My Starting Point
Before beginning my Master’s in Business and Information Management, I studied hospitality management and worked for a year as a Management Trainee at a luxury hotel in Vienna. Developing a strategy for an AI travel agent therefore felt particularly relevant to me. It felt like a meeting point between the world I had come from and the one I am now learning to understand.
In hospitality, I learned to notice the small things that make an experience feel personal. Remembering someone’s name, their favorite drink, or how they like to spend their time can make them feel understood. I initially approached our Layla project with a similar mindset. I thought agentic AI could help solve the headaches people face when planning travel. For example, I personally don’t enjoy following an hour-by-hour itinerary, but I care about choosing hotels and restaurants intentionally. A friend of mine loves having every hour planned. I imagined the opportunity lay in helping every kind of traveler find their own version of a meaningful trip.
As the project unfolded, however, the idea became more complicated. Understanding what someone wants is one thing, but being trusted to act on that understanding is another. Over the course of the project, this difference changed both my team’s strategy and my own expectations of what we were building.
My Contribution
My main responsibilities included bringing together market and financial research to justify our prototype, developing the prototype evaluation and market test, and contributing to our final evaluation and ethical considerations. Although these were separate sections of the deck, working on them rarely felt separate. Each time I tried to strengthen one argument, I returned to a finding we had made elsewhere, and I began to see how closely the different parts of a strategy depend on each other. I also found that consistent dialogue with my lovely teammates was essential to building a cohesive, defensible storyline.
Evaluating Our Proposal
The hardest part of my contribution was evaluating our proposal critically. We could not actually test the prototype because of feasibility constraints, so I had to imagine how it could be tested and what evidence would make the results meaningful. Developing the proposal and prototype had given us something to believe in. Evaluating it meant stepping back and asking where and how that belief might be challenged within the context of a real-world business strategy.
I developed a proposed experiment comparing manual booking with our agentic booking flow, along with KPIs to assess success or failure. I considered booking completion, efficiency, risk classification, and safety separately. This helped me see that success could not be reduced to one encouraging result. We had to consider that an AI agent might complete more bookings while possibly making decisions outside the traveler’s approved conditions. A smoother experience would lose its value if the user had to bear the costs of an unsafe or unapproved decision.
I also had to accept the limits of what I could conclude. Designing a test did not mean we had tested the idea, and finding supporting research did not mean our particular solution would work. Our recommendation therefore had to leave room for uncertainty. This was difficult, but valuable, and I learned that taking a proposal seriously also means defining the conditions under which it should change or stop.
Rethinking Trust
Trust became the most interesting thread running through this process. I brought an understanding of trust from hospitality, but the project made me see it differently. At the hotel, guests often arrived with confidence in the institution we represented. I could then build on that foundation through personal attention and consistent service. With our AI agent, I realized that the foundation itself needed more consideration.
Our market research suggested that feeling understood did not necessarily mean feeling ready to hand over control. As our research and discussions developed, trust became more central to our strategy than preference matching alone. I found this particularly interesting because it challenged the instinct I had brought from hospitality: that better personalization would be the clearest way to create value.
Our Risk-Weighted Approval System helped turn this learning into a design choice. The agent’s freedom to act depended on whether a transaction remained within approved conditions or required further human judgment. I began to see that human involvement could make delegation possible. The key was to give people enough confidence to let the agent take some of the work from their hands.
Learning from My Teammates
My teammates also helped me recognize where my own thinking needed adjustment. One moment that stayed with me was when a teammate questioned my approach to the prototype justification. I had misinterpreted the lecture slides explaining what the section needed to achieve. I was surprised at first, but once they explained their concern, it became an “aha” moment. I appreciated that they noticed something I had missed. That exchange reminded me how easy it is to become absorbed in developing an argument before checking whether it answers the right question. It also showed me the value of letting others see unfinished work, and how my contribution became stronger because someone was willing to question it.
Looking Ahead
By the end of the project, I understood AI strategy as something more interconnected than I had initially imagined. Personalization depends on understanding the customer, adoption depends on confidence, and automation depends on reliable execution. Each promise brings another responsibility, further showing the importance of trust in this delicate system. In the future, I look forward to deepening this understanding and potentially contributing to AI strategies that are rooted in a user’s best interest and trust.
