Zomato Personalization: Redesigning How the App Suggests Restaurants and Dishes

You are interviewing for a product manager role, and the interviewer asks how you would improve recommendations in Zomato, the food-delivery and restaurant-discovery app widely used in India. Walk through the goal of the improvement, the users you would focus on, the problems with how the app suggests what and where to eat, the solutions you would prioritize, and how you would measure the result. ```hint Start from the moment of choosing Recommendations appear at several points: the home screen, browsing by cuisine, search results and inside a restaurant's menu. Find the moment where your chosen users struggle most to decide, and design for that decision rather than for "the algorithm" in general. ``` ### Clarifying Questions - Does "recommendations" cover only food delivery, or also dining-out suggestions and table reservations? - Is the main aim more orders, more frequent ordering by existing users, or more satisfaction with what people end up ordering? - Should I improve recommendations of restaurants, of individual dishes, or both? - Do sponsored restaurant placements share the same surfaces, and may I change how they mix with organic suggestions? ### What a Strong Answer Covers - A goal for recommendations tied to Zomato's business and to a specific decision the user makes - A deliberate choice among Zomato's user segments, with concrete pain points in how they choose today - Solutions grounded in how people order food, such as dietary preference, budget, meal time and delivery time, prioritized by impact and effort - Success metrics covering conversion and satisfaction, with guardrails for restaurant partners, sponsored listings and delivery performance - Risks such as filter bubbles, users with no order history, and privacy ### Follow-up Questions - How would you recommend for a user who just installed the app and has never ordered? - Your new ranking raises orders, but most of the gain goes to a few large chains. Is that a problem, and what would you do? - How would you test whether a conversational assistant that answers "what should I eat?" is worth building before you build it?