A Marketer's Coding Challenge: Pushing Past Work Limits with AI

The Marketing Partner team concluded that the existing approach of manually creating ad links for each travel product wasn't a good fit for making the most of LuckyGlide. Since the data updates multiple times a day, a structure that relied on pre-generated links didn't work, and as blog traffic also declined due to AI summaries, partners needed a new tool.

The core solution was a structure where setting up a partner automatically generates a dedicated page, and a MyLink is automatically attached the moment a user clicks reserve. Approved admins were able to add partners and set region and exposure scope, and the page showed the lowest price within 180 days as well as the lowest price by city and month. Beyond flight bookings, the system was designed to track performance from overseas accommodations, MyPet, tours, and tickets that occurred after leaving the page, attributing it all to the relevant partner.

The biggest change in this process was that the act of creating links itself disappeared. Previously, partners had to create ad links for each product and attach them to content, but now a single flow from browsing to booking is connected just through partner setup.

During development, a major limitation became clear: even when AI suggests a technically viable method, it doesn't tell you whether you should actually do it. Following a suggestion to use a tunneling tool led to later discovering that this was an approach that shouldn't have been used, and it became clear that without a solid understanding of basic concepts and infrastructure, conversations with AI can end up stuck at just finding implementation methods.

During deployment and operation, an issue arose where things worked on PC but not on mobile, and the cause turned out not to be performance but a VPN setting. Without knowledge of the infrastructure environment, it was difficult even to distinguish the nature of the problem, and running the operation firsthand taught the importance of QA and the weight of handling both development and operations together.

The experience from the first project led directly into a second project. To solve the problem that flight bookings have a slow sense of performance impact, a dashboard was built and deployed within two days, letting partners directly check click and booking flows. Considering various usage environments, a magic link login based on registered email was also introduced, and going forward there are plans to move beyond external API integration toward embedding the functionality directly into MyRealTrip's internal partner pages.