Why BAD Restaurants Rank First
Uber Eats doesn't always show you the best food at the top of your screen, and it's actually a brilliant engineering move. Today we reverse-engineer "Position Bias" and why making the UX slightly worse protects the whole marketplace.
Full Video Breakdown
Uber Eats is intentionally hiding the food you want to buy. It looks like terrible UX, but ruining your search results is actually a brilliant engineering move. In a normal design process, the rule is simple. Show the user exactly what they want as fast as possible. If you order sushi every Friday, the app should just put that sushi place right at the top. But Uber's engineers noticed a massive problem with this logic. If the algorithm only shows the most popular places and your favorites, those places get all the clicks. When they get all the clicks, the algorithm thinks, "Okay, people love that. I will show it even higher." This creates a closed loop. Big fast food chains win everything and new smaller restaurants get zero visibility. If the app only shows your favorites, it makes easy money for today. But long-term, it's a trap. The algorithm stops learning. It gets no data about new places. And when small restaurants get zero orders, they simply leave the platform. We can call this supply starvation. So how does the interface solve that problem? Uber uses something called position bias. Human beings are lazy. Data shows that we click the first three results just because they are at the top, not because they are the best fit for us. Uber knows this very well. So to collect data on new restaurants, they intentionally ruin your top search results. They take untested new places and inject them right into your main feed. They call this exploration noise. They know it takes you three seconds longer to find your favorite burger, but they use your screen as a testing ground. They force you to look at new places because this is the only way their system can learn if that new place is actually good or not. You pay a small price in your experience, but Uber gets the data they need for tomorrow. You might think this strategy hurts their sales, and honestly sometimes it does, but Uber isn't doing simple optimization. They are doing multi-objective optimization. They don't just want you to buy a burger right now. They want to make sure that in two years there are still hundreds of small, different restaurants on their app. The platform acts more like a market regulator. They put a hidden tax on the big restaurants by limiting their visibility just to keep the small places alive. When you build a product for millions of users, perfect for a single person is not the goal. In a large scale system, you are designing for the algorithm, the supply chain, and the business model. If you found this breakdown valuable, subscribe for more. See you in the next