You can answer the customer obsession interview question without ever meeting a customer. App engineers often meet their users through data. Crash and ANR reports, store reviews and support tickets all count as contact. The answer has to show you read that data as people. Start with the signal you saw. Then name the users behind it and what they were trying to do. Say what you changed because of them, especially if it differed from the ticket. Say who approved the change and what it cost. End with proof that those users were better off, in a number you can defend. Avoid the common swap where another team becomes your customer. That only fits if you build for other engineers. This post follows one Android engineer on a transit ticket app. The app froze on cheap phones at the station gate.
A customer obsession interview question can start from a dashboard
Amazon names this trait outright. Its leadership principles page carries no date. Customer Obsession opens by saying leaders “start with the customer and work backwards.” That line sets a direction. You can follow it with no meeting at all. You begin with what a user needs and reason back to the code.
Amazon’s SDE III interview prep page adds the second half. It is also undated. It says answers “should include metrics or data where applicable.” For an app engineer, that data usually lives in a console. So a dashboard is a fair place for this story to begin.
Interviewing.io’s guide to the Amazon leadership principles goes further. It shows no date. The company behind it sells mock interviews. It argues customer obsession sits under every principle. It also advises showing the user instead of saying the principle’s name. For an Amazon loop, see our post on the Amazon Bar Raiser interview.
Calling another team your customer only works when it’s true
Some prep guides offer a shortcut for engineers without user contact. InterviewPal’s March 2025 guide to this principle is one. The company sells interview prep tools. It says the customer can be a seller, a partner team or an internal stakeholder. For a platform or SDK engineer, an internal team really is the customer. Other engineers are the people using what you build.
For an app engineer, that reframe moves the story away from the actual users. Your product manager and the backend team are colleagues. The person holding the phone is the customer. An interviewer can ask what that person experienced. If the story has no answer, the reframe has cost you the question.
So keep the user in the story, even if you only met them through data.
A ticket app that froze at the station gate
Here is the example. You’re a mid-level Android engineer on a city transit app. Riders buy a ticket and show a QR code at the gate. Your sprint ticket says to rebuild the home screen with an animated header. Design has signed off and the product manager wants it next month.
Before starting, you check the Play Console. Google’s Android vitals documentation sets bad behavior thresholds. For user-perceived ANRs, the line is 0.47% overall and 8% on any one phone model. Your app sits at 0.4% overall. That looks safe. But one budget phone model is at 9%. The docs say Play may reduce an app’s visibility when it crosses a threshold. It may also warn users on the store listing.
Then you read the one-star reviews from the last month. Several say the app freezes when they open it to show a ticket. Support has a tag for riders fined for having no valid ticket. Most of those riders had bought one.
Name the rider before you name the fix
“We had an ANR problem” describes the app. Working backwards describes a person. In the example, it’s a rider on a cheap phone at a gate, with a queue behind them. They need the QR code in a few seconds. The app freezes, so they get fined for a ticket they paid for.
That sentence changes the decision. The trace showed the home screen loading the ticket, the map and the header on the main thread. The animated header from your ticket would add more work to that same screen. So the requested feature would make the rider’s problem worse.
Say the rider sentence early in the answer. It shows the interviewer where your reasoning started. The Play visibility risk is real. It’s a reason for the business, though. Lead with the person and mention the ranking second.
Bring the product manager a smaller change and its cost
The ranking of work wasn’t yours to change alone. The product manager owned it. So you brought them the per-model ANR rate, five reviews and the support tag count. You proposed two changes.
First, show the last bought ticket from a local cache before anything else loads. Move the map and header work off the main thread. Second, ship the animated header two weeks later. On phones that report low RAM, skip the animation. Android’s isLowRamDevice() method gives that signal.
The product manager agreed. The design lead didn’t like a static header on some phones. That cost belongs in the answer. Someone gave something up for the rider. You can name who it was.
Show the rider got the ticket faster
A falling ANR rate is good. It still describes the app. Close the loop on the person you named. In the example, the ANR rate on that phone model fell from 9% to 2% over four weeks. Fine-related support tickets dropped from about 40 a month to 6. Riders could now open a ticket at the gate.
Then say what you kept doing. You added the per-model ANR chart to the team’s release checklist. That way a regression on budget phones shows up before a release ships.
The station-gate answer in under two minutes
I was an Android engineer on a city transit app. My ticket was a new home screen with an animated header. Before starting, I checked Android vitals. Our ANR rate looked fine overall, at 0.4%. But one budget phone model was at 9%, above Google’s 8% per-model line. Riders left reviews about the app freezing when they opened their ticket. Support had a tag for riders fined despite having bought one.
So the users were riders on cheap phones, at a gate, needing a QR code in seconds. The trace showed the home screen loading the ticket, map and header on the main thread. The new header would have added to that.
The ranking was my product manager’s call. I brought them the numbers and five reviews. I proposed showing the cached ticket first and moving the rest off the main thread. I also proposed shipping the header two weeks later, without animation on low-RAM phones. They agreed. Our design lead wasn’t happy about the static header. I understood why.
Over four weeks, that model’s ANR rate fell from 9% to 2%. Fine-related tickets went from about 40 a month to 6. I also added the per-model ANR chart to our release checklist.
The rider, the app and its numbers are made up for illustration. Google’s thresholds are real. Swap in your own signal, your own users and the result you measured. Keep the same order. Start with the signal and the person, then the change and its cost. End on the proof.
If your story has no number to show for the user
Some engineers never see a dashboard. Some teams don’t share analytics with engineers. Others shipped and moved on before results arrived. Don’t invent a figure. A follow-up can expose it in one question. Our post on behavioral interview follow-up questions shows how those probes work.
Use the closest evidence you actually had. That might be a count of support tickets, a QA report or three reviews you read. Say plainly that you didn’t have the outcome data. Then say what you would have measured and who owned that number. For the transit app, that would be the per-model ANR rate, owned by the release manager. It shows you knew how to check the rider was better off. The interviewer can score that judgment even without the result.