Google PM Interview — Elderly Shopping App India TAM

It tests how you turn an under-defined business question into a defensible, quantified recommendation. The interviewer is not waiting to check your final number against a known answer. They watch whether you scope the question before any math, state the reason behind each assumption before you use it, choose and name an approach, segment the population on drivers that move the answer, and sanity-check the result against a real anchor. For an elderly shopping app in India that means reasoning about the sixty-plus share, senior smartphone access, caregiver-operated usage, and realistic spend, not headline population times a guess.

Start by defining the metric precisely: are you sizing users or annual revenue, and over what time horizon. Say which approach you are taking and why it fits this market. Break the elderly India population into segments that genuinely change the answer, such as urban versus rural, smartphone access, and digital-payment comfort. Attach a justified penetration rate and a realistic per-user spend to each segment rather than one blended guess. Compute out loud. Then sanity-check the total against a known real-world anchor and, if you have time, rebuild it with a second method to triangulate.

The biggest one is jumping into arithmetic before clarifying what TAM means here. Close behind: using a round number for population, penetration, or spend without stating why before applying it; reciting a named framework instead of producing a decision; never checking the final figure against a real anchor; and treating the elderly and India context as cosmetic by running a generic global-app model. Freezing or arguing when the interviewer rejects a number, instead of re-deriving it from a defensible assumption, also reads as a clear negative.

The behaviour is built to mirror the real analytical loop closely. The persona interrupts on unjustified numbers, says a flat no to shallow explanations, and pushes you to go deeper rather than accepting a first pass. It will not praise you mid-round and it will not teach you the method. The main difference is that it is available on demand and produces a transcript-backed scorecard afterward. It will not give you a hire or no-hire signal, because a real loop decision is made by a committee across several independent rounds.

Your transcript is scored against dimensions drawn from how the analytical round is actually evaluated: how precisely you scoped the metric, whether you justified assumptions before using them, the quality of your segmentation, whether you sanity-checked and triangulated, and how you recovered when a number was challenged. Each dimension has observable anchors so two evaluators would land close. You receive a written scorecard that names the specific moment an assumption went unjustified or a number was not anchored, rather than a single overall grade.

Do not compute anything. Use the opening to define the metric: state whether you are sizing users or annual monetisable revenue and over what horizon, and say why that choice fits the decision. Name the approach you will take and why it suits this market. Surface the one or two scoping questions that actually change the model, such as whether assisted caregiver purchases count as elderly demand. Landing this scoping move cleanly in the first two minutes is the single strongest early signal in this round.

Do not defend the number for its own sake and do not freeze. Treat the rejection as a prompt to re-derive. Say what the input was meant to represent, name a defensible reference point you can anchor it to, and rebuild the input from that reference out loud. The recovery itself is a strong positive signal in this round. Arguing that your original number was fine, or going silent, is the negative pattern the interviewer is specifically listening for.

It opens with scoping, not math: this person decides users versus revenue and a time horizon before anything else. Every input arrives with its reason attached first. The population is segmented on drivers that move the answer, with caregiver-operated usage handled explicitly rather than ignored. The candidate computes out loud, lands a number, then immediately checks it against a real anchor like India total internet users or e-commerce GMV, and offers a second method to triangulate. When challenged, they re-derive calmly instead of arguing.

Because it changes the model rather than decorating it. The elderly are a small share of India's population, senior smartphone access and digital-payment comfort are materially below the working-age average, and a large fraction of elderly online shopping is operated by a caregiver or family member. A credible answer treats the sixty-plus group as the addressable base, models a distinct and lower senior penetration, and handles assisted use as a real behaviour. Candidates who run the same model they would for a generic Indian shopping app are flagged for treating the context as cosmetic.

No. The interviewer often already knows a ballpark and is watching how you break an impossible question into possible pieces. The strongest predictor of a positive score is stating the reasoning behind each assumption before you use it in the calculation, not numeric accuracy. A clean, well-anchored process with an order-of-magnitude answer beats a precise-looking number built on unjustified inputs. Define success metrics with real denominators if asked, because a number with no way to validate it is treated as incomplete.