Our project proposed an agentic AI booking solution for Layla.ai, an AI trip planner that builds personalized itineraries collaboratively with travelers. The identified pain point was the planning to book handoff, where users would either have to book the trip components themselves via third-party websites, or consult a human travel agent to do it for them. Our idea was to implement an AI agent to complete the traveler-approved booking on their behalf. It seemed like the obvious next step: if an AI can plan the whole trip, why shouldn’t it book it too? The reality turned out to be much less straightforward than I initially thought.
My main contributions were in the strategic and economic parts of the project. I built the cost model (fixed and variable cost per booking), conducted the newly vulnerable market analysis and led the defensibility analysis. At first, we found that there were no existing travel booking agents in the market. However, as my research progressed, I was surprised to find how many incumbents had already started investing in the technology. Specifically, I was surprised by the range of companies that were doing so; from large online travel agencies to general-purpose AI assistants. We took this as a meaningful challenge: How can a challenger enter, disrupt and defend its position in a market full of digital giants?
In trying to answer the question of defensibility, we turned to the concepts covered in class. What we got was a sobering reality check. Of the list of defensible assets we covered in class, there were but a few that incumbents did not already have or could not easily replicate. Even Layla, backed by Expedia, had little that Booking.com or a general AI assistant could not replicate. I had always known that tech giants were competitively strong, but I didn’t understand why. Through this project I realized it’s because digital companies generally require the same foundations (data, technology, infrastructure, large user base) regardless of the industry. This is the exact reason behind why they can collide into new industries so easily, like Alibaba expanding from e-commerce to finance. The downside to that is that established giants have already built most of the defensible assets that matter, making it significantly harder for smaller or new companies to attack and defend against them.
In addition to the defensibility dilemma, the barriers to adoption today are higher than we expected. We found that while many younger consumers were open to AI agents completing purchases, only a minority would accept fully automatic transactions carried out by an AI agent. A lot of this came down to trust and perceived loss of control. My teammates designed our prototype to address this, implementing a risk-weighted approval system in which the agent’s authority depended on how much the booking deviated from the traveler’s approved constraints.
My own work also influenced our business case. My estimated cost model showed that, in the pessimistic case, cost savings alone barely justified the investment, so the business case had to be based on whether the agent would also increase conversions (finalized bookings). Together with the trust mechanics, our goal shifted from simply launching an agent to lowering the adoption barrier and limiting the costs of failure for Layla.ai. Hence our final recommendation: Run a controlled pilot, retain human support for edge cases and escalation handling, and expand only when the evidence supports it.
Where we fell short was testing. Our most critical assumption was related to trust. However, it was simply not feasible to realistically test customer trust without enabling real transactions. That is why my teammates designed a controlled experiment with pre-committed success and failure criteria instead of executing one. This was the part of the project I worried about most, and it taught me a lesson: read the full project requirements at the start. If we had done that, we might have chosen a different solution with an assumption that we could test!
The role I took on within the team was keeping us grounded. Throughout the project, I would remind the team of the scope of our idea and the project deliverables and requirements. When the group is energized by new ideas, it’s not always fun to be the person asking “does this fit the brief?” However, I believed it was important to maintain a cohesive storyline to ensure our best chance for a strong result.
I also learned a lot from the collaboration aspect in this project. While we selected a company relatively quickly, we spent longer than I would like to admit iterating between proposed solutions. We kept jumping between 3 possible directions. Our meetings went on and on and we went several meetings without any real progress.
What changed this was stating a goal for the meeting at the start and assigning individual homework at the end. When everyone came prepared, our meetings became shorter and more productive. The professor’s feedback also helped with choosing a solution: keep it simple and don’t make the project more complicated than it needs to be. It was a crucial reminder that it’s better to do 1 thing well than 10 things poorly. So, we stopped debating and chose the solution we thought best connected to the course topics and allowed us to focus our proposal.
The most impactful decision we made, however, was to work on the project simultaneously. We would get on a call and fill in the presentation together. Even when we were working on different sections, we stayed on the call and muted ourselves. This led to significantly more alignment, both between each other and between our separate sections. Our questions would be answered within seconds and inconsistencies would be immediately identified. It also made us a closer team.
This project has changed how I see AI and strategy. I realized the implementation of AI or an emerging technology is only as successful as the strategy behind it. Strategy has to go beyond the value of the product and technology it uses. Companies have to rely on networks, alliances, building strong customer relationships and intangibles like trust to attract an audience large enough to benefit from network effects and build assets competitors cannot easily copy.
If I take one lesson into the future, it’s that the most important question may not necessarily be what potential technologies can add to my company or product, but who else can do it too and how can I build a defensible moat once they do.
