Helping Buyers and Agents Get on the Same Page

OneHome sits in an interesting position. It gives buyers a place to search for homes, but it's also intentionally connected to the real estate agent. Cotality didn't want AI to break that relationship. They wanted to understand where it could make it better.

Our original research question reflected that: "How might we strengthen real estate agent-homebuyer collaboration by incorporating AI search tools into OneHome and giving agents better support to uncover client priorities and guide them to the right home?"

It was a reasonable place to start, but it still assumed AI search was the answer. We needed to understand the problem before deciding what AI should do.

One pattern kept becoming more interesting to us: what buyers say they want is not always what they respond to. Someone can start a search saying a short commute is essential, then save five houses farther away because they have more outdoor space. Another buyer may insist on a particular neighborhood until they see what the same budget buys a few miles away.

That doesn't mean buyers are bad at searching. It means they're learning.

Traditional search treats preferences like inputs: 3 bedrooms plus $600k plus neighborhood X equals results. Actual home buying is messier. Search, compare, react, reconsider, talk, compromise, search again. The agent is often the person making sense of that mess. That changed how we thought about AI.

We started with: how can AI improve search? We moved toward: how can AI help buyers and agents build shared understanding?

That sounds like a small wording change. It changed almost everything about what we were willing to design. The question stopped being whether AI could make a recommendation. The better question was whether the system could help explain something useful enough that the buyer and agent could have a better conversation.

One direction we explored started with conversation rather than more filters. The idea wasn't to make the buyer configure every preference manually before they could search. Instead, the system could build context over time from what the buyer said and did, then use that context when it became useful.

For example, a buyer might begin by saying commute matters more than outdoor space. A few sessions later, their behavior tells a different story. Rather than quietly changing the search results, the system could surface that tension: "You've saved several homes with longer commutes because they offer more outdoor space. Has outdoor space become more important than it was when you started?"

Now the AI has done something useful. It noticed a pattern. But the buyer still gets to decide whether the pattern means anything, and that answer can give the agent better context.

We also explored how the same idea could work when buyers are comparing properties. Most comparison interfaces are good at showing facts, beds, baths, square footage, and that information matters, but it still leaves the buyer to translate the numbers back into the decision they're trying to make.

Our concept explored an AI summary that could explain the tradeoff instead.

Not every version worked. One piece of feedback during critique was that some of our early concepts still felt like more manual filtering, even though our research was pointing toward people not necessarily wanting another layer of controls.

That feedback was right about the tension. We'd designed a conversational entry point, but parts of the experience still drifted back toward asking the buyer to manage the system. That forced a better question: if AI already has useful context, why are we making the buyer keep configuring it?

That pushed the concept toward surfacing relevant information when it mattered, instead of turning preference discovery into a long setup process. This is one of the pieces we want to show in the final case study, because it's much more representative of design work than pretending the first idea was right.

By this point, we stopped thinking of the concept as an AI search feature. It was closer to a translation layer. The buyer brings incomplete preferences, behavior, reactions, and changing priorities. The agent brings experience, market knowledge, questions, and judgment. The system can sit between those two things and make useful patterns easier to see.

That gives AI a role that feels much more defensible to us: it handles some of the remembering and pattern finding. People still own the meaning.

This section will be completed when the final evaluation is finished. It should eventually answer four things, without turning into a usability-test report: what did we put in front of people, what did we expect to happen, what did we learn that changed the design, and what remained unresolved.

The goal isn't to prove that every idea tested well. The useful part is showing which assumptions survived contact with another person and which ones didn't.

Final project recommendations and evaluation findings will be added after the capstone report is complete. The direction is already clear, though: the strongest role we found for AI in OneHome was not replacing the agent or simply giving buyers another way to generate listings. It was helping turn search behavior into something buyers and agents could understand together.

That could mean recognizing changing priorities, explaining why two homes represent different tradeoffs, summarizing what the system has learned without pretending that interpretation is fact, or taking repetitive information work off the agent's plate so they have more context when human judgment matters.

The technology matters. But the relationship is the product constraint that makes the technology interesting.

We came into this project expecting the hard question to be what should AI do? We think the more useful question turned out to be what should AI not do? Once we put boundaries around it, the design got better.

Let the system remember things people shouldn't have to remember. Let it notice patterns that are difficult to see across dozens of listings. Let it organize and explain. But when a person is deciding where they're going to live, what they're willing to give up, and what actually feels right, AI doesn't need to become the tastemaker. It can help make the decision clearer. The people involved should still make it.