Plan a pilot box
Menu

Writing a digitization RFP that gets comparable bids

Reader question: How do we write a request for proposals for scanning and digitization so that vendor bids can be compared fairly?

By Dastok Editorial TeamIssue: Published

A digitization RFP gets comparable bids when every bidder prices exactly the same work. That means the same counts of each kind of original, the same measured specification for each material, the same deliverables and quality checks, and one bid form that asks for unit prices you multiply by your own quantities. Make a priced pilot the first phase, publish your evaluation weights before proposals arrive, and put data ownership and AI-use terms in the request itself, because they change what each bid is worth.

A disclosure first. Dastok digitizes historical collections and answers requests like these, so we have an interest in how they are written. The structure below is meant to make bids from any vendor easy to compare, ours included, and it draws on published solicitations and professional guidance rather than on our own forms.

This article covers the document. Whether to outsource at all is weighed in our article on in-house or outsourced digitization, and scoring the responses is the job of the vendor selection checklist. Nothing here is legal advice: your procurement office, your funder and your counsel decide the form a solicitation must take.

What a digitization RFP has to do

A digitization RFP has one job: to make every bidder price and promise the same thing. Then the differences between bids are differences between vendors, not between the assumptions each vendor had to make.

RFI, RFQ or RFP

The three documents answer different questions. The Academy Digital Preservation Forum’s RFP template for scanning services puts it simply. A request for information is used “earlier in the cycle when an archive is initially entertaining possible solutions.” A request for quotation “is typically sent to a single-service provider.” An archive issues an RFP “when it has clear-cut requirements and is interested in comparing services, proposed workflows, and pricing.”

The Northeast Document Conservation Center’s leaflet on outsourcing and vendor relations, written by Robin L. Dale, calls the request for information a “casting net” for finding vendors able to answer the RFP that follows. If you know what you want and need prices, write the RFP.

Why bids come back impossible to compare

Most bad comparisons trace back to something the RFP left open, which forces each bidder to guess:

  • No counts. One bidder prices 20,000 images and another 35,000, because the RFP said “about 60 boxes.”
  • No measured standard. “High resolution” lets one bidder quote 300 ppi and another 600 ppi.
  • Bundled services. One bid folds preparation into a per-image price; another bills it by the hour.
  • Missing lines. Transport, delivery media and rework appear in one bid and not in another.
  • No stated weights. Bidders cannot tell whether price or care counts for more.

The Academy template states the trade-off in two sentences: “If an RFP is too vague, proposals will be vague. On the other hand, overly detailed and restrictive RFPs can be limiting.” NEDCC’s leaflet sets the target. An RFP should be broad enough “to allow for different vendors to propose alternative methodologies where appropriate, but specific enough to ensure that vendors understand the standards that they will be required to meet.”

When public or grant money pays for the work

Public bodies follow their own procurement codes, and federal grant money brings federal rules. Under the Uniform Guidance at 2 CFR 200.320, as the eCFR showed it in November 2026, “Requests for proposals require public notice, and all evaluation factors and their relative importance must be identified.” The contract goes to the offeror whose proposal is “most advantageous” considering “price and other factors.” Under 200.317, states and Indian Tribes use their own procedures; most other recipients follow 200.318 through 200.327.

The competition rules in 2 CFR 200.319 shape how you write specifications. A solicitation must give “a clear and accurate description of the technical requirements,” and “Detailed product specifications should be avoided if at all possible.” Specifying only a “brand name” product is listed as a practice that may restrict competition. If digitization is part of a grant project, ask the grants office which rules apply before you draft; our guide to digitization grants covers the programs themselves.

The sections of the RFP, in order

Section What it fixes
Summary and schedule Purpose, dates, contacts, how questions are answered
Collection description Counts, sizes, formats and condition
Technical specification Guideline and level per material; files and names
Text and records Formats, fields, accuracy measures, machine labels
Handling and insurance Locations, transport, storage, coverage
Quality control Vendor checks, your sample, the acceptance rule
Bid form Unit prices per format and service
Phase one pilot A priced test with a decision point
Data and AI terms Ownership, deletion, model use, exit
Evaluation Gates, criteria and weights

Describe the collection with counts, not adjectives

Bidders price what they can count. Give them counts by format, measured from an inventory and labeled as estimates where they are estimates, along with sizes, condition problems and what is out of scope.

The Washington State Archives’ advice sheet on digitizing costs (November 2024) names four factors that drive price: variety of sizes, obstacles such as “staples, clasps, folds, rips, and fragility,” content and access. Each belongs in your description. Its rule of thumb, that a standard storage box “holds an average of 2,500 pages, when filled,” shows why “60 boxes” is not a count: a box of letters in envelopes and a box of typed minutes are different jobs.

What the description should state

  1. Counts by format, in images rather than items where you can, because capture time follows images. A cabinet card with writing on the back is two images.
  2. How each count was made: every item counted, or a sample of boxes scaled up.
  3. Sizes, especially anything over 11 x 17 inches or anything that will not lie flat.
  4. Condition flags: brittle paper, tight bindings, fasteners, folds, tape, past mold.
  5. Exclusions: items that will not be digitized, or that need a conservator first.
  6. Arrangement: how items are housed and numbered, and that they return in the same order.
  7. Viewing: whether bidders may examine the material before bidding.

A collection inventory worksheet gives a usable set of columns, and the completed sheet can go into the RFP as an appendix.

Flagging items is work in itself. When the Smithsonian’s Ralph Rinzler Folklife Archives and Collections prepared two collections for off-site vendor digitization, it issued a separate request for quote in 2018 for about 1,280 hours of archives technician work. The technician would flag “special photographic formats and materials that will not be digitized before they are transported to the vendor,” then inspect the returned files, embed metadata and ingest them. Budget for preparation on your side too.

Example (fictional). The Aster County Historical Society describes its collection in four lines, each with the basis for its count:

Series Format Estimated quantity How counted Condition notes
Family papers, 60 boxes Letters and manuscripts, up to 11 x 17 in About 8,500 items, 27,800 images 6 boxes sampled and scaled up Letters folded in envelopes; some pencil
Prints in the same boxes Cabinet cards and mounted prints About 1,100 items, 2,200 images Same sample Pencil notes on backs; gilt mount edges
Store ledgers Bound volumes 9 volumes, 1,900 pages Every page counted Wren Gulch day book: split spine, opens to about 110 degrees
Oversize Maps and posters over 11 x 17 in 40 sheets Every sheet counted Two sheets torn at folds; conservator to review first

A last line invites bidders to examine Box 3 at the society by appointment during the first two weeks of the response period.

Write technical specifications by material

Name the guideline, its edition and the level for each kind of material, and require proof by target measurement rather than by equipment settings. Then fix the file formats, names and folder structure, so that every bidder prices the same files.

Name a measured level, not a resolution

The Federal Agencies Digital Guidelines Initiative’s Technical Guidelines for Digitizing Cultural Heritage Materials (third edition, 2023) grades images from one to four stars. A rating “is determined by meeting the FADGI evaluation parameters through image analysis.” Sampling rate is one parameter among several. Writing “FADGI three-star, category 3.3, verified by target analysis in every session” asks for something a bidder can prove. Writing “400 dpi” does not.

The guidelines say FADGI “does not recommend digitization to less than a three-star level” where possible. The table gives the minimum sampling rate for three-star and four-star work in the categories a historical collection usually contains.

Material FADGI category Three-star minimum Four-star minimum
Bound volumes, rare and special 3.1 300 ppi 400 ppi
Letters and manuscripts, rare and special 3.3 300 ppi 400 ppi
Unbound documents, general collections 3.4 300 ppi 400 ppi
Maps, posters and other oversize items 3.6 300 ppi 400 ppi
Newspapers and clippings 3.7 300 ppi 400 ppi
Prints and photographs 3.8 400 ppi 600 ppi
Negatives, 35mm up to 4 x 5 in 3.11 3,000 ppi 4,000 ppi

Each value carries a small tolerance, and each category sets other limits as well. Our explainer on FADGI star ratings shows what the levels measure and what a vendor’s target report should contain; require that report for every session.

Specify performance, not a scanner

The Academy’s template, written for motion picture film, includes a line for scanner preference. For paper and photographs, a measured level does that work without naming a brand, which also avoids brand-only specifications in federally funded procurements. Ask each bidder to name the equipment it proposes for each material and how fragile items will be supported. That is information to score, not a requirement.

Files, names and structure

State each of these, so no bidder has to assume:

  • Master files: format (for example, TIFF), bit depth and color space, as each FADGI category requires.
  • Access files: format and pixel size, and whether you want one PDF per item.
  • File names: the pattern, with an example. A file naming builder produces one, such as achs_c003_b03_f012_i004_0001_m.tif.
  • Folders: by box and folder, mirroring the physical arrangement.
  • What gets captured: both sides wherever there is writing, blank pages in bound volumes, envelopes and enclosures, with a color target and scale.
  • Delivery: packaging and checksums. BagIt, published as RFC 8493 in October 2018, is a common choice.

Specify text recognition and catalog records as separate deliverables

Name each text and record deliverable, its format and how its accuracy will be measured. Have bidders price machine output and human correction on separate lines, and require machine-made values to stay labeled.

Recognized text

Printed clippings, typescript and handwriting are different jobs, so ask for each separately:

  • Output. Plain text per page and, for print, layout-aware XML such as ALTO; the Library of Congress ALTO page lists version 4.4 as “Official.” Word coordinates let a viewer highlight search hits.
  • Accuracy measure. Character error rate (CER) against reference transcriptions. Transkribus, a handwriting recognition platform, defines it in its help pages as “the percentage of characters that have been transcribed incorrectly by the Text Recognition model.” Require CER for each batch, by kind of text, on a random sample of pages, and say who writes the reference transcriptions.
  • Correction. None, names and dates only, or full correction, each priced on its own line.

Do not write a target CER into an RFP for handwriting you have never tested. Ask bidders to report measured rates, then set the threshold in phase one. The OCR accuracy calculator computes CER from a reference and a machine reading, so you can check reported figures yourself.

Catalog records

A record deliverable needs more definition than “metadata.” Specify:

  1. Level of description: item, folder or series, for each part of the collection.
  2. Required fields, with the standard and vocabulary for each: Dublin Core for a CSV import, MARC or EAD where your system needs them.
  3. Machine suggestions: which fields software may draft, and that each machine value stays labeled until a person reviews it. Our article on labeling AI-generated metadata gives a status vocabulary you can attach.
  4. Review: who reviews machine values, priced as its own line.
  5. Rights and restrictions carried into each record.
  6. A test import of pilot records into your own system.

Avoid a single line called “AI cataloging,” because accuracy differs by field. The Library of Congress’s Exploring Computational Description experiments found identifiers, author and title could be extracted “at around 85% accuracy,” while subject headings, genre and dates were accurate “at or under 50% of the time.”

Set rules for handling, transport, custody and insurance

State where the work may happen, who moves the originals, how custody is recorded, how originals are stored while away and what insurance the vendor must carry. Then bidders price the handling you require, not the handling they assume.

Where the work happens

Say whether work may happen on your premises, at the vendor’s facility or either, and require every bidder to name each location where your material or files will be, including subcontractors. NEDCC’s leaflet warns that “Some vendors offer cheaper services by sending the work offshore.” The Academy template bars subcontracting without the client’s “prior written consent” and reserves the right to inspect the vendor’s place of business “at reasonable times.”

Transport and custody

Write the custody rules into the RFP so they can be priced:

  • Packing: who packs, to what standard. Our packing and transport checklist is a starting list.
  • Transport: vehicle, routing and whether material may ever be left unattended.
  • Handoffs: a signed manifest at every change of hands, matched against your inventory.
  • Storage: secured, within environmental ranges you set, with a known overnight location.
  • Condition: reports before and after, with photographs of existing damage.
  • Return: every item back in its original order and housing.

Our collection care page shows the level of custody detail worth asking any bidder for.

Insurance

Set the coverage you need and ask for proof with the proposal. The Smithsonian’s 2018 request required general liability insurance of $500,000, with the Smithsonian as additional insured, and proof “submitted with quotes.” Ask separately how originals are covered in the vendor’s care and in transit, at values you state, and ask your own insurer what your policy covers off site.

Build quality control and acceptance sampling into the contract

Write two inspections into the RFP: the vendor’s own check of every batch, and your inspection of a random sample from each batch. Add a numeric rule for when a batch is accepted or re-inspected, and name the defects, your review window and who pays for rework.

What FADGI recommends

Chapter 10 of the FADGI guidelines says quality plans “should address accuracy requirements for and acceptable error rates for all aspects evaluated.” It describes two steps in most projects: the scanning technician checks images during production, and a second person checks them afterward. For the visual inspection, it recommends “at a minimum, 10 images or 10 % of each batch of digital images, whichever quantity is larger,” viewed at 100 percent magnification.

FADGI also gives an acceptance rule you can copy. If the random sample shows more than 1 percent of a batch’s images and metadata to be defective, “the entire batch should be re-inspected,” and every error found is corrected. Below 1 percent, only the defective images and metadata are redone.

Sampling decides acceptance, not quality

The NIST/SEMATECH e-Handbook of Statistical Methods makes a point worth putting in front of everyone who reads the contract: “the main purpose of acceptance sampling is to decide whether or not the lot is likely to be acceptable, not to estimate the quality of the lot.” Your sample is a gate; the vendor’s own checks and reports keep quality steady.

Sample size is a choice to write down. For a batch of 2,000 images, FADGI’s minimum of 10 percent means 200 images. Our QC sample size calculator, at its default settings (95 percent confidence, a 5-point margin and an expected defect rate of 5 percent), gives 71 images for the same batch. You might use the larger sample for early batches and the smaller one once results are steady; whichever you choose, say so in the RFP.

Define the defects

A defect list prevents arguments later. FADGI’s inspection list is a good base, extended here to image faults and text:

Area Defects to name
File Will not open; wrong format, bit depth or color profile; missing embedded metadata
Capture Missing page or side; out of order; cropped text or target; wrong orientation; skew; wrong scale
Image Focus, glare, gutter shadow or color outside the agreed tolerance
Names and records File name breaks the pattern; required field empty; machine value unlabeled
Text Batch CER above the agreed threshold; missing pages of text

Review window and rework

Give yourself a fixed number of business days after each delivery to inspect. NEDCC’s leaflet lists “a statement on error correction, whether error correction is free to the organization, and what charges, if any, may be associated with it” among the clauses a contract needs. Define a batch too, for example one box, so the sample, deadline and rework attach to something countable.

Use a pricing table that forces comparable bids

Attach a bid form rather than asking for a price: one row per format and service, a unit for each row, your estimated quantity and a blank for the unit price. You compute the totals, so every bid is priced against the same quantities. State what each unit price includes and how long prices hold.

Where the idea comes from

None of this is new. NEDCC’s leaflet, drawing on Janet Gertz of Columbia University, asks for “Prices in specified units of measure, for instance, per page, per image, or whatever is appropriate,” for the costs of additional services such as delivery media, shipping and insurance, and for whether prices are “firm for the duration of the project.”

The RLG Guidelines for Creating a Request for Proposal for Digital Imaging Services, prepared by Cornell University Library in 1997, went further. A scanning price “should include set up, scanning, data entry, special handling, inspection, basic structuring and labeling, storage media, shipping, insurance,” while “indexing, text conversion and encoding should be priced separately.” The Smithsonian’s 2018 request added a simple line: “Cite the date through which pricing submitted is valid.”

LineService and specificationUnitEstimated quantityUnit priceExtended price
1Letters and manuscripts, FADGI 3.3 three-star, every side with writingImage27,800
2Prints and cabinet cards, FADGI 3.8 three-star, front and backImage2,200
3Bound ledgers in a cradle, FADGI 3.1 three-starImage1,900
4Oversize sheets over 11 x 17 in, FADGI 3.6 three-starImage40
5Preparation beyond what lines 1 to 4 include (unfolding, fasteners)Box60
6Handwritten text recognition, uncorrected, CER reported per batchPage27,800
7Item records, required fields, machine values labeledRecord9,600
8BagIt package with SHA-256 manifest, per delivered batchBatch12
9Packed and escorted transport, round tripTrip6
10Phase one pilot on Box 3, every stage, with written reportFixed1
OptionsText correction (names and dates; full); review of machine values; hosting per yearAs statedPriced, not totaled
Fictional bid form for the Aster County Historical Society; quantities are invented. Structure follows the pricing guidance in the RLG RFP Guidelines (1997) and NEDCC Preservation Leaflet 6.7, checked November 2026.

Rules that make the form work

  1. Define every unit. An image is one capture of one side; a page is one side that carries text.
  2. List what each price includes. For capture rows: handling, target capture, file naming, access files, vendor checks and packaging. Anything else gets its own row.
  3. Cap hourly rows. Any row priced by the hour needs the bidder’s estimate of hours at your quantities and a not-to-exceed figure.
  4. Price options without totaling them. Correction levels and hosting are decisions you make later.
  5. Keep alternates separate. A different method goes on a second form; the base form must still be complete.
  6. Fix prices and validity. Prices firm for the contract term, with a validity date and any escalation rule stated.
  7. Apply unit prices to actual counts. If real quantities differ from your estimates, unit prices apply to what was delivered.

A published rate list shows where totals drift

The Washington State Archives’ Imaging Services Rates List, revised December 15, 2025, serves state and local government agencies in Washington. It prices paper up to 11 x 17 inches per image in six grades, from $0.08 for Grade A to $0.55 for Grade SH. Wide format, flatbed and camera work on bound, oversize or fragile items is billed at $65.00 per hour, and file renaming, indexing per field, delivery media and “Shipping Cost Plus 30%” are separate lines.

Those prices apply to Washington agencies and are not a guide to the market. The lesson is the structure: grades, hourly lines and small add-ons are where two bids stop being comparable unless your form asks for every line.

Capture Preparation Text recognition Delivery and transport Bid A $0.45 per image $39,090 Bid B $0.70 per image $31,400 Bid C $0.60 per image $33,570 0 10k 20k 30k 40k Total at the society's quantities, US dollars
Fictional example data for illustration: three invented bids priced on the bid form above, not market rates. Lines 1 to 6, 8 and 9 only; records and the pilot are left out to keep the chart simple. Bid B includes preparation in its capture prices. The line 1 price per image is shown under each bid name.

Example (fictional). Three bids reach the Aster County Historical Society. Bid A has the lowest price on line 1, $0.45 per image, but prices line 5 by the hour at an estimated 150 hours, charges $0.35 per page on line 6 and adds more for media and transport. Bid B quotes $0.70 per image with all preparation included and $0.20 per page. Bid C quotes $0.60 per image, $40 per box for preparation and $0.25 per page. Multiplied out at the society’s quantities, the cheapest headline price is the most expensive bid: $39,090 against $31,400 and $33,570. Every number in this box is invented to show the arithmetic.

Unusually low or high bids

Treat outliers as questions. NEDCC’s leaflet advises: “Be cautious of unrealistically low or high bids,” and notes that “A very low bid may indicate a vendor who cuts corners or has failed to understand what is really required to produce the products.” A complete bid form shows which line is low, so you can ask about it.

Make a pilot the first phase of the contract

Award the work in phases. Phase one is a fixed-price pilot on a defined sample of your hardest and most typical material, run through every stage, with pass criteria written into the RFP. Phase two goes ahead at the unit prices bid only if phase one passes, and you may stop without penalty if it does not.

Sample scans are not a pilot

Sample scans with a proposal are old practice. NEDCC’s list of what a vendor response should include asks for “a sample that represents a fair cross section of the materials, including both easy and difficult items.” Sample scans show image quality. They do not show error rates on your handwriting, records that load into your system or the real time per item, and those set the cost of the whole job.

A pilot measures all of that on your own material. Our guide to running a digitization pilot covers choosing the sample and writing the criteria. In the RFP, phase one needs six things stated:

  1. The sample: which boxes, volumes and sheets, including every format and the hardest items.
  2. The stages: handling, capture, text recognition, records, packaging and a test import.
  3. The report: target reports, CER by kind of text, field accuracy, defects and time per item.
  4. The criteria: what counts as a pass, published with the RFP.
  5. The decision: proceed, proceed with named changes or stop, and who decides.
  6. The price rule: unit prices hold for phase two unless a stated assumption proves wrong, and the RFP names those assumptions.

Example (fictional). The Aster County Historical Society makes Box 3 its phase one: the Okerlund family papers and the Wren Gulch Mercantile day book, 1893 to 1896, which is 69 items and 410 images. The RFP states the pass criteria in advance: no new damage, every image within the agreed star level, a measured CER for each of the two main hands, and every pilot record loading into the society’s collection system with its images. Bidders price the pilot as line 10 of the bid form, and the society can stop after it.

Keep assessment and bidding roles apart

One caution applies when federal money is involved. Under 2 CFR 200.319, “contractors that develop or draft specifications, requirements, statements of work, or invitations for bids must be excluded from competing on those procurements.” A vendor that wrote your specification during an assessment may not be able to bid on the work. A pilot that is phase one of a competitively awarded contract is different, because the vendor was chosen first. Ask your grants office before you mix the roles.

Write data ownership, retention and AI-use clauses into the RFP

The RFP should say who owns the images, text and records, how long the vendor may keep copies and how deletion is confirmed. It should also say which machine tools may process your material, where, and how their output is labeled, and that your files may not train third-party models without your written consent. These terms change what a bid is worth, so bidders must see them before they price.

Ownership

NEDCC’s leaflet lists “a brief statement on the ‘work for hire’ nature of the contract (organization maintains ownership of all project outputs)” among the clauses to include. Make “outputs” specific: master and access images, recognized text, records, target reports, logs and the pilot report. License terms for any software or hosted viewer belong in a separate clause, so that losing the software never means losing your files.

Retention and deletion

The RLG guidelines of 1997 already listed “vendor deletion of digital files upon acceptance by the institution” among the vendor responsibilities an RFP should spell out. A vendor copy kept for a short, stated period after acceptance is a useful safety copy until your own preservation copies are in place. After that, require deletion of working copies and backups, confirmed in writing.

AI and machine tools

Require every bidder to answer, in writing:

  • Which tools it would use for text recognition, cataloging suggestions and image tagging, whether each runs locally or as a cloud service, and in which country.
  • Who else sees the material: whether any third party receives your images, text or records.
  • How machine output is labeled in the delivered records, field by field.
  • How accuracy is measured and reported for each machine task.
  • Training: whether your files, records or corrections would be used to train or fine-tune any model.

Federal policy offers wording you can adapt. OMB Memorandum M-25-22 (April 3, 2025) directs federal agencies to ensure contracts “permanently prohibit the use of non-public inputted agency data and outputted results to further train publicly or commercially available AI algorithms, consistent with applicable law, absent explicit agency consent.” It binds federal agencies, not museums or counties, but the wording adapts well. The memo also names protections against lock-in such as “knowledge transfer, clear data and model portability practices, clear licensing terms, and pricing transparency.”

You may choose to allow some training, for example on public domain material. The point is that the choice is yours and is written down. Our own AI policy states how we label machine output and treat client material; ask every bidder for the same in writing.

Hosting and exit

If a bid includes hosting a search site, require a full export of files and records in open formats at any time and at the end of the contract, at a stated cost. Public bodies have one more reason to care: the Justice Department’s ADA Title II web rule covers content provided “directly or through contractual, licensing, or other arrangements” and adopts WCAG 2.1 Level AA. An interim final rule effective April 20, 2026 set compliance for April 26, 2027 (entities of 50,000 people or more) and April 26, 2028 (smaller entities and special districts).

Publish evaluation criteria and weights before bids arrive

Publish the criteria and their weights in the RFP, set pass or fail gates first, and score the rest with a written record for each proposal. Score price from the totals your bid form produces, not from headline rates.

The RLG guidelines put it plainly: “The institution should carefully establish criteria and a weighting of factors for evaluating proposals,” documented “so as to justify the final choice to the institution’s procurement office and to the unsuccessful bidders.” For grant-funded proposals, 2 CFR 200.320 requires the factors and their relative importance to be identified. The Smithsonian’s 2018 request is a short public example: best value, judged on experience or past performance, résumé and price, all three “of equal importance.”

Gates before scoring

Some requirements are pass or fail, and a proposal that misses one is not scored:

  • A complete bid form, every line priced or marked “not offered.”
  • Proof of the required insurance.
  • Acceptance of the data, retention and AI terms, or exceptions listed in writing, as the RLG guidelines also asked.
  • Signature, deadline and the required format.

An example of weights

The weights below are an example to adapt, not a standard. Check yours with your procurement office.

Criterion Example weight What earns points
Technical approach and compliance with the specification 25 Methods by material, target practice, file structure, quality plan
Sample or pilot results 20 Images, text and records measured against the stated criteria
Handling, custody and facility 15 Packing, transport, storage, condition reporting
Text and record approach, including AI disclosure 10 Engines named, accuracy reporting, labeling, training terms
Experience and references 10 Similar collections; references actually called
Price, from the bid form totals 20 Lowest total earns full points; others in proportion

One common way to score price gives the lowest total full points and other bids points in proportion: price points = 20 x lowest total / this bid’s total. Use your institution’s method if it has one. And call the references: NEDCC advises you to “Take the extra hour to call several references.”

Give bidders time, and answer everyone at once

NEDCC suggests budgeting “at least four to eight weeks for responses.” The RLG guidelines sketched thirteen weeks from release to purchase order. Collect questions in writing by a stated date and send every answer to every bidder. When proposals arrive, the vendor selection checklist gives you the questions to score them against.

Before you publish: a digitization RFP checklist

Each line closes a gap that would otherwise come back as a guess in someone’s bid.

  • Counts and condition: images and items by format, the basis for each count, exclusions and conservation flags.
  • Specification: FADGI category and star level per material, with a target report every session.
  • Files: formats, naming pattern with an example, folders, packaging and checksums.
  • Text: output formats, CER per batch on a random sample, correction priced separately.
  • Records: level of description, required fields, machine values labeled, a test import.
  • Custody and insurance: locations, subcontracting, transport, manifests, return in order, coverage and proof.
  • Acceptance: batch size, sample size, defect list, the acceptance rule, review window and rework.
  • Bid form: units and quantities on every line, inclusions stated, hourly lines capped, a validity date.
  • Pilot: sample, stages, report, criteria, decision and price rule.
  • Data and AI: ownership of every output, deletion, tools disclosed, no training without written consent, exit terms.
  • Evaluation: gates and weights published; answers sent to all bidders; four to eight weeks to respond.

If you are still deciding whether to bring in a vendor at all, start with in-house or outsourced digitization. To see how a phased engagement with a priced pilot can be set up, our engagements page describes ours, and you can ask us about a pilot box whether or not you go on to issue an RFP.

Working through this with a real collection?

If this article describes a decision you are facing, tell us about the collection. A pilot box answers most questions with your own material.

Plan a pilot box

Studio and contacts

24 South Weber Street, Suite 225
Colorado Springs, CO 80903
(719) 244-8450sales@dastok.com