Most voice AI you see in demos has one hidden assumption: your customers speak English. For millions of small businesses — in India and far beyond — that assumption is simply wrong.
We got to test this on a business we know better than any client's: a family soda-distribution route in Gujarat. Daily sales lists, village routes, bills — all of it runs in Gujarati, on phones, between people who have never opened a laptop.
The experiment
We built an AI voice agent that can call a customer and speak natural Gujarati — not translated-sounding, template Gujarati, but conversation a route customer actually responds to. It sits alongside the daily-sales app we built for the same business.
What we learned
The hard part isn't the AI. Modern speech models handle Gujarati surprisingly well once you pick the right stack. The hard part is everything around it: keeping the agent grounded in the business's real data so it never invents a price, handling the way people actually talk over a phone line, and knowing when to hand off to a human.
Why this matters beyond Gujarat
If it works in Gujarati, the same approach works in dozens of languages that AI products usually ignore. For any business whose customers don't live in English — that's a door most of your competitors haven't noticed is open.
If you're curious whether a voice or chat agent could answer your calls in your customers' language, email us — we'll tell you honestly whether it's worth building.
