Asked about a risk you took that didn’t work out? Judge the bet by what you knew on the day. The result comes second. Build the answer in four parts. First, say what you decided and what you expected to gain. Second, put a number on the downside you accepted. Name who would carry it. Third, tell what actually happened and what it cost. Last, give your verdict on the bet itself. Was it bad luck, a check you skipped, or a bad call? The interviewer is listening for that verdict. A candidate who calls the risk a mistake sounds like they’d skip the next one. A candidate who blames bad luck sounds like someone who can’t learn from it. Many honest stories land in between. The bet was reasonable. One available check would still have changed it. This post builds one Android story around that verdict.
Interviewers ask about a failed risk to see if you’d take the next one
Amazon’s Bias for Action principle ends with “We value calculated risk taking.” A failed-risk question tests that value from the hard side. Other companies ask it as a failure question with a risk attached. Either way, the interviewer has two worries. One is that you take risks without thinking. The other is that one bad result taught you to stop taking them.
Most weak answers feed the second worry. The candidate describes the risk, describes the damage and ends with “I learned to be more careful.” That sentence sounds mature. Yet it tells the interviewer that your lesson was to avoid the next bet. For a company that values speed, that’s a worse signal than the failure itself.
Annie Duke, a former professional poker player who writes about decisions, has a name for the trap. She calls it “resulting,” which means judging a decision only by how it turned out. She explained it in a 2018 podcast interview posted on her site. In her words, “the quality of an outcome and the quality of the decision are only loosely correlated.” Your answer has to show you know that difference. It also has to show you don’t hide behind it.
The running example: raising the minimum Android version
Mateo is an Android engineer on a delivery app. The app still supports Android 8 and 9. A legacy file-storage path exists only for those versions. That path causes a cluster of crashes. It also slows every storage change the team makes. Mateo proposes raising the minimum version to Android 10 and deleting the old path.
He checks the usage data first. About 4% of active installs run Android 8 or 9. Those users keep the version already on their phones. They just stop getting updates, because Google Play filters an app from devices below its minimum version. Mateo writes a one-page proposal with the 4% figure and the expected crash fix. His lead approves it.
The proposal is your evidence that the risk was calculated. It named the gain and the downside. It put a number on the users affected. A second person signed off.
Name the downside you accepted and who carried it
Interviewers can’t tell a calculated risk from a reckless one unless you state the downside. So say it in plain terms before you describe the result. In Mateo’s case, 4% of users would stop getting fixes. He also knew who would hear about it first. That was the support team, so he warned them before release.
A risk you carry alone is easy to take. A risk that lands on users or another team needs that team to know first. Saying you told support shows you didn’t push the cost onto them in silence.
Tell the result, then sort the bet into one of three verdicts
Two weeks after release, the downside turned out larger than 4%. A logistics customer gave its drivers company handhelds running Android 9. That one account brought in about a tenth of the app’s revenue. Its drivers were stuck on the old version. They hit a payment bug the new release had fixed. The account manager escalated. The team kept a maintenance branch for the old version for three months. That cost Mateo about three weeks of work over the quarter.
Now give your verdict. There are three honest ones. Each leads to a different answer to “would you take it again?”
The first is bad luck. You had the information you could reasonably get. The downside happened anyway. Then the answer is yes, unchanged. Use this one sparingly. Said too quickly, it sounds like an excuse.
The second is a missed check. The bet was sound. Yet one piece of information was available. You didn’t look for it. Then the answer is yes, with that check added. Many real stories belong here. Mateo’s is one of them. The usage data counted devices, not revenue. Sales knew which enterprise accounts ran which devices. One message to the account managers would have found the logistics customer before release.
The third is a bad call. Even with what you knew, the downside was too large or too poorly understood. Then the answer is no. Say what you’d do instead. This verdict is rare in good stories. Still, it’s believable when the story supports it.
Mateo’s answer to “tell me about a risk you took that didn’t work out”
Our delivery app still supported Android 8 and 9. Those versions needed an old storage path that caused most of our storage crashes. I proposed dropping both versions. About 4% of our users were on them. They’d keep the current app but stop getting updates. My lead approved it. I warned support before the release.
It didn’t go to plan. One logistics customer ran Android 9 handhelds for all its drivers. That account was about a tenth of our revenue. Two weeks later its drivers hit a payment bug. Our new release had fixed it. They couldn’t install that release. We ran a maintenance branch for three months. It cost me about three weeks of work.
The crash cluster did go away, so the main bet paid off. Where I went wrong was the check. I measured users by device. I didn’t measure revenue by account. Sales had that data. One message would have surfaced the customer.
I’d make the same call again, with that check first. I did it on our next version bump. I sent the planned cut-off to account managers two months early. One customer needed time, so we gave them a date instead of a surprise.
We wrote this answer for the post. The numbers are invented. Use your own risk. Keep your real figures even when they’re unflattering.
Look at where each part comes from. The first paragraph is the proposal. It gives the gain, the 4% and the sign-off. The second paragraph gives the cost in weeks. That’s easier to believe than “it was painful.” The third paragraph separates the part that worked from the part that didn’t. The last paragraph gives the verdict and proves it with a later decision. The later decision carries the most weight. It shows Mateo kept taking this kind of risk and added one step before it.
If your story is a production failure, the same structure holds. See why a production incident is a weak failure story. It puts the decision before the damage. For a decision made under a deadline with missing data, see the incomplete information question.
“Tell me about a risk you decided not to take” can come next
Some interviewers flip the question after a failed-risk story. They want to know whether the failure made you timid. Have a short second story ready, where you turned a risk down for a stated reason. Give the reason as the size of the downside. A reason that sounds like fear of your last mistake confirms the worry.
Mateo’s version could come from the same codebase. Months later, a teammate proposed replacing the payment SDK one week before a peak sales period. Mateo argued for waiting until after the peak. A payment failure that week would hit every user at the busiest time. They swapped the SDK three weeks later, after the peak. In the room, Mateo can tie the two stories together with one number each. The version bump risked 4% of users with a known cost. The SDK swap risked every user during the busiest week.