At some point in early April we realized that none of us had taken the full quiz in its current form. We had each taken earlier versions, contributed to the question design, and seen the outputs from our external beta testers. But the version we were actively shipping to people had been revised significantly over the prior six months, and none of us had gone through it from scratch as a user would.
So on a Friday afternoon we each sat down separately and completed the full 40-question quiz without coordinating. We wanted genuine responses, not responses shaped by knowing what we hoped the quiz would do. The results were revealing in ways that were useful, and in at least two cases, uncomfortable.
What Redouane's profile said
My profile came back strongly weighted toward structured outerwear and considered layering, which tracks with how I actually dress. I have been a professional stylist for eight years, and my personal taste is probably the most legible of the three of us: I reach for things with clean construction lines, interesting fabric weight, and earthy tones from deep plum through warm sand. The quiz captured the aesthetic direction accurately.
What surprised me was the occasion mapping. I answered the occasion questions honestly, which meant including the number of hours I spend in client meetings, in workrooms with designers, and in boutiques with buyers. The profile weighted the output toward more polished pieces than I typically wear working in the studio. I had been answering the questions accurately but not fully: I was describing where I go without describing what I actually need in those settings, which is something I can move in and sit on a cutting-room floor in without concern. That is a quiz design observation, not a quiz failure, and it led to a revision of the occasion question that added a follow-up about physical range of motion needs.
What Mia's profile said
Mia's results came back with a strong orientation toward structured but unforced pieces, something between tailored and relaxed, which she says is accurate. What surprised her was the color clustering: the profile read a preference for warm neutrals that she says is slightly ahead of where she currently shops but correct about where she is heading. That kind of forward-looking prediction is something the quiz is designed to capture through the "direction" questions, and it was validating to see it working as intended.
Her one observation: the quiz's material preference questions produced a strong signal toward natural fibers that she feels overstates the case. She does prefer natural fibers but is not opposed to technical fabrics when the piece earns it. The question as written did not give her a way to express that nuance. We revised the material preference section to include a modifying question about context, specifically whether the material preference changes depending on whether a piece is structured or relaxed. It adds about ninety seconds to the quiz and produces meaningfully richer data.
What Stefan's profile said
Stefan's results were the ones that generated the most useful product conversation. His profile came back with conflicting signals in a way that made the output less precise than we wanted. He is, by his own description, someone whose personal style is in genuine transition: he knows what he is moving away from, a fairly casual technical wardrobe accumulated over years of not particularly thinking about clothes, but has not yet settled into a clear sense of what he is moving toward. The quiz was accurate about the transition but uncertain about the direction.
The issue was that the profile produced a wide scatter of match attributes rather than a useful concentrated cluster. Technically it was working: he genuinely has mixed signals, and the profile was reflecting that. The problem is that a scattered profile produces scattered suggestions, and scattered suggestions are not useful for someone who needs to start building coherence rather than confirming existing incoherence.
We spent real time on this one. The conclusion we came to was that the quiz needed a better mechanism for users who are in transition: an explicit question that asks whether you are trying to build on a clear existing direction or actively change it. That sounds simple but requires different handling downstream. For users building on a direction, you match heavily against current taste signals. For users in transition, you should weight the "where I want to be" answers more heavily and treat the current-wardrobe data as context to move away from rather than toward. We built that branch into the profile logic, and it is one of the changes we are most confident improved the quality of the product.
The two quiz questions we changed
The occasion question and the material preference question were both revised based on what we found. The occasion revision added the follow-up about physical range of motion. The material preference revision added the context question about structured versus relaxed garments. Both changes make the quiz approximately two minutes longer and both produce noticeably more precise profile output in cases where the original questions had been producing ambiguous signals.
There is a third thing we changed that came indirectly from this exercise: the framing of the existing wardrobe section. Before we did this exercise, that section asked what you currently owned as a neutral inventory question. After seeing our own profiles, we understood that for users who are in the same position as Stefan, the existing wardrobe question needs a framing that distinguishes inventory from intention. We added a clarifying line: "Tell us what you have now, and we will ask separately what you want to keep building from and what you are ready to move past." That framing produced a meaningful improvement in profile accuracy for transition-state users in our subsequent beta cohort.
Why we are writing this down
We are not the typical Atorie user. Redouane has professional training in styling and a very clear personal aesthetic. Mia thinks about fashion as a product problem, which sharpens her self-awareness about what she actually wants. Stefan is unusual in being both technically precise and genuinely uncertain about his personal style. The three of us are not a representative sample.
But the exercise was useful precisely because we are atypical. When we went through the quiz as users, not as builders, we encountered the edges of the questions in ways that beta testers with less context or less willingness to articulate uncertainty might not have flagged. The awkwardness of the material preference question is something that a user who just wants to get through the quiz quickly would probably not name as a problem. They would just answer it less accurately and get a slightly worse profile as a result. We named it because we knew enough to notice it.
The lesson we took from this is that testing your own product matters most when you do it as a user who does not know what the product wants to hear. The version we did in April, where we each sat down without coordinating and answered honestly, was more useful than any number of internal reviews where we were watching the product with professional distance. We will do this again every time we make significant changes to the quiz, and we will publish what we find.