The client
A hospitality group based in Thailand with a guest database that had grown organically for more than a decade. Around twenty thousand past guests sat in the system, alongside a few thousand enquiries that never converted. The marketing team was small, which meant effectively one person with strong instincts and not enough hours in the day.
The problem
The database was, to use the word politely, dormant. A single monthly newsletter went out to the whole list. It was written carefully but generically, because writing twenty different versions was not realistic for one person. The open rate was modest, click-through was weak, and repeat bookings from email were close to invisible in the reporting.
The problem was not that the audience was unengaged. It was that almost nobody was getting a message that felt like it was for them. A guest who had come on a luxury honeymoon trip in 2019 received the same newsletter as a guest who had booked a budget backpacker route six weeks ago. That mismatch was doing most of the damage.
The marketer knew exactly what she wanted to send to each segment. She just could not physically write and send twenty versions a month, every month, while also doing her actual job.
The approach
The starting point was the data: the team already had good segmentation, they just could not act on it at scale. The build took each guest segment (past trip type, destination, recency, average spend band) and set up behavior-triggered sequences inside the email platform they were already paying for. A guest who came back six months ago, who had booked a cultural tour, got a different sequence from a guest who came back last month on a family holiday.
For each segment, an LLM generated the first-draft copy for the sequence: subject lines, body content, and relevant calls to action. The marketer reviewed and edited every piece before anything shipped. That was non-negotiable, and it was the right call. What the AI layer did was cut the drafting time from the thing that was not happening at all to a job she could complete inside half a day per cycle.
The mechanics ran on tools the business already paid for. Their existing email platform handled the sending. The segmentation data already existed in their booking system. The only new pieces were the drafting layer and a small custom glue that pushed segments into the email automations.
The result
Open rates on segmented sequences came in at around three times the previous generic newsletter. Click-through lifted by a comparable amount. Repeat bookings from the email channel moved from essentially invisible to a measurable and steadily growing slice of the monthly revenue.
The marketer got her week back. The campaigns ship on their own. When she wants to adjust the voice or try a new sequence for a new segment, the drafting step takes minutes rather than days.
What is worth noting
Nothing exotic was built. The entire system runs on tools the business already had or could add for a modest monthly fee. The genuine value came from taking segmentation that was already possible in theory and making it possible in practice. That is most of what useful automation looks like inside a smaller business.
"I had a database of twenty thousand past guests doing nothing for me. The build turned that database into a live pipeline. Open rates are roughly three times what they used to be, and the repeat-booking line on the P&L finally moves on its own."