Urdu doesn't fit the usual "one language, one state" map that a lot of language coverage gets built around. It's the language of Lucknow's old city, where centuries of Nawabi court culture left behind a formal, elaborately polite register — tehzeeb — that's still audible in how people greet each other and ask for things, still alive in the way a shopkeeper in Chowk or Aminabad addresses a customer twice his age. It's also the language of Hyderabad's old city, around Charminar and the pearl and bangle markets nearby, where Deccani Urdu picked up its own flavor, mixing in Telugu and Marathi words over generations, spoken faster and more casually than its Lucknow cousin. Same language, on paper. Different conversations, in practice.
Most support tools that claim Urdu coverage don't actually have it — they have Hindi with the assumption that Hindustani speakers will manage. Spoken Hindi and spoken Urdu overlap heavily at the everyday level, which makes it easy to skip building Urdu specifically and call it close enough. It isn't. Urdu carries its own vocabulary — Persian and Arabic loanwords for ordinary things, not just formal or literary ones — and its own greeting and politeness conventions. A caller who says "shukriya" and gets "dhanyavad" back has just been quietly told the business didn't actually build for them.
Lucknow and Hyderabad are this piece's focus because they're the two cities where Urdu shows up most visibly as a living, everyday commercial language, but the map is wider than those two. Urdu is an official language of Jammu and Kashmir, it has real speaking populations across Bhopal, Patna, and Aligarh, and it carries a literary weight — the ghazal and mushaira tradition, Urdu-language newspapers still in daily circulation — that outlasts any single city's association with it. A model built narrowly for one city's accent misses that broader reality. AIVA's Urdu is trained to hold up across the range of how the language actually gets spoken, not tuned to a single old-city dialect and left there.
Voice sidesteps the script problem — but not the register
Urdu is written in the Nastaliq script, distinct from Hindi's Devanagari, which is usually where "supporting Urdu" gets complicated for text products — rendering the script correctly, handling right-to-left flow, sourcing a keyboard that types it comfortably. Voice skips that entirely — there's no script to render on a phone call. What voice can't skip is getting the vocabulary and the register right, and that's the part AIVA's Urdu model was actually trained for: real spoken Urdu, the Lucknow formality and the Hyderabad casualness both, not a Hindi model with a different label.
It detects Urdu from the first thing a caller says and replies in it, at around 198ms on average, without translating through Hindi or English first. That speed matters on both ends of the register spectrum this piece is about — a Lucknow caller expecting an unhurried, formally polite exchange notices a rushed-sounding reply just as quickly as a Hyderabadi caller expecting a brisker, more direct one notices an overly formal reply. Getting the timing right and getting the register right aren't separate problems; a system built specifically for Urdu treats them as the same problem, because on a real phone call, they are.
Why "close to Hindi" is exactly the trap
Because spoken Hindi and spoken Urdu are mutually intelligible at the everyday level, it's tempting to treat Urdu as a cosmetic variant — same grammar, swap a few words, done. That shortcut is where the honorific register breaks first. Urdu's formal address carries a different texture than Hindi's, shaped by that same tehzeeb tradition — a different way of framing a request, a different set of stock polite phrases — and a model trained on Hindi conversation and lightly re-skinned doesn't reproduce it. It produces something that's grammatically fine and socially slightly off, in a way an Urdu speaker notices immediately even if they can't always name exactly what felt wrong.
The businesses this is for
A tailoring shop or attar business in Lucknow's Chowk area fielding order-status calls all day — chikankari embroidery work in particular runs on a slow, made-to-order cycle where "is it ready yet" is a constant question. A clinic or electronics shop in Hyderabad's old city where half the calls open in Deccani Urdu and switch to English mid-sentence for a brand name or a price. A pearl or bangle trader near Charminar, where phone enquiries about stock and pricing are a steady undercurrent beneath the walk-in trade. A catering business built around Hyderabadi biryani and old-city wedding catering, where festival and wedding season means the phone doesn't stop. A driving school, a salon — the small businesses that live on the phone and lose a customer every time the person on the other end has to switch languages just to be understood.
A bookshop or small press dealing in Urdu literature is its own case worth naming — a smaller, more specialized business, but exactly the kind of place where a caller asking about a specific title or a mushaira event wants to be understood precisely, not approximately. And a real estate office in either city's older neighborhoods, where a lot of property enquiries still happen by phone between people who've known the family for years, is another place where the wrong register on a call reads as a stranger, not as the business the caller thought they were calling.
Getting Hindi close enough was never the bar. The bar was a caller in Lucknow or Hyderabad hearing their own language spoken back to them, correctly, on the first try.
The calls that still need a person
Urdu voice AI isn't trying to handle everything alone. A genuine complaint, a price negotiation that needs real authority, a medical question that needs a doctor's judgment — those get recognized and handed to a person, with the call's context already captured, instead of being forced through automation just to inflate a resolution number. The 96% figure only means something if the remaining 4% is handled well, and that's the part worth checking closely with any vendor, not just the headline automation rate. How AIVA decides when to bring in a human is worth reading if that's the part of any AI voice pitch you're most skeptical of.
What happens on the call
AIVA answers FAQs and books appointments directly, resolving about 96% of voice calls without a human joining in. It works the same way across voice, AIVA's web chat widget, and SMS, with the language auto-detected from the first message — so a Lucknow tailoring shop that wants order-status texts in Urdu, not just phone calls answered in it, gets the same model and the same register on both.
Try it
There's no monthly minimum and no seat license to buy — pricing is usage-based: ₹4 a minute for voice, ₹2 per web chat conversation, ₹1 per SMS. ₹500 in free credit covers plenty of test calls before you spend anything real — see the full pricing breakdown for how that plays out across a small shop's call volume versus a busier clinic's. Urdu is one of twelve Indian languages AIVA built as its own model rather than a Hindi substitute, for the same reasons any regional language deserves that investment instead of a translated shortcut. See how AIVA Voice works and hear Urdu handled as its own language, or start free and test it yourself before deciding anything.