01
A missed reminder was a symptom, not the system.
The discovery work covered how enquiries arrive, how staff check fares, when an option or issuance deadline matters, where follow-up lives and how the owner verifies the financial picture. WhatsApp, email, Travelport and repeated manual checks all carried a different part of the operation.
Automating the first visible reminder would have left the fragmented state underneath it. I mapped the work first, including the exceptions and the moments where a person still needs to decide.
02
Let each system keep the truth it is qualified to own.
Travelport remains authoritative for booking and ticketing facts. Communication records stay with the communication channel. A shared workboard gives the team a readable operational view without pretending to replace either system.
A small activity record can remember what the automation checked, which reminder it sent and whether it needs to try again. It should not become an unofficial customer or booking database.
03
Start with visibility before chasing the whole operation.
The broader assessment found meaningful accounting and reconciliation friction. The accepted first sequence is narrower: make follow-up, issuance deadlines and reminders visible enough that the team can trust the workboard.
That first version can be tested without asking a long-running agency to replace the systems and habits that already keep the business moving.
04
Automate the repeatable parts and expose the exceptions.
Checks, deadlines, reminders and escalation rules follow fixed rules. AI can help summarize a conversation or interpret an ambiguous message, but it should not issue a ticket, move money or silently resolve a conflict between records.
The design keeps a human approval point around consequential actions and makes uncertain cases visible rather than forcing a confident answer.
05
A working plan is not an outcome claim.
The wider operation was assessed as carrying roughly 18 to 30 hours of weekly coordination opportunity. That range is directional and covers more than the proposed first release.
No time saving, follow-up improvement or financial result is claimed because the system has not been deployed and measured. The proof here is the diagnosis, boundary design and implementation plan.