Choosing a document scanning company: what to actually test

Most scanning proposals read alike. Every provider says it is secure, accurate and experienced, and none of those words costs anything to write. The difference between suppliers shows up in things that can be inspected: how a pilot batch comes back, what happens to a page that fails quality control, who is standing next to your files at nine in the evening, and whether the quote itemises the work or hides it in a single per-page rate.

This guide is written from the delivery side, and it applies to us as much as to anyone else. If a provider cannot answer these questions concretely, that is information.

Start with the shape of your own project

Before you speak to anyone, write down four things. Suppliers cannot quote honestly without them, and a supplier who quotes confidently without asking for them is guessing.

• Volume, expressed the way you can actually measure it: box count, linear metres of shelving, file drawer count. Page estimates come later, from a physical sample.
• Document mix: sheet sizes, single or double sided, staples and bindings, condition, colour content, how many are handwritten, how many are bilingual.
• What you need to find afterwards, and by what. This drives indexing depth, which is usually the largest cost line after preparation.
• Where the digital files must land: a document management system, a shared drive, a line-of-business application, or a plain folder structure with a manifest.

The pilot is the real evaluation

A sample scan of a clean, flat, modern page proves nothing. Give every shortlisted provider the same awkward material and compare the output side by side.

1. 1

   Choose the worst representative sample, not the tidiest

   Include a faded thermal or carbon copy, a folded or torn sheet, a double-sided form with printing near the gutter, something stamped over the text, a handwritten annotation, and a bilingual document if you hold any. Keep the originals so you can compare against them.

2. 2

   Give an identical brief to each supplier

   Same output format, same resolution expectation, same index fields, same file naming. If briefs differ, the comparison is meaningless. Ask for the pilot to be priced. A free pilot is often run by the best operator on the floor, which is not who will run your programme.

3. 3

   Inspect the images before you look at the data

   Check for clipped edges, lost content in the gutter, skew, dropped colour on stamps, over-thresholded faint text, and pages that came through blank because a sheet double-fed. Ask how the provider detects a double feed, and what happens to the batch when one is found.

4. 4

   Verify the index against the page, field by field

   Take twenty documents from the pilot and check every index field against the image yourself. Count field errors, not document errors. A document with one wrong field out of eight is not eighty-seven per cent correct in any way that helps you find it later.

5. 5

   Test retrieval the way your team will use it

   Search for a document by the term a real user would type, including a misspelled name, a partial reference number and a date range. Open the searchable PDF and copy a paragraph out of it. That single copy-and-paste exposes more OCR and text-layer problems than any report will.

Quality control: ask what happens to a failure

Every provider will say it has a multi-stage quality process. The useful question is narrower. Ask what defect classes are defined, how a sample is drawn, what the acceptance rule is for a batch, and what happens when a batch fails. The answer you want is that the whole lot is reworked and re-sampled, not that the found defects are patched and the rest is waved through. Ask who signs the acceptance, and ask to see the QC record for the pilot batch you supplied.

Ask separately about index verification, because image QC and index QC are different jobs done by different people. Double-key entry on critical fields, blind re-keying of a sample, and validation rules on structured fields such as dates and reference numbers are all reasonable answers. "An experienced operator checks it" is not.

What a real quote itemises

A single per-page rate for a mixed archive is a rate that will be renegotiated later, usually after your files have already left the building. The line items below are the ones that move.

Security, references and contract terms

Security deserves its own conversation and its own checklist. At the shortlist stage, three things are enough to separate serious providers: a described chain of custody from your shelf to the delivered file, a clear statement of who has physical and logical access, and a straight answer on subcontracting. If any part of the work leaves the provider's own premises or country, you should be told without having to ask twice.

For references, ask for a client with a comparable document mix rather than a comparable logo. The questions worth asking a reference are narrow: did the index schema survive contact with real users, what changed after the pilot, and how were exceptions handled. On contract terms, look specifically for these.

• Liability for loss or damage to originals, and whether it is capped at the value of the scanning contract. Irreplaceable records deserve a specific discussion, not a standard clause.
• Insurance covering documents in transit and on the provider's premises, with the certificate attached rather than referenced.
• Defined acceptance criteria, an acceptance window, and a rework obligation that survives acceptance for defects you could not reasonably have found.
• Confidentiality that binds named personnel and any subcontractor, not just the corporate entity.
• An exit clause: what you get back and in what format if the contract ends early. Your images, your index data and the mapping between them.
• Ownership of the index data and derived metadata. It should be yours without qualification.

Signals that should slow you down

• A firm per-page price quoted before anyone has seen or sampled the documents.
• An accuracy percentage offered without a definition of how it is measured, on which document set, and at character, word or field level.
• Reluctance to let you visit the production floor, or to say where the work is physically done.
• No named exception process. Every archive contains documents that cannot be scanned as specified.
• An index schema proposed before anyone has asked how your team searches today.
• Vagueness about what happens to your files, and to the provider's working copies, once the project closes.

The providers worth shortlisting tend to ask more questions than they answer in the first meeting, quote after a physical assessment rather than before it, and are specific about what they will not do. That is the tone to look for.