Walk into any salon in Jayanagar at 7pm and you'll watch the owner miss customer calls. They're cutting someone's hair. The phone is two metres away. It rings four times and stops. That customer just went to the next salon on Google Maps. This is the story of why we built Vani — an AI receptionist for Bangalore businesses that picks up in Kannada — and what we learned shipping it in 90 days on a bootstrapped budget.
The unanswered phone
India's small business economy runs on the phone. Not WhatsApp, not apps, not the website you spent ₹25,000 on. The phone. Customers in Indiranagar want to know if you can do their hair colour in the next hour. Customers in Whitefield want to know if you still have parottas at 9:45 pm. Customers in Koramangala want to know when their Ather can come in for service.
We spent the first month of building Vani sitting in salons, restaurants and one Ather showroom in Bengaluru. The pattern was identical across all of them:
- Calls came in waves — peak hours had 4-7 calls in 20 minutes.
- The owner or shop manager picked up roughly 60% of them.
- Of the 40% missed, only about 10% ever called back.
- The owners knew. They just couldn't do anything about it.
The existing answers — hire a receptionist (₹15,000–25,000/month + space + training), a Hindi-only IVR (“press 1 for…” which nobody in Bengaluru tolerates), or an enterprise voice bot from Knowlarity / Ozonetel (priced for IT/BPO, not for a 4-chair salon) — none of them fit.
Why Kannada-first
Every voice AI demo we tried in early 2026 had the same blind spot. They could speak Hindi. Some could speak English. None of them could handle a customer who opened a call with “ನಮಸ್ಕಾರ, ಸಾಯಂಕಾಲ ಫೋರ್ ಪಿಎಂಗೆ ಬುಕ್ ಆಗುತ್ತಾ?” — and most of Bengaluru's salon and restaurant calls open exactly like that. A Kannada-first opener, with English/Hindi loanwords sprinkled in, switching to full English the moment the AI uses an unfamiliar Kannada term.
So we built Vani to lead with Kannada. The agent opens in Kannada, listens to one full turn from the caller before deciding which language to settle into, and stays in that language for the rest of the call. We support Malayalam too — our launch restaurant client is Adukkala, a Kerala restaurant — and Hindi and English as fallbacks. Whatever the customer is most comfortable speaking, Vani matches.
What “live in 24 hours” really means
The product promise we settled on is uncomfortable: Vani is live for your business in 24 hours, for ₹1,999 setup and ₹15 per minute of AI call. No annual contract. No “discovery workshop”. No three weeks of voice training.
To hold that promise, the entire onboarding had to fit in one form. A salon owner uploads their service list (or types it in), tells us their two busiest time slots, picks a voice, and pays. The next morning their AI receptionist is on a Bengaluru number, answering calls. The first call we ever shipped to a paying customer was eight hours from sign-up. We've been shaving from there.
The pricing is similarly uncomfortable — uncomfortable for the incumbents. Indian voice AI has been priced as enterprise software for years (₹15-50 lakh for a deployment was normal in 2024). We priced Vani so a 4-chair salon in HSR could pay for it out of one Saturday's recovered missed calls. Most of our paying clients spend between ₹1,500 and ₹3,000 a month.
What we learned prototyping with an EV dealer
The customer-side prototype we've been running for the last few weeks is with an EV dealership in Bengaluru. EV dealers have a brutal phone problem: test ride enquiries spike in the evenings and on weekends, when the showroom is either closed or has one person on the floor. Vani picks up those calls 24/7, qualifies the lead — which model, which area, when they want a test ride — and books the slot. The customer gets a confirmation message. The showroom manager gets a heads-up.
The first weeks of prototyping were a tutor for us. We learned that EV buyers ask very specific questions about range, charging time, and per-SKU pricing — questions a generic AI would invent answers to. So every business on Vani gets its own product or service list as the source of truth, and the agent is held to it. We do the same thing across all four verticals — salons, restaurants, supermarkets, and EV dealerships.
What we got wrong
Three things we'd undo if we were starting over:
- We launched with WhatsApp confirmations first, SMS second. It seemed obvious — WhatsApp is the messaging layer in India. Wrong. Getting a business WhatsApp number live in India is a multi-week compliance exercise. SMS was a fraction of the effort and reached every customer the same day. We pivoted hard. SMS is the default now; WhatsApp is the upsell.
- We tried to build for four verticals at once. Salons, restaurants, supermarkets, EV dealers — all in version one. The product warped under the load and we wasted two weeks on premature abstractions. The fix was embarrassingly simple: ship the four verticals as four distinct products that share one voice.
- We underestimated how much Indian telecom compliance mattered. The bureaucracy around sending a single transactional SMS or placing an outbound business call in India is a non-trivial multi-day setup of its own, with multiple registrations to chase. We lost time learning this the hard way. Worth saying out loud so the next Indian voice-AI founder doesn't lose those days too.
What's next
The inbound product is settled. The next bet is outbound: Vani Reach, an AI that proactively calls a dealer's existing customers with service reminders and offers. Fully compliant — we only contact customers who gave prior consent at point of sale. We're prototyping it with the same Bangalore EV dealer. If you run a dealership or salon chain in Bengaluru and want to see how it works, the easiest thing is to email us at support@navamitra.in or call our demo line at +91 80654 80620 — you'll talk to Vani in 30 seconds.
We're a small team in Bengaluru. If any of this resonates — as a customer, a fellow founder, or a journalist who covers Indian SaaS — drop us a line at support@navamitra.in. The phones are ringing. Somebody should pick them up.
This article is also syndicated on Medium, LinkedIn, IndieHackers and Hashnode. The canonical version lives here at aivani.in/blog/why-we-built-vani.