← Back to Blog

Your AI Product's Next Market May Not Buy in English

Dhaval Bhatt
A small business owner speaking into a phone in a dim office, surrounded by glowing multilingual chat bubbles in violet and electric blue light

Most AI products are sold, taught, and supported in English. A large group of business owners is not buying that way.

Two announcements last week made the gap hard to ignore.

On October 13, Dapta announced a $6M seed round led by Chingona Ventures to build no-code AI agents for businesses whose founders do not buy in English, starting with Spanish speakers in the US. Two days later, OpenAI began rolling out GPT-Live-1.1 for ChatGPT Voice, with 13 new voices and better performance in French, German, Hindi, Mexican Spanish, Korean, Japanese, and other non-English languages.

One is a startup. One is a model update. Together they point at the same opening:

Language is becoming a product wedge, not a translation setting.

The market AI keeps reaching last

Dapta’s CEO, Nicolas Rojas, puts the problem plainly in the funding announcement. Business owners “don’t adopt technology from a settings page. They adopt it from someone they trust, who explains it in the language they think in.”

His examples are not tech companies. A roofer in Houston. A family law firm in Miami. An insurance agent in Phoenix. Their customers may speak English, Spanish, or both. The owner still learns, decides, and buys in their own language.

The numbers behind that are not small:

  • More than 1 in 5 people age 5 and older in the US spoke a language other than English at home from 2017 to 2021, according to the Census Bureau.
  • About 61% of them spoke Spanish.
  • Stanford’s State of Latino Entrepreneurship research found Latino-owned businesses grew 44% from 2018 to 2023, while the number of white-owned businesses slightly declined.

Dapta’s own playbook shows how they reach this market: free education in Spanish, daily office hours, weekly webinars, and in-person workshops. They say about 4,000 new businesses join every week, almost all through their own content.

The lesson is not “go copy Dapta.” The lesson is that a market that big, underserved by most AI platforms, has room for many focused products.

Why this favors domain experts

If you have spent ten or twenty years in an industry and you work fluently in two languages, you hold two kinds of knowledge most AI builders lack.

You know the workflow. Where leads arrive, what a good intake call sounds like, which paperwork slows everything down, and what the owner checks before saying yes.

You know the customer’s language at work. Not textbook vocabulary. The terms a contractor uses with a homeowner. The way a patient describes symptoms to a clinic receptionist. The point in a conversation where people switch from Spanish to English because the insurance terms only exist in English.

That second layer is where generic products break. A tool can translate its menu into ten languages and still fail the moment a caller mixes two of them in one sentence.

OpenAI’s GPT-Live-1.1 notes mention better handling of conversations that mix languages. That helps. It does not tell you which industry phrases matter, which mistakes cost a sale, or when a human needs to take over. You know that.

Translation is a feature. Fit is the product.

If you build for this market, avoid the most common shortcut: build in English, translate later.

Here is what “built for the language” looks like in practice:

  • The first conversation works in the customer’s language. Answering calls, texts, and WhatsApp messages in English or Spanish is the core of Dapta’s first agent, not an add-on.
  • Onboarding is in the owner’s language too. Your setup flow, help docs, and support calls are part of the product. If those are English-only, the owner is still locked out.
  • The product handles mixed-language input. Test with real recordings and messages where people switch languages mid-sentence.
  • Industry terms stay accurate. Build a short glossary of the 50 to 100 terms your workflow depends on and check every output against it.
  • The handoff is clear. Decide when the agent hands a conversation to a bilingual human, and tell the customer when it happens.

None of this needs a huge engineering team. It needs someone who knows the work and the language well enough to define what “correct” sounds like.

Start with one language, one workflow, one channel

The fastest path is narrow. Pick:

  1. One language pair you know well, such as Spanish and English, or Hindi and English.
  2. One industry you understand from the inside, such as home services, dental clinics, immigration law, freight, or insurance.
  3. One repeatable job that loses money when it goes wrong. Missed calls and slow lead follow-up are a good start because the owner feels the pain every week.
  4. One channel where that job already happens: phone, SMS, or WhatsApp.

Then validate before you build:

  • Interview 10 owners in their language. Ask what happens today when a lead calls after hours.
  • Collect 20 real (anonymized) call scripts or message threads. These become your test set.
  • Offer a done-for-you pilot to three businesses. Run the workflow with off-the-shelf tools and your own review before you write a line of product logic.
  • Track one number the owner cares about, such as appointments booked from after-hours leads.

Every step is in the owner’s language. That is the product test and the trust test at the same time.

Distribution is the hidden moat

Dapta’s growth engine is education in the customer’s language, not paid ads. That matters for small founders.

Large AI companies can add a language in a model update. It is much harder for them to earn trust inside a specific community, through a specific trade association, local creators, and owners referring each other.

If you already belong to that community, you start with an asset most competitors cannot buy. A weekly workshop in Spanish for HVAC owners, or a WhatsApp group for Hindi-speaking clinic managers, can do more for your first 20 customers than a polished landing page in English.

Build what you already understand

Most founders look for AI ideas in markets they have only read about. Your strongest idea may sit where your work experience and your language meet.

That combination is hard to copy. It shapes the product, the onboarding, and the go-to-market in one move.

AI Product Accelerator is a 12-week, mentor-led program for turning what you already know into an AI product you build, launch, and own. Explore the program if you want help picking the market, validating the workflow, and shipping your first version.

Sources