Vendor selection checklist for digitization
Seventeen questions to ask any digitization vendor, including us, with the answers you should expect to get in writing.
Checklist
Choosing a digitization vendor is choosing who will handle your originals, what your files will look like for decades, and whether you can take everything elsewhere later. Portfolios and sales calls do not answer those questions. This checklist turns them into seventeen specific questions you can send to every vendor you are considering, so the answers can be compared side by side.
Ask for every answer in writing. An answer that will not go into writing is an answer in itself.
Capture
- Which imaging guideline and quality level, in writing. Ask which guideline the vendor works to, which category applies to each of your materials and which level they will meet. FADGI, the federal guidelines, defines four star levels and calls three-star “a very good professional image that is appropriate for most uses”. A star rating is determined by meeting FADGI’s evaluation parameters through image analysis, so ask how and how often they measure it.
- Handling of difficult formats. Bound volumes, glass plates, film and oversize items each need different equipment and care. Ask how each is supported and what they will not take.
- Custody of originals. Do your originals ever leave the vendor’s sight, are they sent on to anyone else, and where are they stored overnight?
- A pilot on your own material. Sample images from a pilot on your hardest items tell you more than any portfolio.
Text and metadata
- Recognition engines. Which engines they use for print and which for handwriting, and how they choose between them.
- Accuracy measurement. How they measure text accuracy, and what they report per batch. A character error rate per batch and per kind of page is a usable answer; a single accuracy percentage with no method behind it is not. The OCR accuracy calculator shows the arithmetic.
- Machine-suggested fields. Which catalog fields a model suggests, and how those suggestions are labeled until a person reviews them. The Library of Congress’s Exploring Computational Description experiments found that “no current AI tools are currently good enough to run fully automatically” for catalog records, so review and labeling matter.
- Training on your data. Whether your images, text or records are used to train any model, and under what terms.
- Export formats. Which formats they deliver, and whether your collection system can import them. Ask for a sample export and test it.
Delivery and access
- File formats and folder structure for masters and access copies, described in the contract.
- Checksums and packaging delivered with the files, for example BagIt bags with checksum manifests, so you can verify that everything arrived intact.
- Ownership of files and records, stated in the contract rather than implied.
- The exit. If the vendor hosts your collection, how you leave with everything: images, text, records and any custom work.
Terms
- Price basis. Per image, per item or per hour, and what is excluded: preparation, conservation, transport, storage, rework.
- Insurance and liability while originals are in their care. Ask to see the cover.
- Schedule and milestones, and what happens when a milestone slips.
- Contact and response. Who you call, and how quickly they commit to answer.
Example (fictional). The Aster County Historical Society sends the checklist to two vendors as a questionnaire and asks each for a pilot on the same material: the Wren Gulch Mercantile day book, which is fragile and opens to only about 110 degrees, and a folder of Okerlund family letters. One vendor answers the capture questions in writing and names the guideline and level for the day book and for the letters. The other sends a portfolio. The society scores each answer as clear, vague or missing, then compares the two pilots using the image QC checklist and each vendor’s reported error rates on the letters.
Using the checklist
Send the questions before you ask for a quote, and again with the draft contract to confirm the answers made it in. Tick the items here as answers arrive, or download the CSV and add a column for each vendor.
Put the same questions to us. We capture to the level agreed in writing and check it against targets, choose recognition engines per collection by measured results, and label machine output until a person reviews it; our AI policy sets this out. We do not use your images or text to train a third party’s general model without your written permission. You receive the master files, access files, text and records, and each agreement sets out ownership and licensing. Our engagements page shows what you provide and receive, and text recognition explains how we measure accuracy. The Reading Room’s planning and funding section covers vendor selection in more depth, and a pilot box is the most direct way to see our answers on your own material.
Send one box first.
A pilot box shows you the images, the recognized text and the catalog records before you commit a whole collection. Tell us what is in the box and what you want people to find.
Studio and contacts
24 South Weber Street, Suite 225Colorado Springs, CO 80903(719) 244-8450sales@dastok.com