Tourism
A sales agent that runs 80% of conversations end to end
Someone sees a reel at 1 a.m. and writes immediately — a surf lesson is an emotional purchase and it does not survive until morning. The agent answers in seconds, in the language the customer wrote in, and takes the booking.
A surf school and tour operator. Surf lessons and multi-day camps. Enquiries arrive around the clock, from several time zones, in several languages.
enquiries handled in its first two months
paid bookings produced
conversion per live conversation
median first reply, around the clock
Problem 1
Enquiries arrive 24/7. A sales team does not.
In business hours the team answered with a median of 12.5 minutes. Outside them, an enquiry waited until morning — by which time the impulse that produced it had passed. A customer writing in English waited longest of all, until someone was free to translate, and meaning was lost on the way.
Solution 1
The sales agent
It lives on the school's messenger channels and does the whole job of a booking desk: not answering a question, but carrying the conversation through to a paid booking.
Problem 2
Automating a broken process only makes it fail faster
Before any agent went live, the team's own handling of enquiries had gaps: leads sat unclaimed, and no one could say how long. So the first piece of work was not code. I reworked the process itself and lifted 15-minute response coverage from 60% to 79% — with people, not models. The agent was then built on a process that already worked.
Solution 2
An assistant for the operations director
Understanding who replies fast, who is losing a customer and where a dialogue stalls meant opening dozens of CRM conversations by hand every day. That review took the operations director more than three hours daily.
Solution 3
Content analytics tied to actual bookings
Built-in social statistics show reach and likes but never answer the question that matters: which post brought someone into the inbox. The assistant collects the numbers every morning and puts them next to enquiries and sales from the CRM, so patterns become visible — which content waves move bookings and which only move reach. This is time-based correlation, not direct attribution: the platform does not expose the link between a specific post and a specific enquiry, and we do not claim it does.
An enquiry outside business hours waits until morning
A customer writing in English waits longest, and meaning is lost in translation
The operations director spends 3+ hours a day reviewing conversations
Nobody knows whether content produces bookings or only reach
An answer in seconds, at any hour
Whatever language the customer writes in, the reply is immediate and intact
The same review takes 15 minutes
Content and enquiries in one dashboard, weekly summary by format
What changed
paid bookings
From 330 enquiries in the agent's first two months
conversion per conversation
Up from 29% at launch — within 2 points of the human team's 38.6% on the same line
of bookings arrive out of hours
Revenue that previously waited until morning, or left
to review the sales team
Instead of more than three hours of manual work every day
Under the hood
What this is built from
- Python
- LLM dialogue agent with warm hand-off to humans
- Telegram Bot API
- Instagram Graph API
- amoCRM API
- Scheduled reporting jobs
- Multilingual: Russian, English, Chinese
What’s next. Currently extending the agent from $100 lessons to $2,500 multi-day tours — a different sale, with a longer decision and a human closing it.
Other work
Let’s talkabout your bottleneck
Hiring, a project, or a second opinion on an agent design — Telegram is the fastest way to reach me. I read everything myself.
mail [at] artemborisov.ru
Based in Russia (UTC+12) · open to relocation