Tell me about a time you failed.

What is actually being scored

This is an accountability probe, and the rubric has less to do with the failure than with your relationship to it. Interviewers are checking three things: that you can identify a genuine failure without prompting, that you attribute it to your own decision rather than to circumstance, and that something concrete changed afterwards. A candidate who cannot name a real failure reads either as inexperienced or as lacking self-awareness, and both are disqualifying at senior level.

There is a fourth, quieter check on proportionality. A staff engineer whose worst failure is a missed sprint commitment has either had a very sheltered career or is not being candid. The scale of the failure you choose signals the scale of responsibility you have actually held.

The two failure modes

The fake failure is the more common one. "I take on too much work," "I was too much of a perfectionist," "I cared too much about quality" — these are strengths in costume, and interviewers recognise them instantly. The cost is not neutral: you have now used your one chance at this question and signalled that you are managing the interviewer rather than answering.

The blame-shifted failure is subtler and more damaging. The story is real, but the causal chain lands somewhere else: requirements were unclear, QA missed it, the deadline was imposed, a dependency was late. Even when all of that is factually true, an answer built on it fails the probe, because the question is what you did. The test to apply to your own story is whether the sentence "and the decision that caused this was mine" fits naturally. If it does not, pick a different story.

Structuring it

Use STAR, but weight it deliberately for this question. Situation and task get one or two sentences — just enough context to make the stakes legible. Action is where most of the answer lives, and it must include the specific decision you got wrong and, ideally, the signal you had available and did not act on. Result covers the consequence honestly, then the change.

The change is the part candidates cut short, and it is the part that earns the score. "I learned to communicate better" is worthless because it is not falsifiable. "I now write a one-page rollback plan before any migration touching production data, and I have used it on four migrations since" is specific, behavioural, and evidently real.

A worked example

"About eighteen months ago I owned a schema migration that added a column and backfilled it for a table with roughly forty million rows. I tested it on staging, where the table had about two hundred thousand rows, and it completed in under a minute. I scheduled it for a weekday evening rather than a maintenance window because I had concluded the change was low risk.

The backfill acquired a lock that blocked writes on the table, and within about four minutes the checkout service was timing out. We were degraded for roughly twenty-five minutes while I cancelled the migration and waited for the transaction to roll back.

The decision that caused it was mine specifically: I generalised from a staging dataset two orders of magnitude smaller and did not check the lock behaviour of the statement I was running. A colleague had actually asked whether we should batch the backfill and I said it was not necessary.

Two things changed. I now batch any backfill over about a hundred thousand rows and run it as a separate step from the schema change, and we added a pre-migration checklist that requires the production row count and the lock mode of each statement written down before approval. I have run six migrations on that checklist since and we have not had a repeat. The harder lesson was about the colleague's question — I now treat a dissenting question on a risky change as a signal to slow down rather than something to resolve quickly."

Note what that does. The failure is real and had customer impact. The cause is a specific decision, stated plainly. The ignored warning is volunteered rather than hidden, which is counter-intuitive but reads as candour and pre-empts the follow-up. And the change is concrete enough to be checked.

Preparation

Have two of these ready, at different scales, because a follow-up of "tell me about another one" is common and improvising the second is where the fake failure slips out. Choose failures at least a few months old so the resolution has been tested, and avoid anything still live or politically sensitive at your current employer — you cannot speak freely about it, and the hedging will show.