How to Write a Translation RFP: The Complete Guide to Selecting a Language Service Provider

A translation Request for Proposal (RFP) is a formal document used to solicit and compare bids from language service providers (LSPs). A strong translation RFP defines scope, volumes, languages, technology, quality standards, security requirements, and AI governance – then scores vendors against weighted criteria.

This guide consolidates a decade of RFP process experience into a single framework you can use to run a defensible selection process and choose a translation partner with confidence. 

What a translation RFP is – and when you actually need one 

A translation RFP is a formal document issued by an organization to solicit bids from language service providers for specific translation and localization work. It outlines the project’s scope, requirements, and evaluation criteria so you can compare vendors on something more useful than a per-word rate. 

It is worth distinguishing three documents that are used interchangeably and shouldn’t be: 

• RFI (Request for Information) – a high-level qualification round. No bid, no pricing. You are asking for a longer list of vendors to describe who they are, so you can shortlist. 

• RFP (Request for Proposal) – the real event. You describe the work; vendors describe their solution and price it. 

• RFQ (Request for Quote) – pure pricing against a scope you have already locked down. 

Large enterprises and public-sector buyers typically post RFPs publicly on a procurement platform, which gives every qualified vendor a fair shot at the business. Smaller organizations usually go the other way: research 8–10 agencies, invite the 4–6 that look like a genuine fit, and run a lighter process. Both approaches work. Skipping the exercise entirely does not; the RFP is often the first time an organization is forced to articulate what it actually needs translated, in what languages, at what volume, and why. 

Why RFP quality determines response quality 

Localization is a multi-billion-dollar industry with mature tooling, and the gap between a good vendor and a mediocre one shows up in cost, speed, and rework, not just in the translation itself. A vague translation RFP produces vague, price-anchored proposals. A precise RFP produces proposals you can compare line by line. Write a detailed, thoughtful document, and you tend to get detailed, thoughtful responses. 

Step 1: Do the internal work before you write a word 

Localization touches marketing, product, legal, support, and HR. If you write the translation RFP from one department’s point of view, the winning vendor will be a poor fit for everyone else. 

Before drafting, align internally on: 

• Business objectives. Entering a new region? Consolidating five incumbent vendors into one? Introducing AI-assisted workflows to reduce spending? Say it explicitly – a clearly stated objective lets unsuitable vendors filter themselves out before bidding. 

• Departmental goals. Marketing wants culturally adapted brand content. Product wants fast turnaround on UI strings. Support wants on-demand translation for urgent tickets. All are valid; all belong in the document. 

• Current state. Who supports translation today, with what tools, at what spend? 

• Pain points. The most under-used section of any translation RFP. Tell vendors what went wrong last time: missed deadlines, inconsistent terminology, a vendor holding your translation memory hostage. You have to tell the doctor where it hurts before she can help. 

• A single point of contact and a decision structure. Agree on who scores, who breaks ties, and who signs. 

Step 2: Publish a realistic schedule 

Every RFP should open with dates. Rushing the timeline is counterproductive: you are asking agencies to invest real effort in a response, and compressed windows produce recycled boilerplate rather than tailored solutions. 

Three rules that make this schedule work: 

1. Allow a genuine Q&A window. Collect questions, answer them, and share the answers with every bidder. It keeps the process fair and materially improves proposal quality. 

1. Give at least 14 business days for the response. Two weeks is a floor, not a target. 

1. Keep the shortlist to 4–6 agencies. Beyond six, the review burden climbs sharply without producing better comparative data. 

Step 3: Scope the work in numbers, not adjectives 

This is where most RFPs fail. “We need our website and marketing materials translated into several languages” cannot be priced accurately by anyone. Give vendors metrics; estimates are fine, and clearly labeled estimates are far better than silence. 

Content types and volumes. Break the scope out by content category: 

• Documents: e.g., 6 four-page brochures at 500 words each; 20 menus at 1,500 words each; 18 eBooks at 5,000 words each. 

• Websites: total word or page counts, the CMS in use (Amplience,, Sitecore, Drupal, Optimizely, Umbraco, Contentful, WordPress, Kentico, etc..), and whether multilingual SEO, GEO, AEO is in scope. 

• Software: platform, string counts, resource file formats. 

• Audio/video: formats, script word counts, number of voice or on-camera talents required, subtitling versus dubbing.

Languages and locales. Not “Spanish” – Spanish (Spain) versus Spanish (Mexico). Not “Chinese” – Simplified for mainland China versus Traditional for Taiwan or Hong Kong. Resourcing and pricing vary meaningfully by region. Rank the languages by business priority so vendors can propose a phased rollout. 

File formats. DOCX, PPTX, XLSX, editable PDF, IDML/InDesign, JSON, XML, HTML, CSV, SRT/VTT, CMS exports. Formats drive engineering and DTP effort, which drives cost. 

Beyond translation. State whether you need transcreation, cultural adaptation, multilingual SEO/GEO/AEO, desktop publishing, linguistic testing, subtitling, glossary and style guide development, or interpreting. Optional services should be scoped as separate, separately priced line items so you can compare like with like. 

Human and/or AI-powered workflows. State whether you want human translation teams and workflows only, AI-powered with human post-editing workflows, and/or pure AI translations or combination of all. 

Cadence. One-time project or ongoing program? Estimated monthly volume, peak periods, and expected growth all change how a vendor staffs your account. 

Step 4: Specify technology, integrations, and translation memory ownership 

Technology is where a good LSP and their tech stack can potentially save you time and money – be careful here on annual license and support fees and lock-ins, lack of customizations and bells and whistles that sound more like a cash register than peaceful a Meinl Sonic chime. 

Translation memory (TM). TM stores previously translated segments for reuse, which reduces cost and enforces consistency across projects and vendors. Ask three questions and accept nothing vague: 

• Will the TM developed and maintained on our behalf be our property? 

• Will it be provided on request, in a standard exchange format (TMX)? 

• How are TM leverage discounts applied to quotes? 

Most reputable vendors accept that a client who paid for a TM owns it. Some treat the memory as leverage to keep you from leaving. Settle this in the RFP, not in the exit interview. 

CMS connectors. Translation connectors automate content export and import between your CMS and the vendor’s translation environment, eliminating manual copy-paste, reducing errors, and compressing turnaround. Require vendors to list every available integration and to state clearly whether each is native/out-of-the-box or custom-built; the difference is weeks/months of implementation time. 

TMS and CAT tools. Ask which Translation Management System and CAT tools are used (Trados, memoQ, Phrase, or proprietary), whether an API is available, and – critically – what the tools cost you. Vendors love to demo technology; the license fees, deployment charges, support costs, and contract length often stay off the slide. Ask whether use of the platform requires long-term commitment and how easily that commitment can be terminated. 

Terminology management. Who builds the glossary, who approves it, how is it enforced across linguists, and does it travel with you if you change vendors? 

Step 5: Define quality expectations and service levels 

Quality is not a promise; it is a documented process. Require vendors to describe: 

• Linguist recruitment, testing, onboarding, and re-evaluation. How are translators qualified, and how are they matched to your subject domain? 

• Review methodology. Translation–Edit–Proof (TEP) or equivalent, and who performs each step. 

• Terminology and style guide enforcement, including how client reviewer feedback is captured and applied. 

• Automated and human QA, with error typology and measurable LQA scoring. 

• Error correction and rework policy, including any warranty period. 

• Certifications: ISO 17100 (translation services), ISO 18587 (machine translation post-editing), ISO 27001 (information security) are the most important for delivering translation services. 

Service levels. State your expectations and let vendors respond: standard turnaround for projects under 5,000 words (5–7 business days is a common baseline), milestone-based schedules for larger programs, and expedited options that are clearly defined and transparently priced. 

Reporting. Require periodic performance reporting: on-time delivery rate, LQA scores, error categories and rework rates, volume by language and domain, and stakeholder feedback summaries. If a vendor cannot report on quality, they cannot manage it. A range of Business Intelligence available via a 24/7 translation portal and the ability to easily customize reports is key.  

Step 6: Security, privacy, and data governance 

Translation vendors handle pre-release product information, financial disclosures, legal filings, HR records, and patient-facing content. Treat them like any other data processor. 

Require vendors to describe their information security framework, encryption in transit and at rest, access controls and authentication, subcontractor confidentiality controls, and data retention and deletion policies. Then get specific: 

• Applicable standards: ISO 27001, SOC 2 Type II, GDPR, HIPAA where relevant. 

• Data residency: every country where your content is processed, stored, or accessed, and whether processing can be restricted to approved jurisdictions. 

• Retention and deletion: default retention periods, accelerated deletion on request, certified deletion of content, TMs, and backups, and support for GDPR right-to-erasure workflows. 

• Incident response: a documented plan, notification within 24–72 hours of any breach, and full cooperation in remediation. 

• Disclosure: any security incidents in the last 36 months. 

Step 7: AI and machine translation governance – the section most RFPs are missing 

Neural machine translation with human post-editing is now a more standard practice for high-volume content, and it genuinely works when it is governed. The risk is not that a vendor uses AI. The risk is that you do not know where, how, or with whose model. 

Modern translation RFPs should require: 

Mandatory disclosure. Whether AI or MT is used at any stage, which engines and tools, and for what use cases (drafting, post-editing, QA, workflow automation). State plainly that non-disclosure constitutes material breach. 

Human in the loop. Unless you have authorized otherwise in writing, all deliverables undergo qualified human review, raw MT output is not acceptable as a deliverable, and AI-assisted output carries the same quality and liability standards as fully human translation. ISO 18587 is the international standard governing this process — it treats machine output as a draft that qualified post-editors must validate, and certification means an accredited body has audited that process. Few LSPs hold it; ask, and ask what independent verification exists if they don’t. 

No training on your data. The vendor must not use client content to train or improve AI/MT models, submit client content to public or third-party LLMs, or retain client data for AI analytics. Where AI is permitted, it should run in closed, non-learning environments under explicit written approval. 

Auditability. The vendor must be able to identify which projects were AI-processed and produce audit trails covering tools used, linguist involvement, and QA steps. 

Step 8: Risk management and vendor viability 

Your translation vendor is an operational dependency. Ask them to disclose: 

• Business continuity and disaster recovery plans, with RTO and RPO targets and redundancy for staff, linguists, and technology. 

• Material service disruptions in the last 36 months. 

• Use of employees versus contractors, NDA coverage for all subcontractors, and the ability to maintain a restricted or named linguist pool for sensitive content. 

• Years in operation, ownership structure, recent or planned mergers and acquisitions, and any material litigation, insolvency, or regulatory action in the past five years. 

Step 9: Structure pricing so you can actually compare it 

Ask every vendor to complete the same editable rate card (XLSX), covering: 

• Per-word translation rates by language pair 

• TM leverage discount structure (fuzzy match bands, repetitions, 100% matches) 

• Editing, proofreading, and independent review rates 

• Machine translation post-editing rates (light and full) 

• Minimum fees and rush multipliers 

• DTP, engineering, and testing hourly rates 

• Project management fees, if any 

• Technology, license, or platform fees 

Then require a sample quote(s) for actual projects so you can ascertain scenario pricing. For example: marketing web content at 10,000 words; a legal/compliance document at 2,500 words requiring DTP of page formatting; UI strings in JSON at 5,000 words. Identical scenarios, identical assumptions, directly comparable totals. 

Sharing an annual budget range per language/locale is optional but usually productive: it lets vendors propose a realistic solution instead of guessing at your ceiling. 

Step 10: Score with weighted criteria – and weight risk properly 

Publish your evaluation criteria in the RFP. A defensible weighting for enterprise buyers looks something like the below. Keep in mind, you know your company and its priorities, past pain points and what you need from a translation partner, so weights can change. 

The three questions to stop asking – and what to ask instead 

After reviewing hundreds of translation RFPs, the same well-intentioned questions keep appearing, and they don’t reveal what buyers think they reveal. 

• “How many translators does your company have?” Most LSPs do not employ translators as full-time staff; they work with vetted freelance linguists and in-country partner teams, which is what allows them to scale across dozens of languages and subject domains. A vendor claiming a “network of 15,000 linguists” is quoting a number nobody can verify. Many of those linguists work for several agencies at once – it is entirely possible that two bidders in your process share translators. 
• “Will you provide a free sample translation?” A sample is not a real project. There is no approved glossary, no reference material, no product training, no query channel to your reviewers, no kickoff call. A sample should be correct, but it cannot distinguish between two vendors, because it strips out the very steps that produce excellent translation. 
• “Do you have experience with clients in our industry?” Every bidder will say yes. Since linguists aren’t in-house, the real question isn’t whether the agency has domain experience; it’s whether the linguists assigned to your account do. 

Ask these instead: 

1. Can you provide a résumé/CV for the lead translator for each language? This is the question the industry-experience question was trying to ask. 
2. What integrations do you offer, can we see a live demo, and what will they cost us? Include license terms and exit terms. 
3. Will the translation memory you develop and manage be our property, and will you provide it on request? 
4. Can we have a call with the project management lead and key account staff we’ll actually work with? Ask about communication cadence, query handling, escalation paths, and QA. You would never hire an employee from a written questionnaire alone; don’t hire an account team that way either. 
5. Where and how is AI used in this workflow, and can you prove it? 

Do’s and don’ts at a glance 

Do: 

• Provide company background, website, goals, and previous pain points 

• Give volumes, formats, languages with locales, and CMS details 

• Publish a realistic timeline with a real Q&A window 

• Meet the account team, PM, account manager, language lead, DTP, and technology leads before deciding 

• Include finalist presentations and tool demos 

• Shortlist 4–6 agencies 

Don’t: 

• Send an RFP with no context about your organization or objectives 

• Skip the Q&A window 

• Allow less than two weeks for responses 

• Decide on written responses alone 

• Treat a free sample as your primary quality evidence; focus on the documented QA process instead 

• Evaluate on price in isolation 

Tailoring the RFP to your industry 

• Hospitality. Website, booking flows, and service descriptions into five locales via Sitecore, roughly 50,000 words, with local SEO/GEO/AEO per language, linguistic testing, ISO 17100 compliance, and a maintained glossary of property and brand terms. 
• Financial services. Annual reports, prospectuses, and marketing collateral across four locales, requiring precision in financial terminology, alignment with local regulatory requirements, an approved style guide, and named-linguist controls for material non-public information. 
• Digital agencies. Campaign content across six locales, requiring transcreation rather than literal translation, Adobe InDesign handling for designed assets, and consistent brand voice in every market. 

Frequently asked questions 

What is a translation RFP? 

A translation RFP is a formal procurement document used to evaluate and select language service providers based on quality, cost, technology, security, AI governance, and operational risk. 

How long should a translation RFP process take? 

Plan for roughly 6–8 weeks: 5 days to confirm interest, 7 business days for vendor questions, 14 business days for responses, and up to 14 days for finalist presentations before selection. 

How many vendors should I invite? 

Four to six. Fewer limits comparison; more adds review time without improving the quality of your decision. 

Should I include my budget? 

It’s optional, but a stated range typically produces more realistic, better-scoped proposals. 

Who owns the translation memory? 

The client should own any TM developed and maintained on their behalf, and should be able to request it at any time in a standard format. Confirm this in the RFP and again in the contract. 

How should AI be governed in translation procurement? 

AI use must be disclosed, human-reviewed, auditable, and prohibited from training on client data unless explicitly authorized in writing. ISO 18587 governs the post-editing process for machine translation output. 

Which ISO certifications matter? 

ISO 17100 (translation services), ISO 9001 (quality management), ISO 18587 (MT post-editing), and ISO 27001 (information security). 

Ready to write yours? 

A translation RFP is worth the effort it takes to write well. The more precise and transparent your document, the better the responses, and the more confident your decision. 

Globalization Partners International has been on both sides of this process for over two decades: helping global brands structure translation RFPs, and responding to hundreds of them. If you’re preparing to go to market, we’re happy to review your draft RFP, share a template, or answer questions about scope, pricing structures, technology, or AI governance requirements, whether or not we end up bidding. 

Request a consultation or explore our translation services, ISO certifications, and CMS translation connectors.