Tell me about a time you pushed back on shipping something you did not think was ready.
answer
- the risk you saw, concretely
- the proof you gathered first
- what you asked for, precisely
- the smaller option you offered
- the outcome and what stayed changed
basics
~20 sTests whether your quality standard survives pressure and whether you defend it with evidence rather than instinct. Answer with the specific risk you found, reproducible proof, a bounded ask with a smaller alternative, and what shipping finally looked like.
how to answer
5 beats- the release and the risk you sawSet the scene in a few sentences: what was due to ship, who wanted it, and the specific failure you were worried about. Keep this short — roughly a fifth of your airtime — and make the risk concrete rather than a general unease.
- the evidence you gathered before you spokeDescribe what you did to turn an instinct into something checkable: an exploratory pass, a failing case, a gap you could point at in numbers. This beat is what separates a professional objection from a mood, so give it real weight.
- the ask, bounded and with an alternativeState exactly what you requested — how long, for what, and what you would trade — and the smaller option you offered so the answer was not a binary. This and the previous beat should be the bulk of the answer, around sixty percent.
- how the conversation actually wentSay who you spoke to and what they said back, including any pushback you got. Do not sanitise it into everyone immediately agreeing; a real negotiation with a decision at the end is more convincing than instant consensus.
- the outcome and what stayed changedClose with what shipped, one number if you have one, and the thing that outlived the episode — a check added, criteria agreed in advance. Give this about a quarter of your time; being right is worth less than making the next argument unnecessary.
your answer
4 story prompts- Pick a time you were the one who said not yet, not one where you agreed with someone else.
- Write down the evidence you had before you spoke — a failing case beats a feeling.
- Have the exact ask ready: how long, for what, and what you would trade for it.
- Note how the person you pushed back on treated you on the next piece of work.
draft and rehearse your own answer in a learn session
go deeper
Probes whether your quality standard survives contact with commercial pressure, and whether you defend it with evidence and options rather than instinct and volume. Interviewers watch for conviction that stays specific, an ask small enough to be granted, and whether both the delivery and the working relationship came out intact.
A client wanted refunds live before their support team's busiest stretch of the year. The happy path worked, but our coverage on the refund flow was thin — 62 percent, with none of the partial-refund cases touched at all. I did not want to be the person who says it does not feel ready, so I went and got specifics first. I spent an afternoon writing four exploratory cases by hand, and two of them failed: a partial refund against a split payment returned the full amount, and a retried refund created a second one. Then I asked for two days rather than for more time. I brought the two failures, the list of exactly what I would cover in those two days, and an alternative in case the date was immovable — release refunds to their internal support tool only, so a person is in the loop on every one, and open it to customers afterwards. They took the two days. We found a third defect in the same family and took refund-path coverage to 88 percent before it went out. What made it work was not firmness; it was that I turned an instinct into two reproducible failures before I opened my mouth, and that the smaller option meant no was not the only answer available to them.
The strength here is sequence: evidence first, then a bounded ask, then a fallback that keeps the date alive in some form. Two reproducible failures do the persuading, not conviction. It would downlevel if the ask had been for more time in general or if no alternative had been offered.
Their checkout had been down fifty-two minutes the previous evening — a webhook handler that stopped acknowledging under load — and the client's delivery director wanted the fix re-released that same afternoon. I was accountable for the release and I said no to the afternoon, which is not a comfortable thing to say to the person paying. What I gave them instead of a flat no was a shape. First, we reproduce the failure under comparable load in a soak run before anything ships, because we still had two competing theories for the cause and shipping into a guess is how you get the same outage twice. Second, when it does ship it goes to four percent of traffic for an hour with automatic rollback wired to the acknowledgement lag. Third, I put the cost of waiting on the table myself — about a day of manual processing for their support team — so they could weigh it rather than have me weigh it for them. The soak run eliminated one of the theories in a little over three hours, which is the entire argument for having run it. We shipped that evening, canary first, and it held. The follow-through mattered more than the pushback: the lag alarm and the soak stage became a standing part of that team's release path, so nobody has to win the argument again.
The senior move is refusing the timing while handing back a concrete plan, and naming the cost of waiting out loud instead of letting the client discover it. Making the soak stage and the alarm permanent is the prevention that stops this being a one-off win. Winning the argument with no lasting change would downlevel it.
Raising the concern at all, to the right person, early enough to matter, is the bar. Bring something concrete — a failing case, a gap you can point at — rather than a feeling, and show you took the decision gracefully when it went the other way.
Your ask should be bounded and costed: how many days, for exactly what coverage, and what you would drop instead. Bring at least one reproducible failure and one alternative that lets the date survive in some form.
You are usually accountable for the release, so show the decision rather than the lobbying: what you held, what you offered in exchange, how you made the cost of waiting visible to the person paying, and what you changed so nobody repeats the argument.
Frame where the authority to say not yet sits and how it is used without becoming a bottleneck — release criteria agreed before the pressure arrives, an escalation path, and an evidence standard that holds across teams.
saying these in an interview costs you the question
- Pushing back on instinct with no evidence anyone else could check
- Framing it as engineering against the business rather than a shared risk
- Asking vaguely for more time instead of a bounded, costed ask
- Offering no alternative, so the only possible answers were yes or no
- Winning the argument and never saying what shipping actually looked like
- A story where you were overruled and then quietly let it fail
- What would you have done if they had said no?Show that you had a plan for losing. Strong answers name the containment they would have asked for instead — a flag, a limited rollout, extra monitoring, a rollback rehearsal — and say that they would have written the accepted risk down and named its owner. Threatening to escalate as your only fallback reads as brittle.
- How did that relationship end up afterwards?Interviewers are checking you can hold a line without burning the person on the other side. Give a small piece of evidence — they came to you earlier next time, they asked you into the planning conversation, you kept working together for months. If it was strained, say what you would do differently to keep the disagreement about the risk.
- Have you ever pushed back and turned out to be wrong?Say yes and mean it. Pick a real one, say what your evidence missed, and what changed in how you assess readiness afterwards — a threshold you loosened, a habit of checking blast radius before objecting. Claiming your caution has always been vindicated makes the whole answer sound rehearsed.