Zomato APM Interview — India Daily-Orders Guesstimate

It tests how you decompose a large, ambiguous number under interruption, not whether you reach a precise figure. The Zomato product lead persona watches whether you clarify scope before computing, choose population-down or single-city-up deliberately, name a basis for every assumption such as penetration and market share, split users by how often they order, and reality-check your final number against what Zomato actually does. Composure when an assumption is challenged mid-answer matters as much as the arithmetic.

Start by clarifying what counts as an order, which geography, and what kind of day. State out loud whether you are going from national population down or from one city up, and say why. Build the path one input at a time, naming where each number comes from. Segment users by ordering frequency instead of using one flat number. Compute, then compare your result to a real anchor and revise if it is off by an order of magnitude. The interviewer rewards the visible structure far more than the final figure.

The biggest one is jumping straight to multiplying numbers before stating any structure. Close behind: asserting penetration, market-share, or frequency figures with no stated basis, using one flat orders-per-user number for the whole country, never sanity-checking the result, and freezing or restarting from scratch when the interviewer interrupts. Reciting a method name instead of actually decomposing the Indian market also reads as hollow. Each of these maps to a specific scored dimension in this practice round.

It mirrors the real round's behaviour closely: it interrupts when a number is unsupported, pushes on the weakest assumption, and never praises a mid-answer. It will not reveal your outcome or coach you during the session. The difference is that it produces a transcript-backed scorecard afterward that names the exact assumption you could not ground and the moment your structure broke, which a real interviewer would not hand you. It also stays strictly on the estimation task.

Your transcript is scored against dimensions specific to estimation: how cleanly you scoped the problem, whether you chose and justified a decomposition direction, how well you grounded each assumption, whether you segmented demand by frequency, the quality of your final reality check, and whether you held structure when challenged. Each dimension has observable signals, so two evaluators would land close. The scorecard quotes the specific moments behind each score rather than giving a single vague grade.

Do not compute anything yet. Spend the first two minutes clarifying what counts as an order, confirming it is all of India for a single ordinary day, and stating explicitly whether you will reason from the national population down or from one representative city up, with a one-line reason for that choice. Naming your direction and your scope before any arithmetic is the single strongest opening signal in this round, and it sets up every assumption you will defend later.

Expect it on every load-bearing number. When challenged, do not abandon your structure or restart. Acknowledge the specific input being questioned, give the basis you were using, adjust that one number if the challenge is fair, and continue from where you were. The round specifically rewards reworking a single input while keeping the overall decomposition intact. Panicking, going silent, or scrapping your approach is the failure mode this round is built to surface.

A strong answer sounds like: clarifying scope, then saying for example I will go population-down because order data is not public but population is, then building the funnel one named assumption at a time, splitting users into heavy, regular and occasional orderers with different frequencies, applying a Zomato-versus-Swiggy share with a stated reason, computing in lakhs and crores, then saying my number is X, Zomato publicly does roughly this many, here is why the gap exists or here is what I will revise. It stays calm and structured through interruptions.

Because in a real Zomato estimation round the assumptions are the entire test. Anyone can multiply. The interviewer wants to see whether you can defend a penetration rate, a city-tier split, or an orders-per-user figure with a reason a colleague would accept, rather than asserting it. Naming the basis for each magic number, even a rough one, is explicitly one of the scored behaviours, and the persona will keep pushing until you either ground the number or visibly flag it as a weak assumption.

Either is fine numerically, but the persona is India-context fluent and expects an Indian PM candidate to move comfortably between lakhs, crores and millions. What matters is that your units stay consistent through the decomposition and that your final number is expressed in a form you can sanity-check against a real Zomato figure. Mixing units mid-calculation without converting cleanly is a common source of order-of-magnitude errors the interviewer will catch and probe.