How many boxes should a document scanning pilot include?

Between 5 and 20 boxes is usually enough. The number matters far less than the mix — the sample must include your most difficult material (mixed formats, staples, handwriting, faded or damaged originals) in roughly the same proportions as the full archive, or the throughput and quality figures will not extrapolate.

How long should a scanning pilot take?

Most pilots run two to four weeks from collection to sign-off. Physical scanning of 10 boxes takes days at most; the time goes into agreeing the specification beforehand, loading the output into your live system, letting real users test retrieval, and holding a proper review meeting at the end.

What quality standards should a scanning pilot be measured against?

Set thresholds before the pilot starts: 100% image completeness verified by page-count reconciliation, full legibility at 100% zoom with no skew or cropping, and index accuracy of at least 99.5% at field level — higher where a wrong reference has legal consequences. Measure OCR accuracy on a known control set rather than accepting a headline figure.

Do UK GDPR obligations apply during a scanning pilot?

Yes. If the pilot boxes contain personal data, you are processing under UK GDPR and the Data Protection Act 2018 from the moment the provider takes custody. Put a signed Article 28 processing agreement in place first, confirm where data is stored and processed, and verify staff vetting and ISO 27001 certification rather than taking them on trust.

What should happen if a scanning pilot fails its targets?

Agree the consequences in writing before the pilot begins — typically rework at the provider's cost, a second pilot against a revised specification, or a clean exit with your paper returned intact. A pilot with no defined failure route offers no real protection, because there is nothing to enforce when the results fall short.

What Should a Good Document Scanning Pilot Project Include?

A good document scanning pilot should be a small, deliberately awkward sample of your real archive — typically 5 to 20 boxes — processed end to end under the exact conditions of the full project, with a written specification, measurable quality targets, a defined index structure, and a formal sign-off before anything scales. The point of a pilot is not to prove that scanning works. It is to expose the decisions you got wrong on paper while the cost of changing them is still trivial.

1. A representative sample — including the difficult material

Sample selection is the single most important design choice in a pilot. If you hand over twenty boxes of uniform A4 laser-printed invoices from 2023, your provider will return a flawless result and a per-box throughput figure that collapses the moment they reach your 1994 personnel files. Deliberately include the material that worries you:

Mixed formats — A3 drawings, A5 receipts, fold-outs, plans, bound volumes
Physically difficult items — stapled bundles, treasury tags, laminated sheets, sticky notes, carbon copies, thermal fax paper that has faded
Handwriting — signed forms, annotated margins, engineer’s notes, clinical or case notes
Poor originals — photocopies of photocopies, water damage, foxing, coffee rings
Colour-dependent documents — highlighted text, colour-coded forms, coloured seals or stamps

Mirror the proportions of the whole archive. If 10% of your collection is oversized, roughly 10% of the pilot should be. Anything else produces an estimate you cannot rely on.

2. A written specification agreed before the first page is scanned

The pilot is where the specification gets tested, so it has to exist first. At minimum it should state:

Resolution and colour mode — 300 dpi is the working standard for business records; 400 dpi where OCR must handle small or degraded print. Bitonal for clean text, greyscale or 24-bit colour where the colour carries meaning
Output format — searchable PDF for everyday access, PDF/A-2b or PDF/A-3 where records must survive as long-term evidence, TIFF where a downstream system demands it
Document separation rules — what constitutes one document: a barcode sheet, a staple, a change in reference number
Index fields and file naming — which metadata is captured, from where on the page, and the exact naming pattern, delimiters and date format
Blank page and duplicate handling — removed, retained, or flagged
What happens to the paper afterwards — returned, stored, or securely destroyed after a hold period

Index fields deserve disproportionate attention

Adding an index field after 4,000 boxes have been processed usually means re-handling the files. Adding it during the pilot costs an afternoon. Ask each department who will actually retrieve documents how they will search: by employee number, matter reference, invoice number, site address, date range? Every answer is a candidate index field — and as our guide to bad file naming and indexing shows, these decisions are expensive to reverse.

3. Measurable quality targets — not “looks fine to me”

A pilot without numbers is a demo. Define the acceptance thresholds in advance and measure against them:

Image completeness — 100%. Every physical page accounted for, verified by page count reconciliation, not sampling
Image legibility — every character readable on screen at 100% zoom; no skew beyond a defined tolerance, no cropped edges, no show-through from the reverse side
Index accuracy — 99.5% at field level is a realistic contractual target for keyed data; 99.95% where a wrong reference has legal consequences
OCR accuracy — measured on a known control set, not accepted as a marketing figure. Clean modern print routinely exceeds 99% character accuracy; degraded or handwritten material can fall well below 90%
Separation accuracy — documents split at the right boundaries, against a known count

4. A full test of the process, not just the scanning

The pilot should exercise every stage of the eventual production run, because the failure points are rarely the scanner itself:

Collection and chain of custody — vetted staff, sealed containers, signed manifests, and a box-level audit trail you can inspect from your loading bay to the bureau and back
Preparation — de-stapling, repairs, insert sheets; usually the largest labour cost in the whole project
Scanning and QC — first-pass capture plus a documented second-pass check
Delivery — the actual transfer method: encrypted portal, SFTP, or encrypted physical media
Ingestion — load the files into your real document management system, SharePoint tenancy, or case management platform
Retrieval — have three or four end users try to find specific documents without help

That last step catches more problems than any technical audit. If a paralegal or payroll administrator cannot find the right file in under a minute, the index design is wrong regardless of how good the images look.

5. Compliance and security checks under real conditions

If your pilot boxes contain personal data — and HR, finance, clinical and legal files almost always do — the pilot is processing under UK GDPR and the Data Protection Act 2018, not a rehearsal for it. Before the boxes leave, confirm you have:

A signed Article 28 data processing agreement covering the pilot specifically
Confirmation that no data leaves the UK, or a documented lawful transfer mechanism if it does
Evidence of staff screening — BS 7858 vetting is the recognised UK benchmark
Certification you have verified rather than been told about: ISO 27001 and ISO 9001
Encryption at rest and in transit, plus a deletion schedule for the provider’s copy after handover

The ICO can impose penalties of up to £17.5 million or 4% of global turnover for serious breaches, and using a processor you never properly assessed is exactly the kind of failure that lands badly in an enforcement notice. Check the paperwork while only 10 boxes are exposed.

6. Numbers that let you forecast the full project

A pilot should end with data you can extrapolate. Capture:

Actual images per box — the estimate that most often wrecks budgets. A standard archive box is frequently quoted at around 2,500 sheets, but real boxes vary from roughly 1,500 to 3,500 depending on file thickness and packing. Multiply that variance across 1,000 boxes and the difference is enormous
Prep time per box — measured, not assumed, for both your easiest and hardest material
Exception rate — the percentage requiring manual intervention
Elapsed turnaround — from collection to delivered, searchable files
Storage footprint — the gigabytes produced, which drives ongoing hosting cost

Honest per-box figures let you build a defensible business case and, just as importantly, decide what not to scan. Many organisations find during a pilot that only a third of their archive justifies digitisation; the rest is better handled through off-site document storage with scan-on-demand retrieval, or securely destroyed under a certified shredding programme once retention periods expire.

7. A formal review and sign-off gate

Close the pilot properly: a documented review meeting, a written list of specification changes, revised pricing if the sample revealed a different workload, and an explicit go/no-go decision in writing. A pilot that quietly rolls into production without a review gate has functioned only as a delay.

Agree in advance what happens if the pilot fails its targets: rework at the provider’s cost, a second pilot, or a clean exit with your paper returned. Providers confident in their process will accept those terms. For more on evaluating suppliers, see our guide to professional document scanning and the wider resources library.

A good pilot is small in volume and large in scope. Ten hard boxes taken through every stage — collection, prep, capture, index, delivery, ingestion, retrieval — measured against written targets and closed with a formal review will tell you more than a hundred easy ones ever could. Design it to fail somewhere useful, and it pays for itself many times over on the full project.