Translation QA Quote Checklist Abroad

$25 per 1 million translated characters can look almost free until a client sends 400 product pages, asks for two target languages, and expects a human-quality delivery by Friday. For an operator living abroad, the margin is not in "using AI translation"; it is in counting characters, scoping human review, and quoting the job before the API bill and revision time eat the retainer.

This is a supporting guide for readers who already like the broader AI translation service business abroad model but need the unit economics. The goal is simple: price a translation QA offer that can be sold from a lower-cost country without pretending raw machine output is professional localization.

Data note: API pricing, limits, and tax thresholds were checked in July 2026. Provider plans can change, so verify the live pricing page before quoting a large client.

What does API cost really mean?

Translation API cost is a character-volume problem, not a per-file or per-project problem. A one-page sales sheet, a 300-page help center, and a product catalog all become source characters, target languages, glossary setup, review time, and client risk.

As of July 2026, DeepL's own API plan page says the Developer API allows up to 1,000,000 characters in total before upgrading, while Growth includes fixed monthly or annual pricing with included characters and overage billing. The same page says Growth has monthly usage limits of 50 million characters and 300 speech-to-text hours.

The cleanest public dollar benchmark I found is the AWS Marketplace listing sold by DeepL: DeepL API Growth at $312 for 12 months, including 12 million characters per year, with additional characters billed at $25 per 1 million characters. Treat that as a marketplace procurement quote, not proof that every direct DeepL customer sees the same checkout terms.

The character-count trap

The source length matters more than the file count. DeepL's usage documentation says translated-character usage includes text and document translations and is measured by Unicode code points. It also notes that text translations still count when the source and target languages are the same, which matters if your workflow sends accidental duplicates.

For a simple quote, estimate 6 characters per English word as a conservative planning shortcut. A 20,000-word knowledge base becomes about 120,000 characters per target language before repeats, tags, re-runs, and revisions.

API cost table for a translation QA offer

Use the table as a planning model, not a promise. It shows why API cost is usually smaller than human review time, but still large enough to require usage caps in every client agreement.

Monthly source volume	One target language	Two target languages	API cost signal	How to quote it
100,000 characters	100,000 billable chars	200,000 billable chars	About $2.50 to $5 at $25 per 1M chars	Starter audit or one-off QA package
1,000,000 characters	1M billable chars	2M billable chars	About $25 to $50 before included allowances	Monthly retainer with revision cap
5,000,000 characters	5M billable chars	10M billable chars	About $125 to $250 at $25 per 1M chars	Agency-style retainer with approval gates
20,000,000 characters	20M billable chars	40M billable chars	About $500 to $1,000 at $25 per 1M chars	Custom statement of work with client-paid overages

That $12 estimate is not the price to the client. It is the raw engine exposure. Your sellable work is intake cleanup, glossary creation, QA, screenshots, style checks, issue notes, revision handling, and a final go/no-go recommendation.

Why benchmark against Google Cloud too

Clients may ask why you are not using the cheapest engine. The Google Cloud Translation pricing page says standard NMT translation includes the first 500,000 characters per month as a $10 credit, then charges $20 per 1 million characters. It also lists LLM TextTranslation at $10 per 1 million input characters and $10 per 1 million output characters.

That does not mean Google is always the better service engine. It means your proposal should separate engine selection from QA value. If the client's main need is coverage across many languages, Google may fit. If the client values natural output in a narrower set of language pairs, DeepL may be worth testing.

How should you price the service?

Price the service in tiers because character volume alone does not predict effort. A clean SaaS help center can be easier than a short medical brochure, and a product catalog with repeated specifications can be easier than a founder's brand manifesto.

A practical operator abroad can start with three offers: a small diagnostic, a recurring QA retainer, and a custom localization operations package. This lets beginners sell a constrained job while giving agencies and SaaS clients a path to larger monthly work.

Starter offer

Sell a $250 to $500 translation QA diagnostic for one language pair and one content type. Include character count, engine test, top terminology risks, sample before/after notes, and a fixed quote for the next batch.

This works for a solo operator because the client is buying a decision, not a vague promise. To test demand before building a full agency site, post the offer as a free listing on Brixaz and see who contacts you.

Retainer offer

Sell a $750 to $2,500 monthly retainer with a defined character allowance, one or two target languages, one glossary, one QA pass, and one revision window. Put client-paid overages in the agreement once usage exceeds the allowance.

For example: $1,200 per month for up to 500,000 source characters, one target language, internal-support quality, one terminology log, and weekly delivery. If the client sends 1.2 million source characters or asks for customer-facing polish, the scope changes.

Agency offer

For $3,000 and up, sell workflow management: intake rules, file preparation, glossary maintenance, QA sampling, reviewer assignment, screenshot checks, and delivery reporting. This is where an operator abroad can use lower fixed living costs to spend more time on process without racing to the bottom on per-word rates.

Do not guarantee "native-quality translation" unless a qualified human linguist is reviewing the target text. A safer phrase is "machine translation QA and post-editing coordination," with clear levels for internal, public, and regulated content.

A practical workflow that protects margin

The workflow should stop bad jobs before they become unpaid cleanup. Every step either reduces API waste, reduces revision time, or makes the client approve a cost gate.

1. Count before quoting: extract source characters, target languages, file types, and duplicate content before naming a price.
2. Classify the destination: internal knowledge base, support article, public marketing page, legal document, medical content, or regulated notice.
3. Run a paid sample: test 1,000 to 3,000 words and document error patterns before accepting a bulk job.
4. Create the glossary: lock product names, forbidden translations, brand terms, units, date formats, and tone rules.
5. Translate in batches: avoid uncontrolled re-runs and keep versioned outputs so errors can be traced.
6. QA the riskiest slices first: headlines, pricing pages, disclaimers, onboarding screens, and technical terms.
7. Deliver an issue log: show what changed, what remains uncertain, and which items require a native specialist or legal review.
8. Bill overages automatically: define character, language, rush, and second-pass triggers before work starts.

What can go wrong with API-first translation?

The biggest risk is not the API bill. It is a fluent mistranslation that makes the client think the work is finished when it is only readable. That gap is where your service earns its fee, but it is also where liability lives.

Use three content levels. Level one is internal gist: low polish, low risk, fast delivery. Level two is operational content: support docs, onboarding, training, and product descriptions that need terminology consistency. Level three is public, legal, medical, financial, or regulated content that needs a qualified reviewer and a separate signoff.

Request limits and file limits

DeepL's usage documentation lists a 128 KiB total request-size limit and document character limits that vary by file type, with Pro document translations generally capped at 1 million characters per file. Microsoft's Translator limits page is a useful stress test too: it says each translate request is limited to 50,000 characters across target languages, with F0 at 2 million characters per hour and S1 at 40 million characters per hour.

These limits matter when you promise turnaround. A client may see "API" and assume instant delivery, but file cleanup, batching, retries, formatting checks, and review queues still take time.

Data rules and confidentiality

Do not accept confidential or regulated material without checking data handling. DeepL's API pages emphasize security features for paid API plans, but your client agreement still needs a rule for personal data, trade secrets, contracts, medical text, and unreleased product information.

For a deeper checklist on intake rules, read the adjacent guide on AI translation client data rules abroad. That post is the better place to decide what content should never touch a casual workflow.

Tax and cash-flow setup for operators abroad

A translation QA service is portable income, but it is still business income. U.S. citizens and resident aliens abroad generally do not escape U.S. filing just because the work is performed outside the United States.

The IRS Schedule C and Schedule SE FAQ says sole proprietors report income and expenses on Schedule C, and net self-employment earnings of $400 or more generally require Schedule SE. If you are abroad, the IRS totalization agreements page is also relevant because totalization agreements can eliminate dual Social Security and Medicare coverage in covered countries. The Social Security Administration agreement descriptions explain certificates of coverage for employees, employers, and self-employed workers.

Payment fees also need to be in the model. Stripe's standard U.S. card pricing was listed at 2.9% plus $0.30 per successful domestic card transaction when checked in July 2026. A $1,200 card payment can therefore cost about $35.10 before international-card or currency-conversion add-ons.

A quote template you can actually use

The quote should make the client approve volume, quality level, and exclusions. Short quotes win because they prevent the client from assuming that every translation risk is included.

Copy this scope structure

Project: Machine translation QA for one source language, one target language, and one content type. Included: character count, glossary setup, API translation, QA pass, issue log, and one revision window. Excluded: certified translation, legal advice, medical review, layout reconstruction, SEO rewriting, and native publication signoff unless separately priced.

Allowance: up to 500,000 source characters per month. Overage: additional source characters billed in 100,000-character blocks, and additional target languages billed separately. Turnaround: weekly batch delivery, with rush work quoted before acceptance.

Quality level: internal-use QA, customer-facing post-editing coordination, or specialist review. Client responsibility: provide final approval, brand terms, subject-matter answers, and legal or regulatory review where required.

Next step

Data notes / Sources checked

The pricing and limit checks for this article used provider and primary-source pages rather than generic blog summaries: DeepL API plans, DeepL usage limits, DeepL API product page, AWS Marketplace DeepL listing, Google Cloud Translation pricing, Microsoft Translator limits, Stripe pricing, IRS Schedule C and Schedule SE FAQ, and IRS totalization agreements.

For market context, I also checked public post-editing offers such as tolingo's ISO 18587 page, which listed light post-editing from €0.05 per word and full post-editing from €0.08 per word when checked. That benchmark is useful for positioning, but your own quote should depend on language pair, source quality, reviewer availability, and liability.

Conclusion

Translation API cost is low enough to make a translation QA service attractive, but not so low that you can ignore usage controls. The operator advantage is not arbitrage alone; it is disciplined scoping, paid samples, quality gates, and a pricing model that separates raw characters from human judgment.

If you live in a lower-cost country and earn from U.S. or dollar clients, this can be a practical cash-flow offer. Keep the first version narrow, link it to a clear business system inside the AI Income & Cash Flow category, and expand only after you can predict cost, turnaround, and review effort.