Customers ask in a language your staff do not speak

An enquiry arrives in Malayalam, or Tamil, or Hindi, and the person at your desk answers confidently in a language the caller is not comfortable in. The call continues. It does not convert, and nothing in your records shows why.

This failure is quiet, which is what makes it expensive. Nobody hangs up complaining about language — they just go somewhere easier.

What actually breaks

It is rarely total incomprehension. It is the softer failures:

The caller switches to their second language and loses precision. They can manage in English or Hindi, so they do — and then cannot quite express the specific thing they wanted, so they ask a smaller question and leave.

Numbers and names get mangled. Addresses, spellings, appointment times. The part of the call most likely to go wrong is the part you most need correct.

They do not ask the follow-up. Someone operating in a second language asks fewer questions, and the unasked question is often the objection that would have closed the sale.

None of this appears in a report. It looks like an enquiry that did not convert.

Why vendor language lists mean very little

Almost every voice product in this market advertises a long list of languages — “22+ Indian languages” is close to standard. The number is nearly meaningless, for reasons worth knowing before you compare anyone:

Understanding and speaking are different problems. A system can transcribe a language well and sound wrong saying it back, or the reverse. A single entry in a list covers both.

Accent and code-mixing are the real test. Actual Indian speech is mixed — Manglish, Hinglish, English nouns in a Tamil sentence. A system tested on clean single-language speech performs differently on how anyone actually talks.

The voice may not be native. It is common for a “supported” language to be served by a general-purpose voice that produces recognisable but odd output.

Latency varies enormously by language. The gap before the assistant starts speaking is the single biggest driver of whether a call feels natural, and it is not uniform across a vendor’s list.

What to ask instead

Do not ask how many languages. Ask, for the specific one or two your customers actually use:

1. Can I hear it, now, before buying? Not a demo video — a live call.
2. How does it handle mixed speech, where half the sentence is English?
3. How long is the pause before it starts speaking, in that language?
4. What happens when it does not understand — does it guess, or does it get the call to a person?

That fourth one matters most and is asked least. A system that guesses confidently in a language you cannot audit is worse than one that says it did not follow and hands over.

Where we stand

Kooday’s proof point is Malayalam — a real deployment answering real customer calls in Malayalam, not a language on a list. That is why the name is Malayalam: Kooday (കൂടേ) means “with you”.

Malayalam is evidence that the local-language conversation genuinely works, not a limit on where we operate. The assistant handles other Indian languages, and the honest position is that you should test yours rather than take a list on trust.

⚠️ We run on Sarvam, Gemini and OpenAI voices — not every engine on the market, and we would rather name them than imply universal coverage. If the language you need is not strong on those, that is a real answer and we will tell you.

Before you buy anything

Call the assistant yourself, in the language your customers use, and try to break it the way a real customer would: mix in English, say a local place name, give a phone number, ask something slightly off-topic.

Five minutes of that tells you more than any comparison table, including ours.