An eLearning content development RFP is the document that decides whether you get four comparable quotes or four incomparable ones. A workable one has twelve sections: organisation and context; business objective and success measures; audience and volume; the state of your source material; scope of work with a per-module breakdown; technical and delivery format (name the standard and the edition, not "SCORM-compliant"); accessibility standard with the evidence you will accept; languages and localisation; the review and approval cycle with a defined number of revision rounds; IP, source files and exit; maintenance and support; and commercial terms with a stated budget range. Attach a scoring rubric and publish the weights before proposals arrive. Across five real eLearning RFPs published between 2024 and 2026 by ECMWF, the University of Arizona, ISO, the State Bar of California and AARFP, price carried between 10 and 30 per cent of the score and never more; instructional design and technical solution was the heaviest bucket in every single one, ranging from 20 to 50 per cent. If your rubric weights price above 30 per cent you are not running a quality competition, you are running a reverse auction with extra steps. The rubric in this article uses 30 per cent instructional design and process, 20 per cent technical and LMS delivery, 15 per cent accessibility, 15 per cent team and track record, 20 per cent commercial, with four pass or fail gates in front of it. The full Word template is a free, ungated download.
Key takeaways
- Publish your weights before proposals arrive, or you cannot defend the award. Erie County's 2026 web-based training RFP states plainly that criteria are “not necessarily listed in order of importance” and that it “reserves the right to weigh its evaluation criteria in any manner it deems appropriate.” That is legal, common, and it means the losing vendors have no idea what to fix and you have no written basis for the decision. Every strong RFP we reviewed published its percentages up front.
- Score the process, not the portfolio. Any vendor can show you three beautiful modules; showreels are selection bias made visible. What predicts your experience is the documented instructional design process with named review gates, who signs off at each gate, and what happens when a subject-matter expert misses a review. ISO weighted “service delivery methodology and project management” at 25% and “prototype capability” at 15%, ahead of company profile at 10%.
- Specify the standard and the edition, or you have specified nothing. “SCORM 2004” has four editions and the market is fragmented across them: Docebo supports 3rd Edition only and ignores sequencing rules, Cornerstone's connector accepts 3rd Edition, SAP SuccessFactors documentation lists 2nd and 4th, Workday lists 2nd, 3rd and 4th. Also specify the suspend_data budget — 4,096 characters on SCORM 1.2 versus 64,000 on SCORM 2004 3rd Edition — because a branched course with heavy bookmarking silently fails on 1.2.
- Source files are a contract deliverable or they are nothing. The strongest clause we found is the State Bar of California's: if the vendor uses proprietary software or file formats “that would prevent a second vendor from combining, editing, or otherwise working with the final Product,” that must be disclosed and alternatives offered, and all modules must be capable of being updated by the buyer or a third party. ISO goes further and names the artefacts: the individual SCORM files, the editable version of the course, and all visual and audio assets. Copy that language.
- Accessibility belongs in the rubric as its own weighted line, not as a checkbox. Two of the five real RFPs scored it separately — University of Arizona at 15% bundled with QA, AARFP at 15% as “IDEA and Accessibility Integration.” Ask for a completed Accessibility Conformance Report on VPAT 2.5Rev dated within twelve months, plus an independent third-party audit of one representative module, and name the criteria that actually break eLearning. A vendor claiming “WCAG compliant” without a version, a level and a date has told you nothing.
- Define a revision round in writing, because nobody else will. “Two rounds of revisions” means nothing until you say what a round is, at which gate it sits, how long the buyer has to consolidate feedback, what counts as a change of scope rather than a revision, and what an extra round costs. This is the single most common source of downstream disputes in content development, and it is entirely preventable with four sentences in the RFP.
- Anchor your budget on your own quoted hours, not on a 2010 benchmark. The Chapman Alliance ratios are 16 years old and have never been repeated; ATD's successor research deliberately abandoned historical comparison because real deliverables “are rarely an hour long.” Instead of asking for a per-hour rate, require hours by role and by phase — analysis, storyboard, media, development, QA, accessibility remediation, LMS integration, localisation — and define what “one finished hour” means in your RFP before anyone quotes it.
Most eLearning RFPs are written by someone buying custom course development for the first time, adapting a procurement template built for buying software. The document that comes out asks vendors for their company history, their methodology and their price, and almost nothing that predicts whether the courses will still be maintainable in eighteen months.
We read these from the receiving end for a living. The ones that produce good outcomes have three things the weak ones do not: a scope precise enough that two vendors quoting it are quoting the same work, a published scoring rubric so the award can be defended, and explicit language on who owns the source files. Everything else is formatting.
Below is a complete structure you can copy, a weighted rubric with score anchors, and the list of questions that separate vendors who will still be easy to work with in year three from the ones who will not. The full Word template is a direct download — no form, no email capture. If it is useful, link to it.
Two things this is not. It is not an LMS RFP: buying a platform is a different document with different criteria, and the two get conflated constantly. And it is not neutral in the sense of having no opinions — we build courses, so we have a commercial interest in this market. What we have tried to do instead is publish the questions we would least enjoy being asked, on the theory that a buyer who asks them gets a better vendor whether or not that vendor is us.
What this is based on
Rather than describe an RFP from memory, we pulled five real, publicly posted eLearning and course development RFPs issued between 2024 and 2026 and read what they actually demand:
- ECMWF, RFP/2025/385, eLearning Modules (July 2025) — a three-year contract to maintain and extend a library of 31 SCORM modules built in Articulate Storyline, up to four new modules a year at around 45 minutes each. Weights published: technical solution 30%, track record 20%, quality of resources 20%, price 30%.
- University of Arizona / ABOR, RFP #L312502, Course and Curriculum Development Services (June 2025) — three years plus two one-year options. Weights: curriculum and learning experience design 20%, instructional design and LMS technology 20%, media 20%, project management 15%, accessibility and QA 15%, pricing 10%.
- ISO, eLearning Course on Fundamentals of Conformity Assessment (June 2024) — roughly four hours of seat time, Articulate 360 preferred, SCORM 1.2 and 2004, WCAG 2.1 AA, budget capped at CHF 100,000. Three-stage evaluation with a pass or fail pre-qualification gate.
- State Bar of California, RFP E-Learning Services (October 2024) — technical and functional capability 50%, total cost 30%, responsiveness 10%, agreement with contracting terms 10%. Contains the best source-file clause we have seen in a public document.
- AARFP, Instructional Design RFP, Pool Operator Level 1 (March 2026) — instructional design and adult learning expertise 30%, methodology 20%, budget and value 20%, accessibility 15%, timeline and project management 15%. Proposals capped at five pages.
Two patterns hold across all five. Price is a minority weight everywhere, between 10 and 30 per cent. And every one of them addresses intellectual property explicitly, usually with work-made-for-hire language plus an assignment fallback. None of the template pages currently ranking for this search publishes an IP clause at all.
The twelve sections your RFP needs
1. Organisation and context
Who you are, what you train, how many learners, which business unit owns this budget, and who the decision makers are. Two sentences on why this project exists now. Vendors write materially better proposals when they understand the trigger — a regulator finding, a system migration, an acquisition, a failed previous vendor. Say which.
2. Business objective and how success will be measured
Not “improve compliance awareness.” State the behaviour that should change, the population it should change in, and the measure you will look at afterwards. If the honest answer is that the only measure is completion rate, say that too — it tells a good vendor to design for defensibility and a bad one to build a page-turner, and you will learn something from which one you get.
3. Audience, volume and constraints
Number of learners, roles, locations, languages, devices, bandwidth, and whether they take training on shift, at a desk, or on a shared terminal. Give the volume in finished minutes or modules, not in “hours of content,” and say whether that figure is fixed or a ceiling. Shared terminals and low bandwidth change the build, not just the delivery.
4. State of your source material — the section everyone underestimates
This is the single biggest driver of quote variance, and most RFPs leave it out entirely. Say plainly what exists: slide decks, policy PDFs, SOPs, recorded webinars, an SME's head, or nothing. Say whether it is current, who owns it, whether it has been through legal, and whether it is in the target language. “We have 200 slides that are three years old and nobody has signed them off” produces a realistic quote. Silence produces a cheap quote and a change order.
5. Scope of work, broken down per module
List the modules. For each one give a working title, target seat time, the interaction level you expect, whether it needs video, whether it needs assessment, and whether it is new build, conversion or refresh. The University of Arizona RFP is worth copying here: it prices new course development, full redevelopment, a 50 per cent refresh and strategic updates as four distinct products. That single distinction removes most of the ambiguity from maintenance pricing later.
6. Technical requirements and delivery format
“SCORM-compliant” is not a requirement. This is:
| Specify | Why it matters |
|---|---|
| Standard and edition — e.g. “SCORM 1.2 as the primary deliverable; SCORM 2004 3rd Edition as an alternate; cmi5 optional” | SCORM 2004 has four editions and support is fragmented. Docebo takes 3rd Edition only and ignores sequencing; Cornerstone's connector accepts 3rd Edition; SAP SuccessFactors documentation lists 2nd and 4th; Workday lists 2nd, 3rd and 4th. AICC is still accepted by several enterprise LMSs but the AICC dissolved in 2014, so there is nobody to raise a defect with — treat it as legacy support on request, never as a target. |
| Target LMS, version and tenant, plus a mandatory test import before design sign-off | Almost every packaging problem is LMS-specific. Workday will not accept SCORM or AICC content in a standalone lesson — it has to sit inside a course structure, which quietly breaks a micro-learning plan. Testing in SCORM Cloud is not testing in your tenant. |
| suspend_data budget and file count | SCORM 1.2 allows 4,096 characters of suspend data; SCORM 2004 3rd Edition allows 64,000. A branched course with heavy bookmarking fails silently on 1.2. Docebo also recommends no more than about 15,000 files per package, which matters for image-heavy or heavily localised builds. |
| Completion, success and score logic stated separately | SCORM 1.2 conflates completion and success. Write down what “complete” means, what “passed” means, the mastery threshold, and require it demonstrated in your LMS rather than described in a proposal. This is the most common post-launch defect in the field. |
| xAPI and LRS ownership if in scope | The standard is xAPI 2.0 (IEEE 9274.1.1-2023, also ISO/IEC/IEEE 39274-1-1:2025 since October 2025). The market is still xAPI 1.0.3 — that is what Cornerstone's connector documents and what cmi5 normatively requires. Ask for “1.0.3 minimum, state your 2.0 readiness.” Then say who supplies the LRS, where statements land, and who owns retention and PII. This is the most common cost surprise in the category. |
| cmi5 only if you have somewhere to run it | cmi5 (“Quartz” 1st Edition, June 2016) fixes real problems — content need not be same-origin with the LMS, so mobile and offline become tractable — but it requires an LRS, and none of Workday, Cornerstone, SAP SuccessFactors or Docebo officially supports it. Mandating cmi5 into those estates means middleware such as Rustici Engine. Budget it or drop the requirement. |
| Self-contained, HTTPS-only packages with a declared external dependency list | Any HTTP-referenced asset is blocked as mixed content. Fonts, CDNs and analytics called from inside a package break offline and mobile playback and create a privacy review you did not plan for. |
| Browser and device matrix | The State Bar of California specified current plus the previous three major versions of Edge, Safari, Firefox and Chrome, desktop and mobile. Borrow that sentence. |
If you are unsure which standard you actually need, our SCORM versus xAPI comparison works through it, and the nine causes of a SCORM course reporting “incomplete” is a good list of the defects your acceptance test should be designed to catch.
7. Accessibility standard and the evidence you will accept
State a version, a level and a date: WCAG 2.2 Level AA for new builds. WCAG 2.2 has been a W3C Recommendation since 5 October 2023 and adds six Level A and AA criteria beyond 2.1. Two of them routinely break eLearning: 2.5.7 Dragging Movements, which requires a single-pointer alternative to every drag-and-drop interaction, and 2.5.8 Target Size, which sets a 24×24 CSS pixel minimum on hotspots and drag targets. Name them in the RFP and the vendor cannot claim they were out of scope.
Know which legal standard binds you, because they have not converged:
- US federal (Section 508) — still the 2017 ICT Refresh, referencing WCAG 2.0 Level A and AA. No newer rulemaking is listed on Section508.gov as of July 2026.
- US state and local (ADA Title II) — WCAG 2.1 Level AA. The compliance dates were extended by one year in a DOJ interim final rule published 20 April 2026: now 26 April 2027 for populations of 50,000 or more, and 26 April 2028 for smaller entities and special districts.
- US healthcare and other federal-assistance recipients (HHS Section 504) — WCAG 2.1 AA, also extended one year on 7 May 2026 to 11 May 2027 (15 or more employees) and 10 May 2028 (fewer than 15).
- EU (European Accessibility Act) — applying since 28 June 2025. The harmonised standard remains EN 301 549 V3.2.1, which is WCAG 2.1; V4.1.0, which moves the baseline to WCAG 2.2, existed only as a final draft as of June 2026 and has not been cited in the Official Journal. Note that workplace training is not itself a listed EAA service — courses are usually caught through e-commerce or e-books and dedicated software. If your programme touches EU consumers, that is a question for counsel, not for your vendor.
For evidence, ask for three things: a completed Accessibility Conformance Report on VPAT 2.5Rev (April 2025, INT edition if you span jurisdictions) dated within twelve months and covering the player as well as the content; an independent third-party audit of one representative module containing each interaction pattern you are buying, with the auditor named and the remediation window contractual; and an explicit assistive technology test matrix — screen reader, browser and OS combinations, keyboard-only, and 200% and 400% zoom. A VPAT is self-attested by default. Ask who filled it in.
8. Languages and localisation
Name every language you will need within eighteen months, not the ones you need at launch; declaring them up front is nearly free and expensive to retrofit. Then specify: XLIFF round-trip export and import, with the translation memory delivered to you at project end; all localisable text extractable, with nothing baked into images or burnt into video; layered source files for graphics; a glossary approved before translation starts; human voiceover versus synthetic priced separately per language, with audio masters and timed scripts as deliverables; caption format named (WebVTT for HTML5 players, TTML2 where broadcast interchange matters, SRT only as a working format); and a named in-country reviewer per locale whose sign-off is a payment gate.
One layout rule saves more money than any of those. W3C's internationalisation guidance, citing IBM, puts average text expansion from English into European languages at 200 to 300 per cent for strings up to ten characters, falling to about 130 per cent for strings over seventy. Button labels, tab headings and drag targets need two to three times the headroom at storyboard stage. Requiring pseudo-localisation testing before English sign-off costs a day and prevents a per-language layout rebuild.
9. Review cycle, revision rounds and approval gates
Write down the gates — outline, storyboard, alpha, beta, gold — and for each one: who reviews, how long they have, how feedback is consolidated, and what happens if they miss the window. Then define a revision round properly. How many are included at each gate. What counts as a revision versus a change of scope. What an additional round costs. Whether unconsolidated or contradictory feedback from multiple reviewers counts as one round or several. Four sentences here prevent the most common dispute in this industry.
10. Intellectual property, source files and exit
Three clauses, all of them borrowed from real documents.
Ownership. The University of Arizona: everything developed “belongs solely and exclusively” to the buyer and is “considered a work for hire under the U.S. copyright laws.” Add an assignment fallback for anything that does not qualify as work for hire, because in many jurisdictions most of it will not.
Pre-existing IP. The vendor keeps its own templates, engines and libraries, but grants you — in ECMWF's words — “a free, non-exclusive, irrevocable, worldwide, transferable, sub-licensable and time unlimited licence” to whatever is embedded in your deliverables. Without this clause you own a course you cannot legally modify.
Editable source and no lock-in. ISO names the artefacts: the final individual SCORM files, “the editable version of the course, as well as all the visual and audio assets used throughout the course.” The State Bar of California adds the anti-lock-in test: if the vendor uses proprietary software or file formats “that would prevent a second vendor from combining, editing, or otherwise working with the final Product,” that must be stated and alternatives offered, and every module must be capable of being updated by the buyer or a third party. Together those two sentences are the most valuable thing in this article.
11. Maintenance, support and versioning
Content is not a project, it is an estate. Specify the annual update allowance in days or modules (ECMWF budgeted roughly 24 FTE days of routine maintenance across a three-year contract), the SLA for a regulatory correction, the versioning and archive policy, and — the one everyone forgets — what happens when the authoring tool version changes. Ask directly: if Articulate ships a new version and my file no longer opens in the version I hold, what is your obligation?
12. Commercial terms, timeline and submission mechanics
Publish a budget range. Vendors price to a range they can see and pad against one they cannot, and a range filters out the bids that were never viable. ISO published a CHF 100,000 cap and got proportionate proposals.
Require pricing as hours by role and by phase — analysis, storyboard, media production, development, QA, accessibility remediation, LMS integration, localisation, project management — rather than a blended per-hour or per-minute rate. Ask for day rates alongside, so change orders have a pre-agreed price. And define what “one finished hour” means in your document before anyone quotes against it.
Constrain the submission the way the good RFPs do: a page limit (ISO 4 pages, AARFP 5 pages excluding appendices, State Bar of California a 5-page executive summary), named CVs for the people who will actually do the work, three contactable client references, portfolio samples with live access rather than screenshots, and cost in a separate file so technical scoring can be completed blind.
The scoring rubric
This is the part almost nobody publishes, and it is why weak proposals win. Two gates and five weighted criteria.
Stage 1 — pass or fail. No score, no negotiation.
| Gate | Pass condition |
|---|---|
| Source files and IP | Vendor accepts, without amendment, that the buyer owns the deliverables, receives editable source files and all assets, and may have the content updated by a third party. |
| Accessibility evidence | A dated ACR on VPAT 2.5Rev is supplied, and the vendor accepts an independent third-party audit of one module as an acceptance condition. |
| LMS proof | Vendor agrees to a test import of a working module into the buyer's own LMS tenant before design sign-off, with completion and score logic demonstrated. |
| References | Three contactable clients, at least one on a contract running two years or longer. ISO required three prior contracts with international organisations as a pre-qualification gate; the University of Arizona required security documentation such as SOC 2 Type II or HECVAT before any scoring. |
Stage 2 — weighted scoring, 100 points
| Criterion | Weight | What a 5 looks like |
|---|---|---|
| Instructional design and delivery process | 30 | A documented process with named gates and named sign-off roles; a design rationale tied to the stated behaviour change rather than to content coverage; a worked treatment of one of your modules, not a generic sample; assessment design that tests application rather than recall; a stated policy for what happens when an SME misses a review. |
| Technical and LMS delivery | 20 | Names your LMS, its version, and the specific constraints it imposes without being prompted; states suspend_data budget and file count per module; explains completion and success logic; declares external dependencies; commits to a test import date. |
| Commercial and total value | 20 | Hours by role and phase; day rates for change orders; maintenance priced for years two and three; localisation priced per language; nothing material listed as “TBC.” Score value, not lowest number. |
| Accessibility capability | 15 | Names WCAG 2.2 criteria by number and explains how their interaction patterns satisfy them; supplies an audit report rather than a claim; has an accessibility specialist on the named team; treats remediation as a phase with hours attached rather than as an assumption. |
| Team and track record | 15 | Named individuals with CVs and their percentage allocation to your project; the proposal team and the delivery team are the same people; references on multi-year contracts at comparable scale; evidence of surviving a client-side reorganisation or a regulatory change mid-contract. |
Score anchors — use the same five words for every criterion
| Score | Anchor |
|---|---|
| 5 — Exceeds | Specific, comprehensive, evidenced against our project. Tells us something about our own programme we had not written down. |
| 4 — Strong | Reasonably comprehensive with sufficient detail; evidenced, but generic in places. |
| 3 — Adequate | Addresses the criterion but is non-specific and lacks detail. Could have been written for any buyer. |
| 2 — Weak | Misses several elements, or contains statements that are inaccurate or contradicted elsewhere in the proposal. |
| 0–1 — Fails | Does not address the criterion, or restates the question back at us. |
Multiply each 0–5 score by the criterion weight, divide by 5, and you have weighted points out of 100. Score technical criteria before anyone opens the cost file. Set a minimum technical threshold below which cost is never opened — ISO used 80 out of 100 on both its technical and financial stages; Alameda County Superior Court used 70 per cent of available technical points, and published every vendor's score afterwards. Publishing the scores is good practice and costs you nothing.
The questions vendors hope you will not ask
These are ordinary questions. They are uncomfortable only because the honest answers are inconvenient, which is exactly what makes them useful. Put them in the RFP as a required response section, not in the meeting, so the answers are written and attached to the contract.
- How many revision rounds are included, and what precisely is a round? Ask for the definition, not the number. Follow up with: what does round three cost, and does contradictory feedback from two reviewers count as one round or two?
- Who owns the source files, and what exactly will I receive? Ask for the file list by extension. “All project files” is not an answer. Then ask whether any of it requires a licence you do not hold to open.
- What happens when the authoring tool version changes? Will my file still open in the version I own? Who pays to migrate it? This one is rarely in any contract and it is where estates quietly die.
- Is the team that wrote this proposal the team that will deliver it? Ask for names, percentage allocation, and a contractual notification requirement for substitutions. The gap between pitch team and delivery team is the most reliable predictor of a bad year.
- What is your real turnaround, measured from kickoff to gold, on your last three projects of this size? Not the quoted timeline — the actual one, with the reason for any slip. A vendor who has never slipped has either not delivered much or is not telling you.
- Can I speak to a client on a multi-year contract, not a recent launch? Anyone can supply a delighted reference from month two. Year three references are where maintenance quality, change-order behaviour and staff turnover show up.
- Which parts of this will be subcontracted, to whom, and in which country? Then ask how the accessibility and QA passes are handled across that boundary, and who carries the data processing obligation.
- Where do you use machine translation or generative AI, and how is it disclosed? Both are legitimate and both are widely used silently. ISO 18587 exists specifically to give you contract language for machine translation post-editing. Ask for the disclosure, and ask who performs the human validation pass and what they are qualified in.
- What is your defect rate at acceptance, and how do you count a defect? A vendor who measures this will tell you. A vendor who does not has just told you something too.
- What have you refused to build, and why? The best answer we ever heard to this was a vendor explaining that they had talked a client out of a simulation because the underlying process was about to change. A vendor with no examples either has no judgement or does not exercise it.
- If we ended the contract after phase one, what would we walk away with? Ask them to describe the deliverable state at each exit point. The answer tells you whether your payment schedule is aligned to value received or to their cash flow.
What enterprise RFPs demand that yours probably does not
If you are buying at scale — multi-year, multi-region, integrated into an HCM — the real documents in this space consistently include several things that mid-market RFPs leave out:
- Enterprise-scale delivery history with multi-year references. ISO made three prior contracts with international organisations a pre-qualification gate, and required CVs showing a minimum of five years' relevant experience.
- Named LMS and HCM integration expertise. Workday, Cornerstone, SAP SuccessFactors and Docebo each impose different constraints. Ask which the vendor has personally shipped into and how many packages, and score the answer.
- A documented instructional design process with review gates, not a methodology diagram. ISO weighted delivery methodology at 25 per cent and prototype capability at 15 per cent, ahead of company profile at 10.
- Multilingual capacity stated as a number — how many locales concurrently, with what in-country review model, at what quality standard (ISO 17100 for human translation, ISO 18587 where machine translation post-editing is used).
- Flexible capacity or pod models with named roles and stated ramp time, so a spike does not become a queue. Ask what happens if you double the volume in month four.
- Security and data attestations before scoring. The University of Arizona required SOC 2 Type II, HITRUST or HECVAT documentation as a limiting criterion, submitted as confidential.
- Technical integration testing before signature, not after design sign-off. This is the cheapest risk reduction available to you and almost nobody asks for it.
Four red flags worth scoring down
- No structured planning phase. A proposal that moves from kickoff to development without analysis and storyboard gates is quoting a production job, not a design job. It will be cheaper and it will produce content that covers the material and changes nothing.
- No reference on a long-term contract. Plenty of vendors can win and deliver a first project. Ask for year three.
- Vague IP or source-file language — “deliverables will be provided,” “standard industry terms,” or silence. This is almost never an oversight.
- An award decided mainly on lowest cost. Not one of the five real RFPs we read weighted price above 30 per cent, and the University of Arizona put it at 10. If your rubric weights price at 50, the rest of the rubric is decoration.
A note on budget benchmarks
You will be offered a per-finished-hour figure by almost every source you consult. Treat all of them with suspicion. The most-cited dataset in the industry — Chapman Alliance, 249 organisations and 3,947 learning professionals — was published in September 2010 and reports 79:1 development hours per finished hour for basic eLearning, 184:1 for interactive and 490:1 for simulation. It has never been replicated. ATD's successor research, published in January 2021, deliberately abandoned historical comparison and switched from “one hour of training” to “units of training content,” on the grounds that real deliverables “are rarely an hour long.” Every 2026 cost-per-hour page we could find was vendor marketing without a methodology or a sample.
So do not anchor your budget on a ratio. Anchor it on your own quoted hours by role and phase, compare those across four bidders, and interrogate the differences. That comparison tells you far more than any benchmark: two vendors quoting the same total with wildly different QA and accessibility hours are not offering the same product. For the video component specifically, our per-minute rate breakdown for live-action, animated and AI-avatar production gives current bands from named sources.
- 1Week 0 — decide what you are actually buying, and write the failure down
Before drafting, answer three questions in writing: what behaviour should be different after this training, what already exists that we can build on, and what went wrong last time. That third answer is the most useful sentence in the whole RFP. If the previous vendor delivered late, or the content could not be edited afterwards, or the SME never reviewed anything, say so — it becomes a requirement rather than a repeat. Set the budget range now, internally, even if you debate publishing it.
- 2Week 1 — audit your source material honestly
Open the folders. Count the decks, the PDFs, the recordings. Check what is current, what has legal sign-off, what exists only in someone's head, and what is already in the target languages. Write the inventory into the RFP as an appendix. This single step removes more quote variance than everything else combined, and it is the step buyers skip because it is dull and slightly embarrassing.
- 3Week 1 — lock the technical requirements with your LMS admin in the room
Not your L&D team alone. Your LMS administrator knows the tenant version, the upload limits, the SSO model, whether the mobile app is in use and what has broken before. Get the standard, the edition, the completion logic, the suspend_data budget and the browser matrix written down here. Fifteen minutes with the admin now prevents a six-week integration argument later.
- 4Week 2 — draft the rubric before you draft the questions
Write the weights first and let them drive what you ask for. If accessibility carries 15 per cent of the score, there must be enough in the RFP for a vendor to earn or lose 15 points on it. Rubric-first drafting produces a much shorter, much sharper document, because every question has to justify its existence against a weight.
- 5Week 3 — issue, with a real question window and published answers
Give bidders two to four weeks and a genuine clarification window — ECMWF ran ten days for questions and published answers three days later, then allowed a month to respond. Circulate every answer to every bidder. Cap the proposal length (four to five pages plus appendices is plenty) and require cost in a separate file so technical scoring can be done blind.
- 6Week 5–6 — score independently, then calibrate as a panel
Each evaluator scores alone against the anchors before anyone discusses anything. Then meet, and only discuss the criteria where scores differ by two or more — those gaps are where the real information is, and they usually mean two people read the same paragraph as answering different questions. Record a one-line justification per criterion per vendor. Do not open the cost files until technical scoring is signed off.
- 7Week 6–7 — make the shortlist prove it, do not let them present it
Replace the standard vendor pitch with two exercises. First, give the top two a real extract of your source material and ask for a one-screen treatment and a storyboard page — paid, if the effort is meaningful. Second, require a working module imported into your own LMS tenant with completion and score logic demonstrated. Roughly half the problems that surface in month four surface in that import instead, when you still have leverage.
- 8Week 8 — award, debrief, and get the clauses into the contract
Publish the scores to the bidders. It costs nothing, it improves the proposals you get next time, and it is your written defence of the decision. Then check that the IP, source-file, third-party-updatable, accessibility-audit and revision-round definitions actually made it from the RFP into the signed agreement. They are frequently negotiated out during redlines by people who were not in the evaluation.
Signals worth scoring up
- Names your LMS, its version and its specific constraints without being asked, and volunteers what will be awkward about it
- Proposes a test import into your tenant before design sign-off, rather than agreeing to one when you insist
- Quotes hours by role and by phase, with accessibility remediation and QA as separate visible lines
- Accepts the source-file and third-party-updatable clause without amendment, and lists the file types you will receive
- Names the delivery team with CVs and percentage allocation, and commits to notifying you of substitutions
- Offers a reference on a contract running two years or more, unprompted
- Cites WCAG success criteria by number and explains how their interaction patterns satisfy them
- Tells you something in your own brief is a bad idea, and says why
- Prices maintenance for years two and three before you ask about them
- Discloses where machine translation or generative AI is used and who performs the human validation
Red flags worth scoring down
- “SCORM-compliant” with no version, edition or target LMS named anywhere in the proposal
- “WCAG compliant” with no version, no level, no date and no conformance report attached
- No analysis or storyboard gate — the plan runs from kickoff straight into development
- Silence or vagueness on IP, or source files described only as “project files will be provided”
- Cannot produce a reference on a long-term contract, only recent launches
- The proposal team and the delivery team are visibly different people, and substitution is unrestricted
- Revision rounds quoted as a number with no definition of what a round is or what an extra one costs
- Per-finished-hour or per-minute pricing with no breakdown, quoted before anyone read your source material
- Timeline with no dependency on your SME availability — it means they have not planned for the real bottleneck
- Materially cheapest bid with the same stated scope: check the QA, accessibility and project management hours before celebrating
Adjusting the weights to your situation — three common profiles
The 30/20/20/15/15 split is a sensible default, not a law. Three adjustments cover most real situations. If this is regulated or safety-critical content, move 5 points from commercial into accessibility and 5 from team into instructional design, and make an independent audit a payment gate rather than an acceptance condition — a defect found by a regulator costs far more than the points you moved. If this is a large multilingual rollout, carve a sixth criterion worth 10 points out of instructional design and technical for localisation capacity, and score it on concurrent locale count, in-country review model and translation memory handover, because at 15 or 20 languages localisation process is the project. If you are a first-time buyer with no internal instructional design capability, move 5 points from commercial into team and track record and add a hard reference gate: you are buying judgement you cannot yet evaluate directly, so buy it from people other buyers have kept for years. What we would not do in any of these cases is raise price above 30 per cent. Not one of the five real RFPs behind this article did, and the one that went lowest — the University of Arizona at 10 per cent — was buying the largest and longest engagement of the five.
Two moves that change the answers you get, and cost you nothing
First: publish a budget range. The objection is always that vendors will quote to the top of it, and some will — but a range filters out bids that were never viable, stops good vendors from no-bidding because they assume you are shopping on price, and makes every proposal comparable because they are all solving the same constrained problem. ISO published a CHF 100,000 cap and got proportionate proposals. A hidden budget mostly buys you four quotes you cannot compare. Second: replace the vendor presentation with a paid micro-pilot. Hand the top two a genuine extract of your source material and ask for one screen treatment, one storyboard page, and a working module imported into your own LMS tenant with completion and score logic demonstrated. Pay for it if the effort is real. You will learn more in that week than in every reference call combined, because you are watching the actual delivery team work with your actual material inside your actual system — and roughly half of the problems that would otherwise surface in month four surface here instead, while you still have leverage and have not signed anything.
Frequently asked questions
What should an eLearning RFP actually contain?
Twelve sections: organisation and context; business objective and success measures; audience, volume and constraints; the state of your source material; scope of work broken down per module; technical and delivery format with the standard and edition named; accessibility standard plus the evidence you will accept; languages and localisation; review cycle with a defined revision round; IP, source files and exit; maintenance and versioning; and commercial terms with a published budget range, submission mechanics and the scoring rubric. The two that get left out most often are the source-material audit and the source-file clause, and those two cause most of the trouble afterwards — the first drives quote variance, the second decides whether you can ever change vendor.
How should I weight the scoring criteria?
Our default is instructional design and process 30, technical and LMS delivery 20, commercial and total value 20, accessibility 15, team and track record 15, with four pass or fail gates in front of it. That sits inside the range real buyers use: across five published eLearning RFPs from 2024 to 2026, price carried 10 to 30 per cent and never more, while instructional design and technical solution was the heaviest bucket in every one, from 20 to 50 per cent. The important thing is less which weights you choose than that you publish them before proposals arrive and score against written anchors. An unpublished rubric cannot be defended and teaches losing bidders nothing.
Should I gate the RFP template behind a form?
We did not, and we would suggest you do not gate yours either. The two most visible templates in this category take opposite approaches — d'Vinci publishes a direct Word download with no form, while the same document syndicated through ATD's vendor directory sits behind an email opt-in. A procurement document is not a lead magnet: the people who most need it are often at the start of a process where they are deliberately not talking to vendors yet, and a form filters for the ones already willing to be sold to. Ungated documents also get linked to, which is worth more over time than the contact details.
How specific do the technical requirements need to be?
More specific than feels natural. “SCORM-compliant” is not a requirement because SCORM 2004 has four editions and support is genuinely fragmented — Docebo takes 3rd Edition only and ignores sequencing rules, Cornerstone's connector accepts 3rd Edition, SAP SuccessFactors documentation lists 2nd and 4th, Workday lists 2nd, 3rd and 4th. Name the standard and the edition, the target LMS and its version, the suspend_data budget (4,096 characters on SCORM 1.2 versus 64,000 on 2004 3rd Edition), the completion and success logic, the browser and device matrix, and require a test import into your own tenant before design sign-off. If xAPI is in scope, also state who supplies the LRS and who owns statement retention and PII — that is the most common cost surprise in this category.
Who should own the source files, and what exactly should I ask for?
You should, and you should name the artefacts rather than asking for “project files.” ISO's language is the model: on sign-off the supplier hands over the final individual SCORM files, the editable version of the course, and all visual and audio assets used throughout. Pair it with the State Bar of California's anti-lock-in test — if the vendor uses proprietary software or file formats that would prevent a second vendor from editing the final product, that must be disclosed and alternatives offered, and every module must be capable of being updated by you or a third party. Add a pre-existing IP clause so anything the vendor embeds from its own libraries comes with a perpetual, irrevocable, transferable licence. Without that last clause you can own a course you are not permitted to modify.
What accessibility standard should I specify in 2026?
WCAG 2.2 Level AA for new builds, stated with the version, the level and a date. WCAG 2.2 became a W3C Recommendation on 5 October 2023 and adds six Level A and AA criteria beyond 2.1, of which 2.5.7 Dragging Movements and 2.5.8 Target Size are the two that routinely break eLearning interactions. Check which standard legally binds you, because they have not converged: US federal Section 508 still references WCAG 2.0 AA; ADA Title II requires WCAG 2.1 AA with deadlines extended in April 2026 to 26 April 2027 and 26 April 2028; HHS Section 504 also moved out a year to 11 May 2027 and 10 May 2028; and in the EU the harmonised standard under the European Accessibility Act remains EN 301 549 V3.2.1, which is WCAG 2.1. Ask for a completed ACR on VPAT 2.5Rev dated within twelve months plus an independent audit of one representative module — a VPAT is self-attested unless you say otherwise.
How long should the whole RFP process take?
Eight to ten weeks from first draft to award is realistic for a mid-sized content programme, and the real documents bear that out: ECMWF ran from issue on 14 July 2025 to a 25 August deadline, evaluation through late September and contract start on 17 October. Give bidders two to four weeks to respond with a genuine clarification window, allow two weeks for independent scoring and panel calibration, and one to two weeks for the shortlist exercises. The stage worth protecting is the source-material audit before you issue — an RFP written without it produces quotes that are not comparable, and no amount of evaluation rigour later fixes that.
The short version
A good eLearning RFP does three things a bad one does not. It describes the source material honestly, so the quotes are comparable. It publishes its weights, so the award is defensible and the losing bidders learn something. And it settles ownership of the source files in writing, so the decision you are making is which vendor to start with rather than which vendor to be stuck with.
Everything else — the section order, the page limits, the exact percentages — is adjustable. Those three are not.
Download the eLearning RFP template and scoring rubric (Word, free, no form) — twelve sections with prompts, the weighted rubric with score anchors, the IP and source-file clauses, and the vendor question set as a required response section.
Related reading:
- SCORM vs xAPI: which standard does your LMS need in 2026 — for section 6 of your RFP.
- Why your SCORM course shows “incomplete” — nine causes, and a good basis for your acceptance test.
- Training video production cost in India 2026 — per-minute bands if video is in scope.
- AI avatar training videos: when they work and when they fail — worth reading before you specify presenter-led video.
- Course and content development — what we do, if you want to know who wrote this.
Sources
- ECMWF, RFP/2025/385 eLearning Modules, July 2025 — published weights, IP transfer on acceptance, maintenance day budget.
- Arizona Board of Regents, RFP #L312502 Course and Curriculum Development Services, June 2025 — 10% pricing weight, limiting criteria, work-for-hire clause.
- ISO, RFP, eLearning Course on Fundamentals of Conformity Assessment, June 2024 — three-stage evaluation, source-file handover language, CHF 100,000 cap.
- State Bar of California, RFP E-Learning Services, October 2024 — 50/30/10/10 weights, rights in work product, proprietary-format disclosure clause.
- AARFP, Instructional Design RFP, March 2026 — weights including 15% accessibility, five-page limit.
- Erie County NY, RFP 2026-035VF Web-Based Training, April 2026 — the unweighted counter-example.
- W3C, What's New in WCAG 2.2 — the six new A and AA criteria.
- US DOJ, Extension of Compliance Dates, ADA Title II web rule, Federal Register, 20 April 2026.
- US HHS, HHS extends mobile and web accessibility deadline, 7 May 2026.
- European Commission, Web Accessibility Directive standards and harmonisation — EN 301 549 V3.2.1 as the cited version; Directive (EU) 2019/882 for EAA scope and transition dates.
- ITI, VPAT 2.5Rev, April 2025 — ACR editions and what an ACR is.
- Rustici, SCORM versions; Workday Learning components; Cornerstone Course Asset Connector; Docebo SCORM limits — edition support, suspend_data and package constraints.
- IEEE, 9274.1.1-2023 (xAPI 2.0) and ISO/IEC/IEEE 39274-1-1:2025; cmi5 FAQ and adoption list.
- W3C, Text size in translation — the 200–300% expansion figures for short strings.
- Brian Chapman, How long does it take to develop training?, Chapman Alliance, September 2010; ATD, How Long Does It Take to Develop Training? New Question, New Answers, 13 January 2021.
Bring the RFP you are drafting, or the one you just received. We will go through the rubric with you and tell you where your scope is under-specified — including the sections that would make it harder for us to win it.