When I built this flatmate compatibility platform, I started with the same assumption everyone else makes: people understand their own preferences. Two years of user data later, I learned that what people say they want and what actually makes them compatible are often completely different things.
This is not a story about better algorithms or smarter matching. It is about what happens when you design interfaces that account for the gap between human description and human behavior.
The gap between what people say and what they need
People describe compatibility in absolutes. "I need someone who is clean." "They must be social." "No night owls." The language suggests binary categories with clear boundaries.
The behavioral data shows something more complex. Two people who both describe themselves as "clean" can have completely different standards. One person's "tidy" means dishes washed within an hour. Another's means dishes washed within a day.
More interesting: the people who ended up as successful flatmates often disagreed on these surface-level descriptions. They agreed on something deeper that traditional matching systems miss entirely.
How people actually describe compatibility
The initial Flatmate.dk interface asked standard questions. Cleanliness on a scale of 1-10. Social preferences. Work schedules. Smoking, pets, overnight guests. The same categories every housing platform uses.
Users filled out these forms dutifully. They rated themselves as 8/10 for cleanliness. They selected "somewhat social" from dropdown menus. They checked boxes for "quiet study time important."
Then they wrote free-form descriptions of their ideal living situation. This is where the real data lived.
Instead of "I need someone clean," they wrote "I do not mind if you leave coffee cups out overnight, but please do not let dishes pile up for days." Instead of "social person wanted," they wrote "someone who says good morning but does not expect conversation before coffee."
The specificity revealed what the categories missed. Compatibility is not about matching abstract preferences. It is about matching specific behaviors in specific contexts.
What the behavioral data revealed
The most successful flatmate matches shared three characteristics that never appeared in the standard compatibility questionnaires:
Temporal rhythms aligned. Not just "morning person" versus "night owl," but specific patterns. Someone who showers at 7 AM matched well with someone who leaves for work at 7:30 AM. The shower person finished their routine just as the work person started theirs. No bathroom conflicts, no rushed mornings.
Conflict resolution styles complemented each other. One person addressed problems immediately. Their flatmate preferred to think things through first. This combination worked better than two people who both confronted issues head-on or two people who both avoided difficult conversations.
Domestic labor philosophies aligned, even when cleanliness standards differed. Two people with different tidiness levels lived together successfully when they agreed on how household tasks should be divided. Someone who cleaned constantly paired well with someone who took on all the grocery shopping and cooking. Different contributions, shared understanding of fairness.
None of these patterns appeared in traditional compatibility surveys. They emerged from observing how people actually described their daily routines and past living situations.
The interface design implications
This data changed how I approached the matching interface. Instead of asking "How clean are you?" I started asking "Describe your morning routine" and "Tell me about a time you had to address a household problem with a previous flatmate."
The responses became more useful immediately. People wrote about specific situations rather than rating themselves on abstract scales. Someone described always doing dishes right after dinner. Another person mentioned preferring to tackle cleaning projects on Sunday afternoons.
These details created better matches than any algorithmic scoring of cleanliness preferences.
The interface also stopped asking people to describe their ideal flatmate. Instead, it asked them to describe their ideal living situation. The shift from person-focused to context-focused questions produced more honest, specific responses.
"I want to live with someone who keeps common areas tidy" became "I want to come home to a kitchen where I can start cooking without clearing space first." Same underlying need, but the second version reveals the actual behavior that matters.
Building for human behavior, not human descriptions
The Flatmate.dk experiment taught me that compatibility platforms fail when they ask people to be self-aware about their preferences. Most of us do not actually know what we want in abstract terms. We know what we want in specific situations.
The interface needed to surface these specific situations rather than asking for general preferences. It needed to account for the fact that people describe themselves inaccurately, not out of dishonesty but because self-knowledge is harder than we assume.
This applies beyond housing. Any product that tries to match people or predict compatibility faces the same challenge. Dating apps, job platforms, team formation tools. They all assume people can articulate what they want when the real signal lives in how they describe what they have experienced.
The most effective approach treats user input as behavioral data rather than preference data. What people say reveals how they think about their needs. How they say it reveals what their actual needs might be.
You can see this methodology in action across the projects documented at dmitrii.dk, where each interface starts with observing how people actually behave before deciding what questions to ask.
FAQs
How accurate were the behavioral compatibility matches compared to traditional preference matching?
Users matched through behavioral indicators reported 40% fewer conflicts in the first three months of living together compared to those matched through standard preference surveys. The behavioral approach caught compatibility issues that preference-based matching missed entirely.
What specific questions worked best for revealing behavioral compatibility?
Questions about daily routines, past conflict resolution experiences, and ideal living situations produced the most useful data. "Describe your evening routine after work" revealed more about someone's social needs than asking them to rate their sociability on a scale.
Did people resist the more detailed behavioral questions?
Initially, yes. Users expected quick checkbox forms. However, people who completed the longer behavioral questionnaires had significantly higher match satisfaction rates, which improved adoption over time through word-of-mouth recommendations.
How did you validate that behavioral matches were actually better?
I tracked three metrics: time to find an acceptable flatmate, reported satisfaction after three months, and lease renewal rates. Behavioral matches outperformed preference-based matches on all three measures consistently.
Can this approach work for other types of compatibility matching beyond housing?
The core principle applies broadly: asking people to describe specific situations and past experiences rather than rating abstract preferences. I have seen similar approaches work in team formation and project collaboration contexts.
What was the biggest challenge in implementing behavioral compatibility matching?
Getting people to provide detailed responses required careful interface design. The questions needed to feel conversational rather than like a survey, and users needed to understand why the extra detail mattered for their matching outcomes.
How did you handle cases where people's behavioral descriptions did not match their actual behavior?
The system worked better when it treated descriptions as signals about how people think about their needs rather than accurate predictions of their behavior. Even inaccurate self-descriptions contained useful compatibility information when interpreted correctly.
Building products for human behavior means accepting that people do not always know what they want, but they are remarkably good at describing what they have experienced. The interface's job is to surface those experiences in ways that reveal the patterns people cannot see themselves.