The office phone rings 300 times a day. Most of those calls can answer themselves.
Six AI agents that sit on top of the new site and Tanna's existing booking data. They handle the routine in Gujarati, Hindi and English, and hand the exceptions to staff with the context already filled in.
Book in Gujarati, in the chat they already use
The customer types the way they'd talk to the counter. The agent reads the timetable and live seat map, proposes buses, holds seats, sends a UPI link and delivers the ticket into the same thread. It runs 24 hours, so the 11 PM “kal subah ki bus hai?” gets an answer.
- Understands Gujarati, Hindi, English and mixed script
- Knows boarding points by nickname (“Nari Chokdi”)
- Applies ladies-seat rules automatically
- Cancels and rebooks in the same chat
- Sends live-bus link the morning of travel
- Escalates to staff with full transcript
- 0:00CallerHello, chhe wagya ni bus kya pahonchi?
- 0:03Tanna AINamaste. 6 વાગ્યાની ભાવનગર-અમદાવાદ બસ અત્યારે સાણંદ પાસે છે, પાલડી લગભગ 9:20 એ પહોંચશે. તમારો PNR કે મોબાઈલ નંબર આપશો તો WhatsApp પર લાઈવ લિંક મોકલી દઉં.
- 0:14Caller98765 43210
- 0:17Tanna AIમળી ગયું, Mehulbhai. લિંક મોકલી દીધી. બીજું કંઈ?
- 0:22CallerNa, thank you.
“Bus kya pahonchi?” answered by a voice, not a queue
Three enquiry numbers ring into the same two desks. A voice agent picks up first, answers timing, fare, boarding point and live location questions, and only rings a human for anything it can't settle. Same numbers, no change for customers.
Policy applied, refund fired, no callback
- TT4H8Q2L · AMD → BVN 18:30Cancelled 9h before · 100%₹380Refunded
- TT9M2K7P · BVN → BRC 11:30Cancelled 3h before · 50%₹876Refunded ₹438
- TT2X6N4R · BVN → AMD 09:00Cancelled 40 min before₹380Needs human
Edge cases (inside 1 hour, disputes, duplicate payments) land in the staff inbox with the money already held.
Receivers told before they ask
- 2:31 PMLR 48213 → Paldi“Reached Paldi office, collect by 9 PM”
- 1:35 PMLR 48209 → Iscon“On the 1:30 PM bus, arriving ~5 PM”
- 12:04 PMLR 48198 → Alkapuri“Collected by receiver (OTP verified)”
Triggered by the bus GPS crossing each office geofence. OTP on collection replaces the paper signature.
Fill the 6 PM, protect the 9 AM
- 4seats left09:00 BVN → AMDSold out likely. Offer waitlist.
- 39seats left18:00 BVN → AMDPush ₹40 off to 212 past riders on WhatsApp.
- 41seats left23:45 BVN → AMDLow demand Sunday. Suggest merge with 22:45.
5:00 AM, on the owner's WhatsApp
Manifests per bus go to each conductor at departure minus 30 minutes.
From answering the phone to running the fleet
| Daily task at the Waghavadi Road office | Staff minutes today | With Copilot | Who handles it |
|---|---|---|---|
| Route enquiry calls (“where is the bus”) | 190 | 25 | 87% automated |
| Phone bookings, reading out seat numbers | 140 | 20 | 86% automated |
| Cancellations and refund follow-ups | 75 | 10 | 87% automated |
| Calling parcel receivers | 60 | 5 | 92% automated |
| Morning manifest and day-end tally | 45 | 5 | 89% automated |
| Festival greeting / offer broadcasts | 30 | 2 | 93% automated |
| Total per day | 9h 0m | 1h 7m | 88% freed |
Estimates from a two-desk office handling 300 calls a day. Real numbers come from two weeks of call logs before anything is switched on. Nobody is replaced. The same two people stop reading out seat numbers and start chasing uncollected parcels, empty night buses and the corporate accounts in GIDC.
Site first. Then one agent a fortnight.
Week 1 to 2: the new site and booking flow on tannabus.in, connected to the existing Infinity backend. Week 3: WhatsApp booking. Week 5: the enquiry line. Each agent runs shadow mode for a week, staff see every reply, then it goes live.