts the picture of a scanned page into text your systems can read, index and retrieve. When that conversion is even a few percent off, names get misread, invoice numbers become gibberish, and the one document you need during an audit refuses to surface in a search. A crisp image on screen can hide a broken text layer underneath, which is exactly why OCR quality is the single most under-checked part of most UK scanning projects.

What “OCR Accuracy” Actually Measures

OCR accuracy is rarely a single, honest number. A provider quoting “99% accuracy” might be measuring something very different from what matters to you, so it pays to know which figure you’re being sold.

Character vs word vs field accuracy

Character accuracy — the percentage of individual letters and digits read correctly. 99% sounds excellent, but on a 2,000-character page that’s still 20 wrong characters.
Word accuracy — a single wrong character breaks the whole word, so 99% character accuracy often drops to 95% or lower at word level.
Field accuracy — the one that governs retrieval. If the invoice number, National Insurance number or client name is wrong, the record is effectively lost even if 99% of the surrounding page is perfect.

The lesson: ask what is being measured. A headline character-accuracy figure tells you almost nothing about whether you’ll be able to find a specific file three years from now.

How Poor OCR Makes Files Genuinely Useless

A scanned PDF with a broken text layer looks identical to a good one — until you try to work with it. The failure modes are predictable:

Unsearchable archives — if “Thompson” was read as “Thornpson”, a full-text search will never return it. Staff fall back to opening files one by one, which defeats the point of going paperless.
Broken automated indexing — many workflows auto-file documents by reading a reference or date off the page. A misread digit routes the record to the wrong folder, where it may never be found again.
Corrupted data extraction — feeding low-accuracy OCR into an accounts or case-management system imports wrong values at scale, and the errors are invisible until someone reconciles them.
False confidence — because the image still looks fine, nobody realises the text layer is broken until a specific document can’t be found, usually at the worst possible moment.

The UK Compliance and Legal Risk

In the UK the stakes go well beyond inconvenience. Under UK GDPR and the Data Protection Act 2018, individuals can make a Subject Access Request and you must respond, usually within one calendar month. If poor OCR means you can’t reliably locate every record relating to that person, you risk an incomplete response — and the Information Commissioner’s Office (ICO) can issue fines of up to £17.5 million or 4% of global annual turnover for serious failures.

The exposure is just as real for statutory retention. HMRC requires most business and VAT records to be kept for six years, and employers must retain certain records for longer still. A digitised archive is only compliant if the documents inside it can actually be produced on demand. If a tribunal, auditor or regulator asks for a specific contract and your search returns nothing because the OCR mangled the reference, “we scanned it” is not a defence. Accurate, well-indexed document scanning is what turns a pile of images into a legally usable record.

What Drives OCR Accuracy Up or Down

OCR isn’t magic, and its accuracy is largely determined before a single character is recognised. The main factors:

Scan resolution — 300 dpi is the practical floor for reliable OCR. Scan below that to save storage and accuracy falls off a cliff.
Original document quality — faxes, carbon copies, faded thermal receipts and heavily stamped pages all read poorly no matter how good the scanner.
Handwriting — standard OCR reads printed text; handwritten notes need specialist ICR and should never be assumed accurate without review.
Fonts and layout — unusual fonts, tables, multi-column layouts and coloured backgrounds all trip up recognition engines.
Image processing — deskewing, despeckling and contrast correction before OCR can lift accuracy dramatically; skipping it guarantees errors.
Quality control — the single biggest differentiator. A provider that runs confidence-flagging and human verification on low-scoring fields will hand you a usable archive; one that doesn’t, won’t.

How to Protect Your Project From OCR Failure

You don’t need to accept broken text layers as the price of digitising. A few checks up front make the difference between a searchable asset and an unusable one:

Run a pilot first — scan a representative sample and test real searches against it before committing the full archive.
Ask how accuracy is measured — insist on field-level accuracy for your key index values, not just a headline character figure.
Confirm there is human QC — automated OCR alone is not enough for records you’ll rely on legally or financially.
Specify searchable PDF/A output — it embeds the text layer in a format built for long-term archiving and future retrieval.
Keep originals until you’ve verified — never authorise secure shredding of source documents until you’ve confirmed the digital versions are accurate and searchable.

Done properly, OCR is what makes a scanned archive genuinely more useful than the paper it replaced — instantly searchable, easy to index, and quick to retrieve. Done badly, it produces a digital archive that looks complete but fails you the moment it matters. If you’re weighing up how to digitise safely, our resources library covers the practical decisions, and secure off-site document storage keeps your verified originals protected while your team works from the scans.