The client
A multi-country travel group running tours across Thailand, Vietnam, and Malaysia. Around forty people on the team, with four full-time sales agents in Bangkok handling inbound enquiries from a mix of web forms, email, and partner referrals. The sales manager had been running the function for almost a decade and knew, quite precisely, where it was breaking.
The problem
The team was getting between three and five hundred enquiries a week. The fastest replies went out inside two hours. The slowest took more than a day. By the time a reply went out, plenty of prospects had already moved on to a competitor or lost interest entirely.
The cause was not effort. The agents were good at their jobs. The cause was the shape of the work. Each enquiry needed reading, classifying (destination, group size, rough budget, dates), routing to the right person, and then a personalized first reply that referenced what the prospect had actually asked about. That took time even for the fastest person on the team, and it had to happen in sequence for hundreds of leads a week.
The existing setup was an inbox, a shared spreadsheet, and a CRM that the agents updated after the fact. Nothing was moving information forward on its own.
The approach
The build connected the existing web forms, email addresses, and the CRM into one pipeline. Incoming enquiries were captured regardless of channel and passed through an LLM-powered classifier that scored them for qualification and tagged them by destination, group size, trip length, and budget range. Based on those tags, each enquiry was routed to the right sales agent, with priority sorting on the higher-value and time-sensitive leads.
For each qualified enquiry, a personalized first-reply draft was generated using the prospect’s own wording as reference. The draft was written into the CRM and queued for the agent to review, edit if they wanted, and send. Nothing was ever sent automatically. The sales manager had been clear that the human touch was the differentiator, and nobody wanted that removed.
The whole thing ran on n8n for orchestration, HubSpot as the CRM of record, and Postmark for outbound. The client’s existing tools stayed. The build added a layer on top.
The result
Median reply time dropped from around fourteen hours to under ten minutes. Agents now spend their time editing and sending drafts rather than composing them from scratch, and they work through qualified leads in priority order rather than in whatever order the inbox served them. Conversion lifted noticeably in the first full quarter after go-live, and the manager reported that the team felt less buried in the work.
The build paid back its own cost inside the first quarter on reply-speed conversion alone. It has run without significant intervention for most of a year.
A quiet side benefit
After a few months, the manager realized the system was producing a clean dataset of every enquiry the business had received, classified, routed, and closed (or lost). That dataset now informs where the business markets next, which destinations are trending in which source markets, and where the sales team should be spending time. None of that was designed in. It dropped out of having the information in one place.
"We were drowning in enquiries before. Now the team opens their inbox to a sorted queue with the right matches already pulled. Median reply time is under ten minutes, conversion is up across the board, and nobody on the team mentions feeling buried anymore."