Team 37 – I joined this project without much of a technical background. As an exchange student, my courses are focused on finance, so I was more comfortable with economic analysis than with coding or product design. I took care of the business model and the economic part. The hardest thing was to know whether our solution was financially viable: how much its development would cost, how much its operation would cost and, above all, how much a user would be willing to pay.
To start, I compared three players in the Dutch rental market: Stekkies, HousingAnywhere and Holland2Stay. I wanted to understand the segments they reach and how an AI agent could bring real added value to Stekkies. At first, I mostly analysed their customers, and I did not really understand what made them different. When I looked into it more closely, I understood that Stekkies is mainly used to quickly spot listings that match your criteria, thanks to alerts. HousingAnywhere focuses more on connecting tenants and landlords. Holland2Stay directly rents out homes from its own portfolio.
I shared these findings with my group. At first, we came up with an idea called “HomeScout”, which was supposed to help students with almost every part of their housing search. We wanted to include features such as finding listings automatically, detecting fraud and helping students form groups to rent together. Looking back, we tried to do too much at once. We were also too focused on creating a completely new platform instead of thinking about how we could improve an existing one. Our initial proposal was rejected by the course instructors because the scope was too broad, and I can now understand why. We had focused on solving the whole rental process instead of identifying one specific problem we could realistically address.
After the lecture, we organised an online meeting with the group so that we could rethink our approach and agree on a more focused idea. I suggested brainstorming again, but this time focusing more on our personal experiences and what we felt was missing when we were looking for accommodation. We realised that existing platforms did not really focus on helping students find compatible roommates before looking for a place together. Our agent would address this gap by helping students find people they could live with before starting their housing search. This way, it would complement existing services rather than copy them. I shared my market research with the group to help us decide what was feasible and where our idea could bring real value. Narrowing down our project and setting a more specific goal showed me that everyone can contribute to a project, whether they have a technical background or not. Even though I did not fully understand how the tool would be built, I was still able to question our ideas, challenge certain decisions and support my arguments with market research.
To organise my work, I used the value creation, delivery and capture framework. Stekkies makes it easier to find housing with its alerts. Our agent comes before that: it brings potential roommates together. The problem we targeted was that finding roommates takes time and often depends on personal connections. Our solution would reduce this uncertainty by forming groups based on compatibility criteria. This also gave us a basis to anticipate pricing, even if we did not yet know whether users would be willing to pay. At first, I saw our agent as a disruptive innovation because I thought AI would automatically change the market. However, the framework of Teixeira and Piechota made me reconsider this view. Although platforms offering shared accommodation already exist, they do not focus on matching students with compatible roommates. Our tool could bring value by addressing this gap, but it would mainly improve one step of the user journey rather than transform the market.
To place Stekkies in its ecosystem, I used the Weill and Woerner matrix. I initially chose the ‘ecosystem driver’ position because I was attracted by the idea’s potential, without fully considering what this role would require. A teammate pointed out that it involved much more than connecting people. It meant taking care of contracts and the legal framework of renting. Our group had neither the technical nor the operational means for that. So I changed the position, and I learned that you have to choose a role that the team can really take on.
I also examined the costs of using AI. I thought its marginal cost would be almost zero after development. The work of Shapiro and Varian on information goods helped me separate development costs from usage costs. OpenAI charges for inference based on input and output tokens. Using GPT-6 Sol as a hypothetical example, 10,000 input tokens and 2,000 output tokens would cost around $0.04, excluding other expenses. The actual cost of our agent would depend on its architecture and data usage. However, this example shows that usage costs are not zero and must be taken into account. Technology was only one part of the budget. To set a realistic price, we also had to think about everything that a matching service involves. Stripe Identity, for example, lists €1.25 per successful identity verification in Europe, which gives a rough idea of the cost. Moderation and help in case of conflict between roommates would also cost money, depending on the number of users and the level of service. My analysis therefore had to include both development costs and the variable costs of daily operation.
After estimating the costs, I also had to consider how we would charge users. At first, I thought a monthly subscription would be a good way to generate revenue. However, I started to question this idea when I considered that students might only need our agent for a limited period. In the Netherlands, students spend an average of 4.9 months searching for accommodation, but this does not mean they would need or want to use our service for that entire period. They might also be reluctant to keep paying if they struggled to find compatible roommates. We therefore considered charging when a group is confirmed, as students might be more willing to pay once they have found people they trust and are ready to apply for housing together. This could also help them avoid missing an opportunity because they took too long to make a decision.
I used AI to challenge my ideas and better understand concepts I was not familiar with. Its answers were more useful when I provided specific context, but I still had to check whether its suggestions made sense for our project. This project showed me that AI can help us improve our solution and question our choices. However, even if a solution looks perfect on paper, we still need to consider practical problems. AI can help us develop an idea, but it does not guarantee that the business model will work. Even if our agent helps students find compatible roommates, nothing guarantees they will trust each other enough to apply together or be willing to pay for the service. In future projects, I will look beyond the numbers and consider these factors before assessing a technology’s potential.
