Faster, not better: what testing HomeMatch AI taught me
Where the idea came from
Our project started with brainstorming. None of us came from a fully technical background, so at first we were unsure how far to take the technical side of the assignment. We each came up with a few ideas and then voted, and one of mine got picked: an AI that searches for houses, which later became HomeMatch AI. The idea had a personal connection for me. A few years ago I struggled to find a place to live myself, and many students and young professionals in the Netherlands face the same problem today. That made it feel relevant both to society and to us.
The idea also grew during the project. It started as an AI that simply finds houses, and turned into something closer to a real estate agent: one that understands soft wishes, like a quiet street or a manageable commute, and explains why a home fits.
My role and contribution
We divided the parts by voting, and I took evaluation and governance. My part depended most on everyone else’s. I could not run a test without a working prototype, and my governance analysis had to match what the others wrote about the market and the economics. Because of this, we set internal deadlines for each part. Apart from a few adjustments, which were to be expected, this worked well.
Governance turned out to be my biggest challenge. It was not something I could write on my own and add at the end. I had to make sure it connected to the other slides, for example that the risks I described fit the trust argument in our market analysis and the costs in our economics. Going back and forth between my part and the rest of the deck was an iterative process, but it gave me a much better understanding of the product as a whole.
Testing day
I went into testing with no idea what to expect. A small problem came up first: our prototype only opened from one teammate’s account, not through a public link. Instead of losing time on this, we decided to test together on the same day. That turned out to be a good decision, because it meant we compared the AI and the manual search on the same housing options at the same moment. A teammate and I ran the tests together and recorded our findings in one shared document. We tested five persona’s on the same listings and compared how long each method took and how well the results fit.
The testing itself turned out to be more fun than expected. It was interesting to see how my own manual search compared with the AI’s results.
What surprised me
The result surprised me: the AI was a lot faster than manual searching, but its matches were not better. The AI produced a shortlist on average under 3 minutes, compared with almost 9 minutes manually, while the fit of the results was about the same. Looking back, this makes sense. In a manual search you keep a good overview of all the options, it just takes more time.
In a way, this means part of our project failed, because our key assumption was that the agent would find better matches. But I learned more from this result than I would have from a clear success. Time saved is still real value, especially in a market where every rental listing gets dozens of responses and reacting quickly matters. At the same time, our own future-market test showed that speed will not stay an advantage once every seeker and competitor uses AI agents. If HomeMatch is only faster, its advantage will disappear. The value has to come from better fit, and that is exactly what the prototype needs to improve next. The failed part of our test therefore showed us where to focus.
Feedback and evidence
We did not get feedback on our proposal, which we took to mean we were on the right track. Still, that was a challenge in itself: without outside feedback, we had to judge on our own whether our direction made sense.
Evidence from the market shaped the work more. Thinking about competitors such as Stekkies, which many Dutch students use to find rentals, forced us to ask why anyone would choose us instead. Alert tools like Stekkies are already fast and cover many sources, so speed alone was never going to be enough. This is how soft wishes became central to our idea, and why we added buying a home alongside renting. Seen from this angle, our test result confirmed the competitive analysis: speed is the part competitors already offer, so fit is where we have to win. At Demo Day we were one of the groups that presented, and I was glad our prototype seemed to interest people.
What I learned from collaboration
Communication is key. Our project went fairly smoothly because we set deadlines and expectations for each part from the start. We also kept sharing our thoughts, telling each other when a part was finished, and giving feedback. Because everyone knew what to expect from each other, feedback was easy to give and to receive, which made the whole process a pleasant experience.
The hardest coordination moment was fitting everything into the slide limit. We had to decide together how many slides each part would get while still getting our point across. That meant agreeing on what mattered most in the whole story, not just in our own parts.
Not everything went smoothly. Because we each wrote our own parts, small inconsistencies crept into the deck, and the prototype only working on one account made us depend on one person for testing. Next time, I would agree on shared figures and access to tools in the first week, instead of discovering these issues when everything comes together.
How my view of AI and business strategy changed
Before this project, I had no idea that building an AI tool was this doable for a team without a technical background. I also underestimated how much value an AI can create just by saving time and money.
The project also changed how I think about AI as a competitive advantage. Building the agent was more achievable than I expected, but that also means others can build it too. If the technology is easy to copy, it cannot be what makes the business last. What makes HomeMatch defensible is what surrounds the AI: understanding what seekers really want, relationships with rental agencies, and the trust of seekers. My governance work showed me that trust is not only about avoiding risks. In a housing market full of scams, it is part of what the product offers.
Finally, I used AI during the project mainly to check my own parts and to see whether I had missed important connections between our slides. It was useful as a second pair of eyes, but deciding what was true for our project and what belonged in the deck was still my own judgment.
