Innovate Software
Books end to end
Availability, fares and seats, then a secure payment link through the existing gateway, or a reservation held for the counter. Nothing is booked until the customer has confirmed a summary of it.
Case study · AI booking assistant
A booking desk that never closes.
How a national coach & courier operator in the Caribbean put an AI assistant on its website chat, connected to its live reservations platform, and what happened next.
The situation
Customers wanted to book at midnight. The phones closed at five.
The operator runs intercity coaches and a courier network from dozens of branches. Its customers book seats, change trips, ask about fares and luggage, and track parcels, and a lot of them do it in the evening and at weekends, when the call centre is closed.
During office hours the same questions reached staff over and over, and every answer had to be looked up in the reservations system by hand. A chatbot that only answered FAQs would not have helped much: what customers wanted was a booking, a change or a tracking result, not a link to a page.
What we built
An assistant that works the reservations system, not around it.
The assistant sits in the chat on the operator's website and reads and writes the same platform the branch staff use. It finishes the job in the conversation, and hands over to a person when it should.
Changes and looks things up
Reschedules, timetables and stops, parcel tracking, courier drop-offs and loyalty points, all read live from the operator's own systems.
Checks who is asking
Booking details stay hidden until the customer proves the booking is theirs, and personal details are masked before a transcript is stored.
Answers from the operator's own site
Policies and FAQs are read live from the operator's website, so their team changes an answer in their own content system, without a developer.
Hands over properly
During office hours, a conversation that needs a person moves to the staff console, and later to the operator's contact-centre platform, with the assistant's summary attached.
Priced per conversation
The cost dashboard shows what every conversation cost, split by language model, speech and messaging, so the business case is arithmetic.
Reviewed, not just run
Conversations where something went wrong are flagged for review, and changes to the assistant are replayed against real past conversations before they go live.
One agent, more channels
Voice and WhatsApp run on the same agent and are being rolled out. Calls will reach it from the operator's existing phone system.
The numbers
What a month on web chat looks like.
3,000+
Customer conversations a month, around 110 a day
64%
Of them outside office hours, when the phones are closed
750+
Conversations a month that end in a booking
98%
Finished without a person taking over
Production web chat, last 30 days to October 2026. "Without a person taking over" means no hand-over to staff was needed or asked for; it is not a satisfaction score.
What we learned
Three things we would tell the next operator.
01
The case is availability, not headcount
Almost two in three conversations arrive when nobody could have answered them. The assistant mostly adds service the business never had, rather than replacing people it does have.
02
Put the rules in the system, not just the prompt
"Confirm before booking" and "verify before revealing" are enforced by the platform, so a clever phrasing cannot talk the assistant past them.
03
Measure from day one
Cost per conversation, a review queue and replays of real conversations turned "is the AI any good?" into questions with answers.
Next step
Could this work for your operation?
Start with a 30-day pilot on one channel, connected to your systems, and judge it on real conversations and real costs.