Translation RFP: The Complete Guide & Free Toolkit for 2026
Last updated: 07/19/2026 by Luciano Oliveira, Esq., co-founding partner of The Translation Company Group LLC.
Why trust this guide over the dozen vendor checklists ranking for this search? Because it was written from both sides of the table. The Translation Company Group is a WBENC-certified, SAM-registered US translation company that has answered corporate and government language-services RFPs for two decades, and this toolkit was drafted by a lawyer who has read the fine print buyers usually discover after award. It is also the only translation RFP resource we know of that is legally current for 2026: Executive Order 14224 (2025) designated English the national language and revoked EO 13166, while competing guides either skip the legal layer entirely or still cite the revoked order. Finally, it ships the two tools no one else offers: a bid normalizer that makes structurally incomparable per-word quotes comparable, and a red-flag answer key that tells your evaluators what a disqualifying vendor answer sounds like. Everything downloads free, with no email gate.
Download the toolkit: Translation RFP Template (Word) · Vendor Scorecard + Bid Normalizer + 45-Question Bank (Excel). Use them as-is or adapt them; attribution appreciated.
Important: this is general information only and not legal or tax advice. For important business decisions, you must secure your own legal and tax counsel.
What Is a Translation RFP, and Do You Actually Need One?
A translation RFP (request for proposal) is a structured document a company or government agency issues to shortlist and select a language service provider (LSP). It defines your languages, volumes, content types, quality requirements, security constraints, and commercial terms, then asks competing vendors to respond in a comparable format. A good RFP does two jobs at once: it forces you to specify what you are actually buying, and it forces vendors to commit, in writing, to terms you can hold them to for years.
You need a full RFP when translation spend is material (roughly $50,000+ per year), when content carries regulatory or legal risk, when you are consolidating scattered departmental buying into a program, or when procurement rules require competition — always the case for public agencies above simplified acquisition thresholds. You do not need one for a single document or a first small project; for those, a quote request plus a look at how translation services are priced will get you a fair deal faster. A practical middle path for mid-size buyers is an RFP-lite: send the rate grid, the question bank, and a paid pilot to three vendors and skip the ceremony. The toolkit supports both formats.
How to Write a Translation RFP, Step by Step
Start from your content, not from a template’s table of contents. Inventory what you translate, then split it into stakes tiers: Tier 1 is content where an error creates legal, regulatory, or safety exposure (contracts, labeling, patient materials, filings); Tier 2 is customer-facing content where quality drives revenue (manuals, support, marketing); Tier 3 is internal or ephemeral content. The tiers matter because they should be priced and workflowed differently — full two-linguist translate-edit-proofread (TEP) for Tier 1, possibly full post-edited machine translation for Tier 3 — and because vendors who quote one blended rate for everything are either overcharging your Tier 3 or under-protecting your Tier 1.
Specify languages like a professional buyer. “Spanish” is not a specification: Mexican, US-domestic, and Iberian Spanish read differently, and expansion rates differ enough to move budgets. Name the variant, the direction, and the market. If you operate in regulated verticals, say so in the scope: medical, legal, and technical content each carry qualification requirements a generalist pool cannot meet.
Disclose your linguistic assets. If you have an existing translation memory, glossary, or style guide, say so and attach word counts. Vendors price against leverage: an RFP that hides a 200,000-segment TM invites bids that ignore the discount you already paid to build. If you have no assets, the RFP is where you require that everything built from your work becomes yours — more on ownership below, because this is the sector’s quietest trap.
Then let the toolkit do the technical parts. The template carries the sections buyers rarely know to include: the AI-disclosure schedule, the confidentiality flow-down, the standardized rate grid, the exit clause. The workbook carries the evaluation machinery. Your job is the scope; the toolkit’s job is making sure nothing signable is missing.
The Problem Nobody Warns You About: Translation Quotes Are Not Comparable
Here is the industry’s open secret. Almost every vendor prices per word against a “fuzzy match grid”: new words at full rate, previously translated or similar segments (“fuzzy matches” and “100% matches”) at discounts. But every vendor defines the bands differently, assumes different leverage from translation memory, counts source or target words as it prefers, and layers minimum job fees, project-management percentages, desktop publishing, engineering, and rush premiums on top. The result is that the per-word rates on page one of five proposals are five different units of measurement. Comparing them directly is the single most common — and most expensive — mistake in translation procurement.
The fix is to refuse to compare vendor-shaped numbers. Our RFP template issues a standardized rate grid: your match bands, your definitions, source-word basis, every ancillary fee disclosed on one schedule, with a warning that non-conforming bids get normalized or rejected. The workbook’s Bid Normalizer then prices every completed grid against the same reference corpus — a CAT-tool analysis of your real content mix — and computes each vendor’s true blended rate and total cost of ownership, minimums and surcharges included.
The worked example in the workbook shows why this matters. Vendor B advertises $0.12 per word, Vendor C $0.16 — yet on a realistic 10,000-word corpus with a year of minimum fees, PM percentages, and DTP, Vendor C lands at $6,665 all-in while “cheap” Vendor B costs $7,253. The vendor with the lowest headline rate is the most expensive bid on the table. Until you run this arithmetic, a translation RFP is a beauty contest; with it, it becomes procurement. For what fair per-word ranges look like in 2026 before any discounts, see our translation rates guide — this page deliberately doesn’t restate rate tables.
Who Owns the Translation Memory? (Ask Now, Not at Exit)
Every project your vendor completes builds a translation memory — a database of your source sentences paired with approved translations — plus term bases and style decisions. That asset is why year-three translation costs less than year-one: repetitions and matches get discounted against it. It was built with your money. And yet, in a remarkable share of vendor agreements, it quietly lives in the vendor’s systems, exportable on request, for a fee, in a format of their choosing, eventually. Buyers discover the lock at the worst moment: when switching vendors, when the incumbent’s leverage discounts vanish and the new vendor starts from zero.
The template closes this with one paragraph the industry hates and professional buyers should never sign without: all TM, term bases, and glossaries created under the contract are your property; TMX/TBX exports delivered quarterly and at exit within ten business days, free; no retention fees, no deletion, no license-back. A vendor who redlines this clause is telling you the lock-in is part of their pricing model. That, too, is useful information — the scorecard has a row for it.
The 2026 Questions: AI, MTPE, and Where Your Data Goes
Two structural changes hit translation buying in the last three years, and almost no RFP template has caught up. First, machine translation plus human post-editing (MTPE) went mainstream: done honestly, with full post-editing to the ISO 18587 standard, it is a legitimate, cheaper tier for the right content. Done quietly, it means you are paying human-translation rates for lightly reviewed machine output. Your RFP must require per-workflow disclosure — where MT or LLMs are used, which engines, post-edited to what standard — and must make silent substitution a termination event. If a proposal says “proprietary AI-powered platform” and nothing else, the question bank’s red-flag column tells your evaluators exactly how to score it.
Second, data. Free and consumer AI tools translate by ingesting; enterprise content pasted into them can be retained, processed on foreign infrastructure, or used for model training. The template requires a written ban on public MT/LLM endpoints for your content, a no-training clause, data residency answers, and — the piece buyers always miss — confidentiality that flows down to the freelance chain. The translation industry runs on freelancers, and the same linguists work for competing agencies. An NDA that binds the agency but not the individual who actually opens your file is theater. The GSA’s own ordering guidance flags this freelancer overlap; your contract should treat it as the norm, not the exception.
That overlap has a second consequence most buyers never hear: since vendor linguist pools intersect, what you are really buying in an RFP is not “access to 10,000 linguists” — it is the vendor’s testing regime, its named team for your account, its QA process, and its project management. Which is why the template makes shortlisted vendors run a paid pilot on your real content, blind-scored against a rubric shared in advance, performed by the named production team. The free “sample translation” vendors offer instead is their A-team’s audition; the paid pilot is a rehearsal of your actual Tuesday afternoons. The difference is the whole game: pilot-to-production bait-and-switch is the sector’s most reliable disappointment, and a continuity clause is its cure.
Government Buyers: The 2026 Legal Landscape Changed
If you buy for a federal, state, or local agency, your RFP sits on top of legal requirements — and 2025 rearranged them. Title VI of the Civil Rights Act remains the foundation: recipients of federal financial assistance must provide meaningful access to limited-English-proficient (LEP) individuals, and roughly 29.6 million US residents are LEP by the Census Bureau’s count. For health programs, ACA Section 1557 still requires qualified interpreters and language-assistance notices. What changed: Executive Order 14224 (March 2025) designated English the national language and revoked EO 13166, the 2000 order that had structured federal language-access planning. Critically, EO 14224 does not require reducing services — agency heads decide what their mission requires — but any RFP or vendor guide still citing EO 13166 as governing authority is out of date, and most are.
Federal purchasers buy translation under GSA Multiple Award Schedule SIN 541930, and the GSA’s December 2025 ordering guide is the best free procurement training in the sector. Its evaluation logic matches this toolkit’s: assess linguist pool depth and availability by language pair, prefer language-centric providers over resellers who “picked up translation on the side,” require testing and periodic re-testing of linguists, and check CPARS past performance with the knowledge that freelance overlap makes marketing claims interchangeable. Two federal specifics worth writing into your solicitation: Service Contract Act wage determinations apply to translator labor categories, and set-aside claims should be verified against SAM.gov records rather than proposal cover pages. As a woman-owned small business registered in SAM, we would say that last part — but it protects you from firms whose diversity claims evaporate under audit, and it is the kind of check a contracting officer can run in five minutes.
Scoring Proposals: Weights, Red Flags, and What “Good” Sounds Like
The scorecard in the workbook ships with twelve criteria and editable weights that sum to 100. The default weighting tells you what two decades of watching these contracts succeed and fail has taught us: quality process and named-team depth carry the most weight, followed by rate transparency, TM ownership, and AI/security terms — and only then capacity, tooling, and references. Price is deliberately scored through the normalizer’s total-cost figure, not the headline rate, because you now know better.
What separates this scorecard from the generic procurement version is the red-flag column beside every criterion and every one of the 45 questions. Evaluators who have never bought translation cannot tell a good answer from a fluent one; the answer key gives them ears. “We have 10,000 linguists in 500 languages” scores low against a pool-depth criterion that asks for named linguists in your pairs. “Every translation is checked by our QA team” scores low against a criterion that requires a numeric acceptance threshold and a remedy schedule. “The TM lives in our platform for your convenience” is a two out of five on ownership, and the evaluator doesn’t need to know why — the key already does.
On timeline: a competitive translation RFP runs six to ten weeks end to end — two weeks for questions and answers, three for proposals, one for scoring, two for paid pilots, and one for award and transition. Compressing the pilot is the only shortcut that reliably backfires; it is the one stage producing evidence instead of prose.
Translation RFP: Frequently Asked Questions
1. What should a translation RFP include? Scope by language and content tier, volumes, quality workflow requirements (TEP, error thresholds, remedies), a standardized rate grid with full fee disclosure, AI/MTPE disclosure requirements, security and confidentiality flow-down terms, translation-memory ownership and exit clauses, pilot protocol, evaluation weights, and timeline. The free template above contains all thirteen sections.
2. How many vendors should we invite? Five to seven for a full RFP, three for an RFP-lite. Fewer than three eliminates comparison; more than eight multiplies your evaluation cost without changing the winner.
3. RFP, RFQ, or RFI — which one do we need? An RFI surveys the market when you don’t yet know what you need; an RFQ prices a defined, commodity purchase; an RFP evaluates competing approaches to an ongoing program. For translation programs the RFP (or the three-vendor RFP-lite) is almost always the right instrument, because workflow and terms matter as much as rates.
4. How much do translation services cost in 2026? Professional human translation typically runs $0.09–$0.40 per source word in the US depending on language pair, subject matter, and turnaround, with MTPE tiers below that. Our translation pricing guide covers current ranges, discount structures, and negotiation levers in depth.
5. Should we run a paid pilot instead of asking for free samples? Yes. Free samples are produced by the vendor’s best linguists under no time pressure; a paid pilot on your real content, blind-scored, with the named production team, predicts what you will actually receive. Budget a few hundred dollars per shortlisted vendor; it is the cheapest insurance in the process.
6. Who should own the translation memory? You. It is built from your content with your money. Require TMX/TBX delivery quarterly and at exit, free of charge, and treat any vendor resistance as a pricing signal, not a technicality.
7. Do government agencies still have to provide translation after EO 13166 was revoked? Title VI and, for health programs, Section 1557 obligations remain in force; EO 14224 revoked EO 13166 but explicitly does not require reducing services. Agency heads decide by mission. Verify current DOJ guidance when drafting your solicitation.
8. Can we use this toolkit for interpreting services too? The scorecard, question bank, security terms, and pilot logic transfer directly; pricing is per-hour or per-minute by modality instead of per-word. See our interpretation services overview for modality-specific considerations.
9. How long does a translation RFP process take? Six to ten weeks for a full cycle: two weeks for vendor questions and answers, three for proposal writing, one for scoring, two for paid pilots, one for award and transition. An RFP-lite with three vendors compresses to two or three weeks. The stage you should never compress is the pilot — it is the only phase that produces evidence instead of prose.
10. Should we disclose our budget in the RFP? Disclose volumes, content tiers, and your evaluation weights — not a target price. Publishing volumes lets vendors quote realistic capacity and leverage; publishing a budget number anchors every bid to it. If budget is genuinely capped, state the cap and invite tiered alternatives (for example, MTPE for lower-stakes tiers) so vendors compete on how much quality they can deliver inside it.
11. What is a fuzzy match grid, and why does it matter in an RFP? It is the discount schedule vendors apply to text that fully or partially matches your translation memory — new words at full rate, near-matches and repetitions at reduced rates. It matters because every vendor defines the bands differently, which quietly makes quotes incomparable. Your RFP should impose one standardized grid, on source words, that all bidders must complete as issued.
12. How do we evaluate translation quality across competing proposals? Never from the prose. Require a numeric acceptance threshold (errors per 1,000 words by severity, using MQM or an equivalent typology), written remedies for missing it, and a paid pilot on your real content scored blind against a rubric you share in advance. Proposals describe intentions; thresholds, remedies, and pilots create accountability.
13. Should we require ISO 17100 certification from bidders? Treat ISO 17100 as a strong signal, not a substitute for diligence. If a bidder claims certification, ask for the certificate number, the issuing body, and the scope page — claims have a way of outrunning paperwork. And remember the standard governs process roles and qualifications, not your specific outcome: pair it with the numeric thresholds and pilot described above.
14. Where does MTPE belong in a translation RFP? As an explicitly priced tier, never as an undisclosed substitution. Ask vendors to quote full post-editing (ISO 18587) for the content classes you designate — typically high-volume, lower-stakes material — alongside human TEP for regulated and customer-facing content. Require per-job disclosure of which workflow was used, and make silent substitution of MTPE for human translation a termination event.
15. What security requirements belong in a translation RFP? Five minimums: an NDA that flows down to every freelancer who opens your files, disclosed storage and processing locations, a written ban on free or public MT and LLM tools for your content, a no-model-training clause, and a breach-notification window with a named contact. Add a HIPAA BAA for health data and evidence of security certifications with their scope pages if claimed.
16. Should we award to a single vendor or split between two? Programs above roughly 500,000 words a year, or spanning very different content types, often run a primary-plus-secondary model: it keeps pricing honest, provides surge capacity, and de-risks transitions. Below that, a single accountable vendor usually wins on consistency and TM leverage. Either way, TM ownership and quarterly TMX delivery are what keep a split — or a future switch — practical.
17. For government buyers: PWS, SOO, or SOW? Current federal practice favors performance-based instruments: a Performance Work Statement (PWS) that defines outcomes and lets the contractor propose the how, or a Statement of Objectives (SOO) for maximum flexibility, rather than a prescriptive SOW that shifts execution risk to the government. The GSA SIN 541930 ordering guide aligns with FAR Subpart 37.1’s preference for performance-based acquisition.
18. What SLAs and KPIs should the final contract include? On-time delivery rate with a credit schedule, the numeric quality threshold and remedy ladder from your RFP, linguist-query turnaround, named-team continuity and substitution rules, TM delivery cadence, and reporting (spend by language and department, leverage savings, quality scores). If it isn’t measured in the contract, it will not be managed after award.
19. How do we switch from an incumbent vendor without losing quality? Start by exercising your TM and glossary export rights — TMX/TBX in hand before you announce anything. Run the new vendor’s paid pilot against the incumbent’s recent output, transition one content stream at a time, and keep a sixty-day parallel period with the incumbent under contracted transition assistance. Buyers without TM ownership clauses should negotiate the export before signaling the switch.
20. What are the biggest red flags in translation RFP responses? In rough order of predictive power: refusal to complete your standardized rate grid; “we have 10,000 linguists in 500 languages” with no named team; no numeric quality threshold; the TM “living in our platform”; “proprietary AI” with no workflow disclosure; a free sample offered in place of a paid pilot; fees “to be determined”; and references that never materialize. Two or more of these, and the polish of the proposal is working against you.
Get the Toolkit, or Get a Bid Worth Comparing
Download the RFP template and the scorecard + bid normalizer workbook, adapt them, and run your process — no email required, no strings. And if your shortlist needs a bidder who will complete your rate grid as issued, accept the TM-ownership clause without a redline, and put its named team through your paid pilot: request a quote or write to [email protected]. We built the toolkit because we win on exactly the terms it teaches you to demand.