How would you improve Google Maps? A worked PM interview answer
The Google Maps question is asked so often that the average answer is memorised: five personas, four features, no decision. Here is a version that picks one broken moment and defends a single v1.
Quick Answer
Pick one moment in the product that is genuinely broken, for one segment, and build a v1 for that. For Google Maps the strongest available answer is the last 200 metres: routing is excellent, arrival is not. Everything after that choice is defending it against the alternatives you did not pick.
Why the common answer scores badly
Search this question and you will find the same shape everywhere. Describe the product. List the users: private vehicle owners, bikers, commuters, walkers, tourists. List pain points. Suggest features. Prioritise on a cost-versus-benefit grid. It looks like a framework, and it produces nothing an interviewer can disagree with, which is the problem. A Google interviewer has heard this structure several hundred times. Naming five personas and then treating them equally is not segmentation, it is a table of contents.
Find the moment, not the feature
Maps is a mature product. There is no missing feature a candidate will name in twelve minutes that the Maps team has not already considered. What there is, in any mature product, is a moment where the product's promise quietly fails.
- Routing to a street segment works. Arriving at the right door of a hospital, campus, mall, or apartment complex often does not.
- The driver circles, reverses, parks far away, or walks to the wrong entrance.
- It is high-frequency, emotionally loaded, and it happens at the end of the trip, which is the part the user remembers.
- It is measurable from telemetry Google already has: reroutes within 300 metres, U-turns, low-speed circling time.
Commit to a v1 and say what is not in it
Name the thing. Arrival Assist: about a mile out, Maps surfaces an arrival goal rather than a pin. Main entrance, ER entrance, parking garage, rideshare drop-off, accessible entrance. It highlights the lane or side of the street, shows parking likelihood and walk time, and hands off to walking navigation from wherever the car actually parked. Then immediately say what you cut: no guaranteed parking reservations, no payments, no indoor navigation, no generative trip planning. The exclusions are where interviewers hear judgment. A candidate who only adds is a candidate who has never shipped.
Explain why this one and not the others
You will be pushed on alternatives, so bring them yourself. Broad local discovery sits closer to the ads business but further from the core job and is harder to measure. A generative travel co-pilot demos well and depends on model reliability in a context where wrong guidance is a safety issue. Deep indoor mapping needs venue-by-venue data partnerships that will not land inside a year. Arrival Assist wins because it uses assets Google already has, Street View, business profiles, the POI graph, navigation telemetry, and because it improves the promise Maps already makes rather than making a new one.
Pick a metric that could actually go down
The north star here is successful arrival rate: navigation sessions where the user reaches the intended arrival point without excessive circling, rerouting, manual search, or immediate correction. It is a good metric precisely because it can fail. Then name the guardrails that would make you roll back: navigation distraction reports, accessibility misinformation, entrance-recommendation precision, local business disputes. A metric with no failure mode attached is a metric you have not thought about.
The sixty-second version
If the interviewer cuts you off, this is what should be left: "Maps is great at routing and weak at arrival. I would build Arrival Assist for urban drivers going somewhere unfamiliar, launching only on place categories where we have high-confidence entrance and parking data. North star is successful arrival rate. I am explicitly not doing parking reservations or indoor nav in v1, because wrong guidance here erodes trust faster than a missing feature does." Practise saying that without notes.
Practice question
Twelve minutes, out loud, no framework recitation: "How would you improve Google Maps?" Rules for the rep: name one segment inside the first ninety seconds, one v1, and three things you deliberately excluded. If you catch yourself listing five personas, stop and restart.
Turn this into reps
Run this as a product design case on PMMockr and get graded on framing, tradeoffs, and metrics separately. Most candidates score fine on structure and lose the round on tradeoffs, because nothing in their answer was ever at risk of being wrong.
FAQ
Should I ask clarifying questions before answering?
Two or three that change your answer. Whether the goal is engagement or retention changes what you build. Asking whether it is web or mobile, when you are going to say mobile regardless, burns time you need for tradeoffs.
Is it risky to pick a narrow improvement on a product this big?
The opposite. Breadth on a product this mature reads as avoidance. Interviewers are checking whether you can choose under uncertainty, and a narrow, defended bet is the only way to show that.
How much should I know about how Maps works technically?
Enough to know which constraints make your idea hard, such as entrance data confidence and category coverage. Not enough to start designing the routing system, which is a different round.
What if the interviewer says my idea already exists?
Ask what the current version does not solve, then narrow to that gap. Being partly wrong about product state is normal and recoverable. Refusing to update is not.