Intake in the caller's language, for work where one call is a case
Machine translation is about to be free. Getting a legal or medical intake right in Spanish, Vietnamese or Tagalog is not, and most voice AI is still built English-first.
Neil Gaikwad
DisclosureI build Malan, a bilingual voice intake platform for service businesses, law firms among them. Read this as the memo of someone with a stake in the answer.
The gap
About 27 million people in the US have limited English, and nearly one in five employer businesses is immigrant-owned. Spanish is 61% of non-English speech at home, but the long tail is long: Chinese, Tagalog, Vietnamese, Arabic, Korean.
For most businesses, a missed call in any language is a lost sale. For a few, it is a lost case: an immigration consult, a personal injury claim, a home health intake, a property damage claim. These are the buyers where one call can be worth thousands, the record has to be right, and the caller often does not speak English.
The voice AI wave mostly missed them. The funded receptionists and vertical agents are English-first, with Spanish added as a setting. In my research I could not verify a single funded AI intake company built for immigration or personal injury firms. That is either an opening or a warning, and the memo is about telling which.
Why now
The price fell off a cliff. Real-time speech-to-speech translation now costs $0.034 a minute from OpenAI and about 3 cents a minute of output from Google, against $1.25 a minute for a human Spanish interpreter on the phone. OpenAI's realtime voice went generally available with phone calling in August 2025.
Phone calls can now be answered by agents that hold a conversation, ask follow-up questions and write a structured record, in the time a human interpreter takes to join the line.
Health care still has a hard rule. The 2024 Section 1557 regulation, still in force for language access, bars clinics from leaning on unqualified staff and requires qualified human review of machine-translated critical text. That is both a reason to pay and a design constraint.
Who pays, and how they buy
The first buyers are practices where intake is the product: immigration and personal injury firms, home health agencies, public adjusters. They already pay for bilingual staff, answering services with Spanish lines, and per-minute interpreters.
They buy on outcomes they can count: consults booked, cases signed, intakes completed after hours. Price per minute matters less than whether the record is complete enough to act on.
Money already moving
- Answering services sell a dedicated Spanish line as a paid add-on, at $1 per call on top of plans from $300 a month. smith.ai
- Phone interpreting is sold directly to small businesses at $1.25 a minute for Spanish and $1.75 for most other languages. dayinterpreting.com
- The US has about 73,900 interpreters and translators, at a median wage of $28.93 an hour. bls.gov
- AI receptionists already lead with language: one advertises English and Spanish on every call to 2,000+ small businesses. heyrosie.com
The wedge I would back
A vertical intake agent that runs natively in the caller's language and writes the record in the firm's language: an immigration consult form, a personal injury intake, a home health referral. The value is the structured, reviewable record and the booked next step, not the translation.
Start with one profession and two languages, prove the case-signed rate, and add languages by demand. Hard rules stay in code: no legal or medical advice, and a person reviews anything that changes someone's rights.
What I would not fundA general translation layer or another horizontal AI receptionist. OpenAI, Google and Apple are turning translation into a free feature, and answering services already include Spanish at no extra cost.
The founder I would back
Someone who grew up interpreting for their own family, then worked inside the profession they are selling to: an immigration paralegal, a PI intake coordinator, a home health scheduler. They know which questions get answered wrong in translation and why.
Who is already building
| Company | What it does | Stage and last round |
|---|---|---|
| Palabra | Real-time speech-to-speech translation with voice cloning, multilingual-first | $8.4M pre-seed led by Seven Seven Six Aug 2025 · techcrunch.com |
| Synthflow | No-code voice agent platform with its own telephony | $20M Series A led by Accel Jun 2025 · techcrunch.com |
| Goodcall | AI receptionist for small businesses, English-first | $4M seed Sep 2021 · techcrunch.com |
| Avoca | AI front office for home-service businesses | $125M Series B at $1B, per the company's site Oct 2026 · avoca.ai |
| Vapi | Developer platform for voice agents | $50M Series B led by Peak XV, about $500M valuation May 2026 · techcrunch.com |
| Sanas | Real-time accent conversion for call-center agents | $65M led by Quadrille Capital and Teleperformance Feb 2025 · techcrunch.com |
As of Oct 2026
Risks
- Health care requires qualified human review of machine-translated critical text, so AI-only interpretation carries compliance risk. law.cornell.edu
- Translation is commoditizing: OpenAI sells it at $0.034 a minute and Google ships it in the free Translate app. developers.openai.com
- Federal policy moved away from language access in 2025 and 2026, weakening compliance as a buying trigger outside health care and states like New York. eelp.law.harvard.edu
- Well-funded voice platforms can add any language as a setting, so language alone will not hold as a moat. techcrunch.com
Sources
- KFF, Limited English proficiency in the US (Oct 2025)
- Census Bureau, Language spoken at home, 2018 to 2022 ACS
- USAFacts, Immigrant-owned businesses (May 2025)
- Nimdzi 100, 2025 edition
- OpenAI, gpt-realtime-translate
- Google, Gemini API pricing
- Day Interpreting, pricing
- Invoca, Lead conversion benchmarks 2026
- Cornell LII, 45 CFR 92.201
- Harvard EELP, DOJ rescinds LEP guidance
- BLS, Interpreters and translators
Nothing here is investment advice: these are my own notes on public information, written to sharpen judgment, not to recommend buying or selling anything.