The Certified Translation Filing Engine

01Executive summary

Every foreign-language document filed with USCIS must be accompanied by a full English translation the translator has "certified as complete and accurate," plus a certification that the translator is competent to translate — the rule at 8 CFR § 103.2(b)(3). Federal and state courts impose a parallel requirement: foreign-language evidence is generally inadmissible without a certified translation and an affidavit of accuracy. Verified This is a mandatory, document-by-document purchase attached to millions of filings a year.

It is also a textbook AI-native service. Modern LLMs translate fluently and cheaply — but on high-stakes legal text the best models still hallucinate or fabricate content 10–18% of the time, and the errors "look indistinguishable from correct translations." Verified A mistranslation on an immigration document can trigger a Request for Evidence that adds 3–6 months to a case, or sink an asylum claim outright. Verified That is precisely the shape of business we want: the machine does 85%+ of the production, and a qualified human is concentrated at the one place accountability lives — the signed certification of accuracy. The international standard for exactly this workflow already exists: ISO 18587, machine-translation post-editing. Verified

We do not sell a translation app the customer operates. We sell the finished, filing-ready, liability-backed packet — word-for-word, with [Official Seal] and [Signature] notations, the certification statement, and an RFE-fix guarantee — to the people who buy this every week: immigration law firms and litigation teams.

02Thesis

A linguist-supervised certified-translation service, powered by an internal AI translation-and-formatting engine with a maintained legal/immigration terminology memory and a per-language QA layer, can deliver court- and USCIS-acceptable certified translations at software-grade speed and cost, priced per page. The moat is the combination of (a) a mandatory, filing-gating legal requirement that keeps casual entrants from credibly certifying; (b) a qualified linguist as the accountable, signing trust interface whose affidavit carries professional liability; and (c) an operations layer — translation memory, a formatting/notation library, an acceptance-track-record, and an RFE-fix guarantee — that no raw LLM reproduces. Because every foreign document on every case needs this, and firms file continuously, each law-firm account is a recurring, expanding relationship. As frontier models improve, the automated share widens and margin expands; the human signature stays exactly where the liability lives.

03Discovery rationale

This run searched document-heavy, deadline-bound, regulated workflows with an existing outsourced budget and a per-unit price already accepted by the market. Certified translation for immigration and legal filings surfaced because it satisfies all six qualification gates at once: it is already outsourced (a mature $18–$70/page vendor market exists); it is purely document work (no physical labor); it is legally mandated and liability-heavy (8 CFR 103.2, court admissibility rules, the translator's sworn affidavit = moat); it is inherently repeating (every case brings new foreign documents); and it sits squarely in the machine sweet spot — translation is what LLMs do best, yet the 10–18% hallucination rate on high-stakes text forces judgment and accountability to one human chokepoint. It is not duplicative of any prior blueprint in this catalog: the closest prior runs are the employment-immigration petition engine (drafting the legal petition, not translating source documents) and the claims medical-chronology engine (English-language record synthesis). This is the distinct certified cross-language production layer that sits underneath immigration and litigation filings.

04Candidate comparison

Five candidates were generated this run and scored 1–5 (higher is better) on the criteria that matter most for an AI-native, outcome-priced service. The certified translation engine won on the blend of a hard filing-gating requirement, an unusually clean AI-native fit, a market that already pays per unit, and clear novelty versus prior runs.

CDI is strong but skews toward a customer-operated co-pilot inside the EHR and a nurse-heavy labor model. OSHA/EHS documentation is attractive but the buyer is fragmented and the price-per-unit norm is soft. SBIR/STTR proposal writing is judgment-heavy (quality is the whole game) and grant rules restrict contingency pricing. Estate-planning production carries direct unauthorized-practice-of-law exposure on the drafting itself. Certified translation uniquely pairs a filing-gating mandate with the cleanest possible AI-native production profile and a per-page price the market already pays every day.

05CODE validation

C — Consumer/Buyer trend

Three forces are moving now: (1) frontier LLMs have made translation drafting nearly free, collapsing the cost floor and inviting a productized, AI-first entrant; (2) at the same time, high-profile evidence that AI translation is failing asylum seekers and immigrants — reversed sentences, inserted people, mistranslated legal terms — has made careful buyers (law firms) wary of raw machine output and hungry for an accountable, human-signed layer; Verified and (3) USCIS backlogs and rising RFE sensitivity mean a single translation defect can cost months, raising willingness to pay for "accepted the first time." Verified

O — Opportunity

The underserved gap sits between (a) free/near-free machine translation that no attorney will certify or stake a case on, and (b) traditional human LSPs and freelance translators that are slower and structurally low-margin (~38% gross, 10–25% net) because they still translate mostly by hand. Verified Nobody has cleanly productized a fast, fixed-price, AI-produced, linguist-signed, acceptance-guaranteed certified translation built natively on the ISO 18587 post-editing standard and sold as a managed service to law firms. Inferred from the tool-vs-agency gap.

D — Demand

A funded market already exists and buyers actively shop it: RushTranslate ($24.95/page), Translayte ($24.99/page), Rapid Translate ($27.99), Universal Translation ($20), and dozens more, with explicit law-firm volume plans, dedicated account contacts, and monthly invoicing for firms placing 10+ orders/month. Verified The existence of that pricing tier is direct proof that law firms buy this repeatedly and at volume.

E — Economic sizing

Market-research vendors size immigration-document translation at ~$27.8B (2025) and legal-document translation at ~$7.1B (2025), with certified translation ~29.8% of the legal segment and North America ~36% of it. Inferred (third-party market-research estimates vary and should be treated as directional). Even setting those aside and building bottom-up: 18,000+ AILA immigration attorneys, plus tens of thousands of family/humanitarian practitioners; USCIS received ~994,000 naturalization applications in 2024 alone, and family-based cases routinely require multiple certified translations (birth, marriage, police, academic records). Verified A conservative model — 20,000 firms buying 20 documents/month at a $60 blended document price — is a ~$288M/yr served pool for immigration alone, before litigation, ECA credential evaluation, and healthcare Section-1557 language access. Inferred

06Rubric scorecard — the six gates

07Target buyer & ICP

08Jobs-to-be-Done

• Functional: "Turn my client's foreign documents into a certified English translation that USCIS or the court accepts the first time — word-for-word, with every seal and stamp accounted for and a valid certification statement."
• Emotional: "Never be the reason a case gets an RFE or a filing gets bounced. Make the translation risk disappear."
• Social: "Show clients and partners that we handle their documents with professional, accountable rigor — not Google Translate."
• Job free machine translation fails: no certification, no signature, no formatting fidelity, and a real chance of an invisible hallucination that surfaces at the worst moment.
• Job the freelance/agency route fails: slower, inconsistent turnaround, opaque quality, and no productized guarantee — while charging similar per-page rates.

09The painful problem

A foreign-language document filed without a compliant certified translation is a defect waiting to happen. USCIS requires a word-for-word translation that mirrors the original — including government stamps noted as [Official Seal], illegible signatures as [Signature], and all fine print — plus a signed certification of completeness, accuracy, and translator competence. Verified Minor mistakes — a mismatched name, an omitted stamp, a wrong date format — are a documented, significant driver of rejections in family-based petitions, and an RFE can add 3–6 months to a case. Verified In litigation, foreign-language evidence is generally inadmissible without a certified translation and an affidavit of accuracy vouching for the translator's qualifications. Verified Firms "solve" this today with a patchwork of freelancers, slow agencies, or — dangerously — raw machine translation that hallucinates 10–18% of the time in ways that look correct. Verified The problem recurs on every case, the stakes are the client's future, and the current options are risky, slow, or both. That gap is the wedge.

10The outcome we sell

Concretely, the firm receives: a certified English translation that mirrors the source layout, seals, and stamps; the signed certification/affidavit of accuracy that satisfies 8 CFR 103.2 and court admissibility; notarization where required; a clean PDF packet paired with the source; and a stored translation memory so the client's recurring terms (names, place names, institutions) render consistently across every future document. If USCIS or a court challenges the translation, we stand behind it and remediate.

11First one-feature MVP wedge

12Evidence summary

• Certified translation is required for every foreign-language USCIS filing under 8 CFR 103.2(b)(3). Verified
• Courts generally require a certified translation + affidavit of accuracy for foreign-language evidence to be admissible. Verified
• Top AI models hallucinate/fabricate 10–18% of the time on translation; errors look indistinguishable from correct output. Verified
• Documented asylum/immigration harms from raw machine translation (reversed sentences, inserted people, rejected applications). Verified
• An RFE can add 3–6 months; minor translation mistakes are a significant driver of family-petition rejections. Verified
• Prevailing price $18–$70/page (RushTranslate $24.95/page, 250 words, 24h); volume/law-firm plans exist. Verified
• Industry translation gross margin ~38%; LSP net margins ~10–25%; cost per word falling with AI. Verified
• ISO 18587 (MT post-editing) and ISO 17100 define the exact AI-produce/human-certify workflow; 2025 ISO 18587 update in progress. Verified
• 18,000+ AILA members; ~994,000 N-400s filed in 2024. Verified
• Immigration-document translation ~$27.8B and legal-document translation ~$7.1B (2025), per market-research vendors. Inferred
• 55%+ blended gross margin at maturity via MTPE. Unverified (modeled target).

13Claim table

14Source-claim matrix

15Market & demand evidence

Demand is structural, not speculative. The certified translation is required to file, recurs on every case with foreign documents, and is already bought per page from a crowded vendor field. The clinching signal: incumbents publish dedicated law-firm volume pricing, account contacts, and monthly invoicing — you only build that when firms buy repeatedly and at scale. Verified

16Active buyer conversations

• Immigration-attorney and paralegal forums asking "does the free translation count for USCIS?" and "what does a compliant certification statement need to say?"
• Firms comparing translation vendors on turnaround and RFE risk after a rejection cited a translation defect.
• AILA and practice-management communities sharing horror stories of RFEs and delays from bad translations.
• Litigators asking how to get foreign-language evidence admitted and what affidavit language courts require.
• Immigrant-services and legal-aid orgs discussing how to translate large humanitarian caseloads affordably without risking accuracy.

These are captured as content and outbound triggers, not as proof of product-market fit — paid re-orders and firm retention remain the real test.

17Competitive landscape

The market is real, funded, and fragmented — exactly the setup for an AI-native, linguist-signed, guarantee-backed service that beats free MT on defensibility and beats human-first vendors on speed, margin, and firm-level integration.

18Competitor & budget validation

Budget already exists and is redirectable. Firms currently pay $18–$70/page to vendors and freelancers, or absorb the far larger cost of an RFE/denial when they cut corners with raw MT. Incumbents' law-firm volume plans prove the recurring spend. Our service captures that per-page budget at structurally higher margin by producing 85%+ of the work with AI while keeping the accountable human signature and adding an RFE-fix guarantee no freelancer offers. We are not creating demand or a category — we are productizing an existing, mandated, repeating purchase and taking the margin the AI shift makes available.

19Pricing evidence & proposed pricing

Evidence: RushTranslate $24.95/page (250 words, 24h); Translayte from $24.99; Rapid Translate $27.99; Universal $20; broad market range $18–$70/page or $0.10–$0.16/word. Verified This market already accepts per-page and per-document pricing — no need for hourly billing.

Proposed pricing (per-unit / outcome, never hourly)

Pricing is per-page and per-document — clean, defensible, and recurring. No contingency or fee-splitting (avoids any regulated-pricing or UPL fee concern). The firm plan converts episodic orders into a monthly relationship; the guarantee and turnaround, not price alone, win the account.

20Regulatory & compliance considerations

• USCIS mandate: 8 CFR 103.2(b)(3) — any foreign-language document must be accompanied by a full English translation the translator certifies as complete and accurate, plus a competence certification. No notary or guild membership required by the rule itself. Verified
• Court admissibility: foreign-language evidence generally requires a certified translation authenticated by sworn testimony/affidavit describing the translator's qualifications; notarization when the document carries legal testimony. Verified
• Standards of record: ISO 17100 (translation) and ISO 18587 (machine-translation post-editing) — we operate our production explicitly to the MTPE standard, which requires a competent human post-editor. Verified
• Data handling: documents contain sensitive PII (and sometimes PHI); we encrypt, restrict access, sign DPAs/BAAs where relevant, and use enterprise (non-training) AI endpoints only.
• Formatting rules of record: word-for-word mirroring, [Official Seal] / [Signature] / [Illegible] notations, and per-document certification (one certificate per document). Verified

21Licensing boundary

Because the deliverable is a factual, standards-based translation with a translator's own competence certification, there is no unauthorized-practice-of-law exposure so long as we stay strictly out of legal advice and filing strategy — a bright line reinforced by anti-notario rules. Disclaimers, DPAs/BAAs, and audit logs are built into every engagement.

22AI-native advantage

AI changes the economics, speed, and consistency of the service — not merely "we use ChatGPT."

• AI tasks: OCR and normalize the source; produce a faithful first-pass translation; mirror layout, seals, stamps, and fine print with the correct notations; match names/terms against the client's translation memory; assemble the certification statement; run automated source-vs-target completeness and number/date/name checks; flag low-confidence spans for the linguist.
• Deterministic rules: USCIS formatting checklist, one-certificate-per-document rule, notation conventions, evidence-completeness gates, notarization triggers.
• Human tasks: post-edit against source, resolve legal-term ambiguity, verify names/dates/jurisdictions, and sign the certification of accuracy.
• What must never be fully automated: the final accuracy verification and the signed certification — these carry professional accountability and case-critical liability.
• Why it compounds: every job enriches the per-client translation memory and the formatting/notation library, so the next document is faster, cheaper, and more consistent. Better base models widen the automated share without touching the human sign-off.

23Internal AI engine architecture

Model-portability layer: the translation engine is model-agnostic behind an internal API, so we can route by language pair to the best model and swap providers as frontier models improve — the terminology memory, formatting library, QA harness, and human sign-off are the durable assets, not any one model.

24AI-vs-human operations pipeline

Target automation: ~80% at launch, ~88–92% by day 90 as terminology memory and the QA harness mature — with the signature immovably human.

25Dynasty translation layer

• Buyer translation: immigration/litigation firms pay for certified translations; their urgent problem is filing-blocking documents and RFE risk; the outcome they want is "accepted the first time, fast, and I never have to worry about accuracy."
• Service translation: a done-for-you certified translation — AI produces and formats, a linguist post-edits and signs, we deliver the filing-ready packet and guarantee it.
• Workflow translation: upload → AI draft + mirror + cert-assembly → automated QA → linguist post-edit & sign → (notary) → deliver → memory update → renewal on the firm's next case.
• Tooling translation: secure portal + OCR + model-agnostic translation API + translation-memory store + QA harness + PDF assembler + e-sign/notary + billing. Off-the-shelf first; custom firm dashboard/API later.
• Sales translation: "Stop risking RFEs on translation. Certified, USCIS- and court-ready, 24 hours, flat per page, signed by a qualified linguist — or we fix it free."
• Delivery translation: first 3 firms served semi-manually (linguist does more of the post-edit); automate mirroring, cert-assembly, and QA as volume proves patterns.
• Expansion translation: from vital records → all document types → litigation packs → notary/apostille → firm API integration → multilingual support desk → adjacent ECA and Section-1557 verticals.

26Anti-duplication analysis

This is not a generic automation agency, a translation SaaS the customer runs, a directory, or an AI wrapper. Similar things that exist: consumer MT (Google/DeepL), per-page certified-translation vendors (RushTranslate), freelance marketplaces, enterprise MT platforms (Smartcat/Smartling). We differ because: (1) the outcome sold is an accepted, liability-backed certified document with an RFE-fix guarantee — not a translation the buyer has to trust and certify themselves; (2) we are AI-first in production (ISO 18587 MTPE) where incumbents are still human-first, which is the margin and speed edge; (3) we build a per-client translation memory and acceptance record and integrate at the firm level, which no freelancer and few vendors do; (4) our buyer wedge is the immigration/litigation firm specifically, with formatting and certification tuned to USCIS/court rules, not general localization. Versus prior blueprints in this catalog, it is distinct from the immigration petition engine (legal drafting) and the medical-chronology engine (English-language synthesis): this is the certified cross-language production layer.

27Anti-commoditization analysis

The obvious objection: won't better models make certified translation self-serve? Two reasons the service still wins. First, accountability is the product, not the words. USCIS and courts require a human to certify competence and accuracy under professional and legal liability; a firm will not stake a client's case on an unsigned machine output, and the 10–18% hallucination rate — with errors that look correct — makes an accountable reviewer non-optional for exactly the documents where it matters. Verified Second, the defensible assets are operational: the ISO-18587 workflow, per-client terminology memory, formatting/notation fidelity, an acceptance track record, notary/apostille handling, and the RFE-fix guarantee. As models improve, our cost falls and margin rises while the signature and the guarantee remain ours to give. Better AI is a tailwind, not a threat — it commoditizes the draft, which is our raw material, not our product.

28Service delivery workflow

• Intake: firm uploads source docs + language pair + case reference via portal or API; auto-typing and completeness check.
• Production: AI drafts, mirrors formatting, assembles the certification statement; deterministic checklist enforces USCIS/court rules.
• QA: automated source-vs-target consensus and entity checks flag omissions, number/date/name mismatches, and low-confidence spans.
• Chokepoint: qualified linguist post-edits against the source, resolves flags, and signs the certification/affidavit.
• Delivery: certified PDF + source paired, notarization where required, delivered to portal; SLA clock tracked.
• Follow-up & renewal: acceptance logged; translation memory updated; the firm's next case triggers the next order; RFE-fix guarantee honored if ever invoked.

29Operations as product

We treat the operation itself as the product and drive out variance:

• SOPs per document type and language pair; gold-standard exemplars for common vital records.
• Structured intake with required-field and legibility gates; automated completeness checks before production.
• Deterministic formatting/notation library and a one-certificate-per-document rule.
• Source-vs-target QA harness with confidence scoring; low-confidence spans force human review.
• Reviewer assignment by language pair and domain; two-person integrity on litigation/high-stakes jobs.
• Version control and audit trail on every document; acceptance/RFE outcomes logged to a root-cause loop.
• Postmortems on any translation-cited rejection feed new QA checks, memory entries, and SOP updates.

30No-holes quality engine

Every unit passes: (1) an automated completeness gate (no omitted segments, seals, or stamps); (2) an entity check (names, dates, numbers, jurisdictions match the source); (3) a consensus/back-translation flag for hallucination and meaning drift; (4) a formatting/certification checklist; and (5) a mandatory human post-edit and signature per ISO 18587. High-stakes documents (asylum declarations, court exhibits) route to a second reviewer. Nothing ships uncertified, and the human never rubber-stamps — low-confidence items are force-routed to review. The acceptance record and RFE-fix guarantee make quality a measured, accountable output, not a hope.

31What the human expert actually does

Target: a linguist signs many pages/day as automation deepens; human minutes concentrate on verification and the signature, never on typing the whole translation from scratch.

32Minimum viable offer

"Certified USCIS translation of vital records — birth, marriage, divorce, police, and academic records — from Spanish, Portuguese, French, Arabic, and Mandarin. 24-hour turnaround, $24/page flat, linguist-signed, RFE-fix guarantee." One offer, one buyer (immigration firms), one measurable outcome (accepted the first time). Everything else — rush, more languages, litigation packs, notary, API — is expansion once the wedge holds.

33Fulfillment process (first 3 customers, semi-manual)

• Stand up a secure upload portal (off-the-shelf), the model-agnostic translation API, a simple translation-memory store, and a PDF assembler.
• Retain 2–3 contract linguists per top language pair (ATA-certified where available) on a per-page basis.
• For the first firms, the linguist does a heavier post-edit while we tune the draft, mirroring, and cert-assembly prompts.
• Deliver, capture every friction point (formatting misses, term inconsistencies), and codify them into SOPs and the QA harness.
• Introduce the firm plan (dashboard + monthly invoicing) once a firm places its second and third orders.

34Tools & systems

• Production: enterprise (non-training) LLM endpoints routed by language pair; OCR; translation-memory store; QA/back-translation harness.
• Delivery: secure client portal, PDF assembler, e-signature, remote online notarization partner, apostille coordination.
• Ops: SOP library, exception queue, reviewer scheduling, audit-log/versioning, acceptance/RFE tracker.
• Business: per-page/per-document billing, monthly invoicing for firm plans, CRM, and later a case-management API (Clio, LawLogix, etc.).

35Human-in-the-loop quality control

The linguist is the last line and the accountable signer. QA automation pre-flags risk (omissions, entity mismatches, hallucination, low confidence), but a qualified human makes the final accuracy call and signs. Litigation and asylum documents get two-person review. The certification the human signs is both the legal deliverable and the quality guarantee — which is why the human sign-off is structurally non-negotiable no matter how good the models get.

36Nonlinear scaling & unit economics

Illustrative per-page COGS (standard tier, ~$24 price)

• Revenue/FTE: as automation deepens, one linguist signs far more pages/day, decoupling revenue from headcount — the sublinear-labor property.
• Funnel assumptions: lead-magnet (free RFE-risk audit) → consult ~15–25%; pilot order → repeat within 30 days ≥60%; firm-plan attach on 3rd order.
• CAC payback: targeted <3 months given low order price but high repeat frequency per firm.
• Retention: firm accounts recur on every new case; translation memory and integration create switching cost.

Margins are modeled ranges, not guarantees; linguist cost per page is the swing variable and the primary automation target.

37Distribution proof table

38Sales & outreach + founder-led content

Founder/expert-led content: teach immigration paralegals and attorneys exactly how USCIS translation rules work and how translation defects cause RFEs — building trust and inbound. Warm GTM: convert free-audit and lead-magnet users with a first-order offer and a firm-plan demo. Targeted outbound: lead with a specific diagnosis ("your I-130 packet's marriage-certificate translation is missing the seal notation USCIS expects") rather than a generic pitch. Layer AEO so the service surfaces when firms ask AI assistants "how do I get a USCIS-compliant certified translation."

39First 30 days of content

10 educational posts

• "What 8 CFR 103.2 actually requires in a USCIS certified translation."
• "The 7 translation mistakes that trigger an RFE — and how to avoid them."
• "Why you can't use Google Translate for a USCIS filing (with real failure examples)."
• "How to notate seals, stamps, and illegible signatures the way USCIS expects."
• "Certified vs. notarized vs. sworn translation: what your case actually needs."
• "Getting foreign-language evidence admitted: the affidavit of accuracy explained."
• "One certificate per document: the rule firms miss most."
• "AI translation hallucinations in legal documents — the 10–18% problem."
• "Turnaround math: how translation delays quietly blow filing deadlines."
• "ISO 18587 in plain English: why a human must post-edit machine translation."

3 diagnostic teardown formats

• "Anonymized RFE teardown: the exact translation defect that caused it."
• "Before/after: a raw-MT birth certificate vs. a compliant certified translation."
• "Red-team a certification statement: spot the 4 problems USCIS would."

2 lead-magnet angles

• "Free RFE-risk audit of your last filing's translations (48-hour turnaround)."
• "USCIS certified-translation compliance checklist + certification-statement template."

1 webinar & 1 outbound diagnosis template

• Webinar: "Killing translation RFEs: a live review of real (anonymized) filings."
• Outbound: "I reviewed a sample of [Firm]'s public filing types — here are 3 translation-defect risks and how we'd remove them."

40Lead magnet & waitlist plan

Lead magnet: a free "RFE-Risk Translation Audit" — the firm uploads one recent foreign-language document (or its existing translation), and our engine + a linguist return a short report flagging formatting, certification, and accuracy risks, with a compliant sample page. It captures the exact pain signal (documents that could trigger an RFE), demonstrates quality, and creates trust before any purchase. Waitlist/CTA: a single "Get certified translations that get accepted — start with a free audit" CTA feeds a portal signup; sales-ready = a firm that uploads a real document or books a consult.

41Pilot design & early-access flywheel

• Cohort: 8–12 immigration firms, capped, recruited from warm network + free-audit users.
• Incentive: founding-firm pricing lock + 5 free pages to start.
• Feedback loop: per-order friction capture + weekly cohort check-in; every formatting miss or term inconsistency becomes a QA check, a memory entry, or an SOP update.
• Product vs. custom: recurring document types and language pairs = product; one-off exotic requests = flagged, not silently absorbed.
• Fix before expanding: zero translation-cited rejections and ≤24h SLA held before opening more pilots.

42Build-before-scale checkpoints

• After 5 firms: harden intake (legibility/completeness gates) and the QA harness on the top document types.
• After 10 firms: harden SOPs, the formatting/notation library, reviewer checklists, and delivery templates per language pair.
• After 20 firms: pause new pilots until COGS/page, rework rate, escalation rate, and cycle time are measured and on target.
• Acceptable temporary manual workarounds: heavier human post-edit on new language pairs. Unacceptable signal: needing a full human re-translation from scratch to hit quality — that means the draft/QA layer isn't ready to scale.

43Launch plan — 7 / 30 / 90 day

First 7 days

• Stand up the portal, translation API, memory store, QA harness, and PDF/cert assembler; retain 2–3 linguists for the top 5 language pairs.
• Publish the free RFE-risk audit + waitlist; write the first 3 teaching posts; define the mock-USCIS acceptance checklist.
• Draft the certification-statement and affidavit templates; line up a notary/RON partner.

First 30 days

• Run the audit across warm network + outbound to 150–250 firms; book 12–18 consults.
• Deliver 20–40 paid documents semi-manually; capture every friction point; publish the 10-post set + first webinar.
• Codify SOPs and QA checks from real jobs; stand up the firm plan for early repeat buyers.

First 90 days

• Reach 8–12 active firms; hit the 5- and 10-firm hardening checkpoints; measure COGS/page, rework, cycle time.
• Sign 1–2 referral partners (paralegal network, ECA, notary); land one legal-aid bulk pilot.
• Launch rush tiers and 2–3 more language pairs; scope the first case-management API integration.

44Metrics & KPIs

• Funnel: audit completions, audit→consult %, consult→paid %, CAC, CAC payback.
• Delivery: cycle time (≤24h standard, <6h rush), automation %, human minutes/page.
• Quality: translation-cited rejection rate (<2%), rework rate (<5%), QA flag rate, two-person-review coverage on high-stakes jobs.
• Economics: COGS/page, gross margin (target 55%+), revenue/FTE, pages/linguist/day.
• Retention: repeat-order rate (≥60% in 30 days), firm-plan attach, monthly active firms, net revenue retention.

45Risks & mitigations — exhaustive register

46What could kill this

• A publicized case where a translation we certified harmed a client — trust is the entire business, so one bad, well-known miss is the gravest threat. Mitigated by QA rigor, two-person review, and the guarantee.
• Our own data breach of sensitive legal documents — fatal to credibility. Mitigated by treating internal security as product-critical.
• Linguist capacity failing to scale with demand, forcing quality cuts or missed SLAs. Mitigated by automation depth and a global reviewer bench.
• A dominant incumbent or case-management platform bundling "good enough" AI certified translation at cost — margin compression. Mitigated by integration lock-in, niche depth, and the guarantee.

47Go / no-go reasoning & final recommendation

The candidate clears the evidence threshold decisively: a clearly identified buyer (immigration and litigation firms); a specific, painful, filing-gating problem; verified evidence the problem exists and is enforced (8 CFR 103.2, court admissibility, documented RFE and asylum harms); existing per-page budget with law-firm volume plans; active demand tied to every foreign-document case; a narrow one-feature MVP wedge (certified vital-record translations, 24h, guaranteed); a service-first delivery workflow needing no large custom platform to start; clean per-page/per-document pricing with no regulated-pricing or UPL blocker; a credible path to 55%+ gross margin via ISO 18587 MTPE; and a believable distribution path via search/AEO, AILA communities, outbound, referral partners, and integrations. The Sam Altman and anti-commoditization tests are answered honestly: better models lower our cost and widen automation while the accountable human signature, terminology memory, formatting fidelity, acceptance record, and RFE-fix guarantee remain defensible.

48Sources

1. [S1] eCFR — 8 CFR 103.2 (Submission and adjudication of benefit requests): ecfr.gov
2. [S2] LII Cornell — 8 CFR § 103.2: law.cornell.edu
3. [S3] American Translators Association — Evidentiary Translation for U.S. Courts: atanet.org
4. [S4] Atlas Language Services — Translation of Evidence for Litigation: atlasls.com
5. [S5] SEAtongue — The Complete Breakdown of AI Translation Errors: seatongue.com
6. [S6] Adverbum — AI Translation Risks in 2025: A Regulated Industry Guide: adverbum.com
7. [S7] Context / Thomson Reuters Foundation — AI's translation mistakes endanger US asylum cases: context.news
8. [S8] Lowy Institute — In a high stakes environment, AI is failing asylum seekers: lowyinstitute.org
9. [S9] Corpus Localization — Certified Translation for Immigration Attorneys: corpuslocalization.com
10. [S10] Artlangs — The RFE Nightmare: USCIS Translation Rules: artlangs.com
11. [S11] RushTranslate — Pricing / Certified Translation: rushtranslate.com
12. [S12] Translayte — Certified Translation Cost: translayte.com
13. [S13] BusinessDojo — Language Services Industry Trends (Oct 2025): dojobusiness.com
14. [S14] Mordor Intelligence — Language Services Market: mordorintelligence.com
15. [S15] ISO — 18587:2017 Post-editing of machine translation output: iso.org
16. [S16] ATC Certification — ISO 18587 Update: atccertification.com
17. [S17] American Immigration Lawyers Association (AILA): aila.org
18. [S18] DHS Office of Homeland Security Statistics — FY24 Naturalizations Flow Report: ohss.dhs.gov
19. [S19] Data Insights Market — Immigration Document Translation Service: datainsightsmarket.com
20. [S20] Business Research Insights — Legal Translation Services Market: businessresearchinsights.com