Onz was a food decision app I designed and developed end to end at Kaizer Studios and launched publicly in February 2026. It started from one stubborn, unglamorous problem: the 10+ minutes a group burns deciding where to eat, and the fact that no existing app was built to end that conversation.
Every discovery app answers "what is near me" in an area of more than a hundred options, which is not the question people are stuck on. In a 47-person survey, 91.5% said choosing where to eat was difficult and 57.4% named the real blocker: not knowing what they were craving. This is option overload, not option scarcity, and adding filters makes it worse.
Forty-seven people, one survey, and a pattern that held. The blocker was never too few options; it was having no way to name a craving. 68.1% asked for mood-based curation before they asked for anything else, which told me the entry point was a feeling, not a cuisine.
So the profile came first. Every recommendation has to trace back to a taste profile you built yourself, not a crowd average — so the shortlist re-sorts the moment you rate a single dish.
Friends' recommendations weren't a replacement for the algorithm — they were a manual stand-in for the same signal it would eventually provide, shipped first because the AI engine went live right after and needed real taste data to work from. The tradeoff sat in that first window: without a model yet, the profile only got sharper if you kept reviewing, so retention had to carry weight the product had not yet earned.
Eighty-one percent said weekly. Eight showed up. Market research showed a better-funded competitor closing the window, and the honest read was that the gap was structural, not a marketing problem. I made the case to wind Onz down. The same evidence standard that justified the design justified stopping it, and that judgement became the validate-before-build gate every Kaizer venture now runs through.