Tell me about a time you drove a technical decision that several teams had to agree on.
answer
- name the decision and what blocked it
- two or three real options, with trade-offs
- the criterion both sides accepted
- who decided, and by when
- result in numbers, plus what you kept
basics
~20 sTests whether you can own a decision process across team boundaries, not just hold a strong opinion. Answer with the options you framed, the criteria everyone accepted, who decided by when, and how the losing side stayed on board.
how to answer
6 beats- the decision on the table and why it was stuckOne or two sentences: what had to be chosen, who had to live with it, and what was blocking it. Keep this plus your role to about fifteen to twenty percent of your airtime — long setups are the most common way this answer fails.
- the options you put in front of peopleName two or three real alternatives, including the one you did not prefer, and say what you compared them on. Stating the criterion before the scores is the move that shows process rather than advocacy.
- how you ran the debate and who decidedThis and the previous beat are the bulk of the answer, around sixty percent together. Say how long you gave the discussion, where it happened, and who held the decision. Naming a decider up front is worth saying out loud even if that person was you.
- how you closed with the people whose option lostDescribe one concrete thing you gave them — an escape hatch, the first migration done for them, a date to revisit. Buy-in that costs you something is the part interviewers believe.
- the outcome in numbersResult and reflection together are the last twenty to twenty-five percent. Give a before and after on the thing you said you were optimising, and say whether the decision was still standing later.
- what you kept as a mechanismClose with what outlived the decision: a template, a rule about who arbitrates, a review cadence. If nothing outlived it, say what you would institutionalise now — that is still a credible close.
your answer
5 story prompts- Pick a decision whose process you owned, not one where you merely held a strong opinion.
- Choose a story where at least two teams had to live with the outcome.
- Write down the one number both sides agreed the decision should optimise.
- Name who held the final call and the date the debate closed.
- This can be the same story as your influence-without-authority answer, re-angled.
draft and rehearse your own answer in a learn session
go deeper
This probes influence and process ownership across boundaries you do not control. The interviewer wants evidence that you can convert a preference argument into a decided, documented outcome — framing options, establishing shared criteria, making accountability explicit, and keeping working relationships intact with the people whose option lost. A strong answer proves the decision would not have landed without you.
We renewed TLS certificates by hand at a fourteen-person startup, and it caught up with us when an internal certificate expired and took the checkout path down for forty-three minutes. I picked up the follow-up action, which was really a decision three teams had to live with. Rather than opening the argument cold, I wrote a one-page options page: keep the renewal script and add alerting, move to automated issuance with short-lived certificates, or pay a managed vendor. For each option I put down setup cost, who gets paged when it breaks, and how much warning we would get before an expiry. The web team wanted the vendor because it meant learning nothing new; my two infra teammates wanted automation. Instead of arguing preference, I asked what number we were optimising, and we settled on warning time before expiry. The vendor gave us seven days and a support ticket. Automation renewed at thirty-one days out and alerted at fourteen. I asked our engineering lead to be the decider, gave everyone until Thursday to comment, and then we spent thirty-five minutes in a room and closed it. I wrote it up as a short record with the rejected options and the reason each lost. Median warning before expiry went from two days to thirty-one across nineteen certificates. I also did the migration for the web team's two services myself, so saying yes cost them nothing.
The signal is carried by the criteria beat — agreeing what number mattered before comparing options — and by doing the losing team's migration, which is buy-in that cost the speaker something. Vaguer verbs in place of the options page, or no named decider, would downlevel this immediately.
About a year later the same class of problem surfaced one layer down. An expired mesh certificate took six internal services offline for twenty-six minutes, and the argument afterwards was not technical, it was about ownership. Platform wanted every team to hold its own certificates; two product teams wanted platform to issue everything centrally. I owned the call. I gave the discussion a hard end date of eleven days, and I opened by writing down what we were not deciding: not the certificate authority, only who carries the renewal duty. That killed roughly half the thread on its own. Then I asked each side for the cheapest version of their own position and costed both against the same measure — how much warning we would get on the worst certificate, not the average one. Central issuance came out at thirty-four days. Per-team ownership came out at nine, because two of the teams had no on-call rotation at all. I decided for central issuance, said out loud that I was accountable if it slowed anyone down, and gave the loudest objector something real: a break-glass path to mint a certificate without filing a ticket, and the first migration done by my team. Worst-case warning across twenty-three endpoints moved from six days to thirty-four, and there was no expiry-caused incident in the five months that followed. I kept the format too — three options, one decider, one date — and we reused it on the next two platform arguments.
Deciding rather than waiting for consensus is the move, and doing it with the worst case rather than the average as the measure. Saying accountability out loud, and the break-glass concession, are the senior moves. Drop the timebox and the concession and it reads as an opinion that happened to win.
Your scope is your own task or subteam — a library choice, a test approach, a naming convention. Show that you gathered evidence, wrote the call down, and got a senior to confirm it rather than deciding by preference.
Expect feature or peer scope across two teams. The interviewer is listening for mechanics: an options write-up, a meeting you actually ran, a trade-off you named out loud, and a decision someone can still find in writing.
Team and production scope. Show that you set a deadline for the debate, named who decides, absorbed the strongest objection with something concrete, and left behind prevention so the same argument does not recur.
Org scope. The signal is a repeatable mechanism — a decision-record format, a review forum, a rule about who arbitrates — that other people run without you in the room, plus evidence it survived your departure from the topic.
saying these in an interview costs you the question
- Describing a decision you merely had an opinion about, not one you drove
- No named decider anywhere — the debate just quietly faded out
- Winning on title or volume instead of evidence and criteria
- Never mentioning the people whose option lost
- All we, no I — your own moves never surface in the story
- No result and no artifact anyone could point to afterwards
- How long did you let the debate run before you called it?Give a real number and the reason behind it. Say what you did when the clock ran out: closed comments, moved to a decision meeting, or escalated once. Show that the timebox was announced up front rather than invented afterwards when you got tired of arguing.
- Who actually made the final call?Name the seat, not the person: you, your manager, an architect, a group vote you facilitated. Interviewers are checking that someone was accountable and that everyone knew who it was before the discussion started. Saying we all just agreed is the answer that reads as no process at all.
- What happened to the people whose option lost?Describe something specific you gave them — the first migration done for them, an escape hatch, a review date to revisit. Then say how the working relationship went afterwards. Avoid relitigating why you were right; the probe is about buy-in surviving the decision, not about the technical merits.
- Did the decision hold up?Answer honestly, including the part that aged badly. If it was revisited, say what triggered the revisit and whether the written record made that cheaper. A candidate who can name one thing the decision got wrong reads as more credible than one whose call was flawless.
## One story for every variant This prompt family shows up in many wordings and they all want the same thing. Common variants: - how do you build alignment when two teams disagree; - tell me about a time you influenced a decision you did not have authority over; - describe a technical decision you owned end to end; - how do you handle a debate that will not resolve. Answer all of them with one story, re-angled by which beat you expand. ## What is actually being scored The interviewer is separating three populations. 1. **First**, people who had opinions and lost or won by accident. 2. **Second**, people who won by escalating to someone with more authority every time. 3. **Third**, people who ran a process: framed options, established what everyone was optimising for, set an end date, made the accountability explicit, and closed the loop in writing. Only the third population is being hired for the influence signal, and the difference is audible within the first thirty seconds — it is the difference between I thought we should use X and there were three ways to do it and here is how we chose. ## Weak versus strong on the same story **Weak:** we needed to pick a queueing approach, I thought the managed option was better, I explained it a few times, eventually people came around and we went with it. Nothing here is checkable. Came around is not a mechanism. **Strong:** the same story with four additions — - the options written down before the argument, - a criterion both sides accepted before seeing the scores, - a named decider and a date, - and one concrete thing given to the side that lost. Notice that the strong version is not longer by much. It is the same length with the vague verbs replaced. ## Evidence types that land - An **artifact** you can describe: a one-page options doc, a comparison table, a decision record. - A **criterion expressed as a number** rather than an adjective — recovery time, warning before failure, migration hours per service, monthly spend. - A **named seat** that decided. - A **date** the debate closed. - A **follow-up fact**: it was still in place months later, or it was revisited once for a reason you can state. Two of these five is usually enough; five is a story that sounds rehearsed rather than remembered. ## Framing the debate rather than winning it The single highest-value move to narrate is **agreeing the criteria before revealing the options**. It converts a preference fight into an arithmetic exercise and it is the thing most candidates never mention because they did it instinctively. If your story has it, say so explicitly. If it does not, say what you would do differently — that is the reflection beat and it recovers most of the credit. ## How the bar moves with level - A **mid-level** candidate is credible driving a decision between two teams with a document and a meeting. - A **senior** candidate is expected to have handled the case where consensus never arrived: someone had to decide anyway, and the story should show you either took that on or arranged for it, without the debate simply running out of energy. - A **principal** candidate is expected to have generalised — the interesting artifact is not the decision but the format the decision was made in, adopted by people you do not manage. ## Two traps 1. Do not turn this into a story about a difficult colleague; the moment the narrative becomes about a personality, the process signal disappears. 2. And do not claim a decision as yours when you were a participant — interviewers probe with who decided and what would have happened if you had said nothing, and an inflated claim collapses fast.