Decision making interview questions for hiring teams, with what to listen for

Each of these decision making interview questions hands the candidate a call with evidence missing: ship or stop a test, release on a partly spent error budget, fix forward or roll back. Each comes from a Hirezen interview plan, with three signals of a strong answer and an example of a Yes, a Not sure and a No.

Updated . Written by Hirezen.

The signals, what a strong answer shows, come from the plans, and grouping these five as decision making is Hirezen’s call. The examples are written for this page; each Not sure ends with the follow-up that would settle it. Across all five, listen for whether the candidate makes the call, says who owns it and says what would change it.

1. Ship, stop or wait

A new ranking model, v7, beat the old one offline. Nine days into a live test, click-through is down 1.8% and conversion is flat.

The question

“It is day nine. Ship v7, kill it, or extend the test — and tell me what decision rule you wish had been written down before the test started.”

Ship, stop or wait: each signal, and what Yes, Not sure and No can sound like
SignalYesNot sureNo
Names who owns the decision and what they agreed to before day one“The product lead owns it, and before launch nobody agreed what result would stop the test.”“The product side owns the call.”Ask: What was agreed with them before the test started?“It’s an ML decision, so our team calls it.”
Considers the cost of extending — users in the treatment arm are seeing a worse product — and weighs it rather than ignoring it“Half our users are in the arm that’s losing on click-through, and every extra day keeps them there.”“Extending has a cost.”Ask: What does it cost, and who pays it?“Nine days isn’t enough data. Let’s run it two more weeks.”
Says what they will write down for the next test, and it is a rule with numbers rather than a lesson“Next time: click-through is primary, and the test stops if the fallback rate goes more than a point over the old model’s.”“Next time we should agree the rules up front.”Ask: Which metric, what threshold, how long?“Lesson learned: keep a closer eye on latency.”

2. A release on a partly spent budget

About two-thirds of the error budget for the rolling 30-day window is gone. The team has fixed the deposit flow and wants to ship it on Thursday, with deposits for parties of six or more, which restaurants have asked for.

The question

“Do they ship on Thursday? Tell me what the budget says, what you decide and who decides it, and what you would write down so the next one is not settled in a meeting.”

A release on a partly spent budget: each signal, and what Yes, Not sure and No can sound like
SignalYesNot sureNo
Separates the fix from the new feature, and would ship one without the other“The fix can go out on its own on Thursday, if a test reproduces the failure and we can roll it back fast. The large-party deposits wait.”“Let’s slow this release down.”Ask: Could the fix ship without the new deposits?“It’s one release, so it ships or waits as one.”
Says who decides — product and reliability, agreed in advance, with a named tie-breaker — so the budget sets a default, not a veto“It isn’t mine alone to decide. Product and I make the call, and if we split, the head of engineering decides. That should be written down before the next one.”“Product should have a say.”Ask: Who decides if product and reliability disagree?“Two-thirds of the budget is gone, so nothing ships until it resets.”
Writes the policy down — costly incidents, an exhausted budget, known peaks, exceptions — and has it signed before it is needed“We write down what happens after a costly incident, when the budget runs out and before a known peak, and who can grant an exception. Product and I sign it this month.”“We should have a policy for this.”Ask: What does it cover, and who signs it?“We’ll decide this one on Thursday morning, once we see the numbers.”

3. A decision with one number missing

A migration plan moves a database to the cloud provider’s managed PostgreSQL. The database depends on an in-house extension the managed service cannot load, and nobody has estimated what removing it would cost.

The question

“The plan moves the portal database to the provider's managed PostgreSQL. Using the cost sheet and what you know about the extension, make the call — managed or self-run — and price it over three years.”

A decision with one number missing: each signal, and what Yes, Not sure and No can sound like
SignalYesNot sureNo
Turns the unknown into a break-even: removing the extension can cost up to about $104,000 — roughly seven engineer-months at the sheet's rate — before self-run is cheaper“We don’t know what removing it costs, but we know the ceiling: about seven engineer-months before self-run wins.”“It depends what removing the extension costs.”Ask: At what cost does self-run become cheaper?“Managed is the default, so managed. We’ll sort out the extension during the move.”
Proposes a time-boxed estimate of removing the extension before deciding, with a date that still fits the contract“Give me two weeks to estimate removing it. That still leaves time to move before the contract ends.”“I’d want an estimate before deciding.”Ask: By when?“We can’t decide until someone has fully scoped the extension work.”
States the decision and the evidence that would reverse it“Managed, unless the estimate puts removing the extension past about seven engineer-months. Then self-run.”“I lean toward managed.”Ask: What would change your mind?“Both options have trade-offs. It’s a call for leadership.”

4. Take it, change it or kill it

The question

“Here's the proposal on the table. My web team has capacity for one substantial thing this quarter, and they've scoped two: add FAQPage schema across our whole template library, and publish an llms.txt at the root. They're keen, it's specced, and it can start Monday. Take it, change it, or kill it — and if you kill it, you have to tell me what you'd spend the same capacity on instead, and why that's a better use of an engineering quarter.”

Take it, change it or kill it: each signal, and what Yes, Not sure and No can sound like
SignalYesNot sureNo
Proposes a concrete alternative use of the same capacity and can defend the swap on expected value“Kill it, and spend the quarter correcting what our product pages say about pricing. That’s what the answers get wrong today.”“I’d kill it and put the time into something else.”Ask: Into what, and why is that worth more?“The team is keen and it’s specced, so I’d take it.”
Offers a cheap version of the rejected work if it is genuinely near-free, and distinguishes near-free from worth-a-quarter“The file is an afternoon’s work, so publish it. It just isn’t worth the quarter.”“Neither is worth the quarter.”Ask: Does either part cost almost nothing?“The file is quick, so let’s do the whole proposal.”
Says plainly what they are uncertain about, and what would change their mind“I don’t know that anyone reads that file yet. If one of the big assistants says it does, I’d move it up.”“I have some doubts about llms.txt.”Ask: What would settle them?“Nobody reads llms.txt. It’s a waste of time.”

5. Fix forward or roll back

Midway through a Saturday-night network change, 14 of 61 building controllers have no address. Users are unaffected, and the change window closes at 02:00.

The question

“It's 23:41 and the window closes at 02:00. Users are unaffected, 47 controllers have addresses and 14 don't. Fix forward, roll back, or stop and leave it for Monday — and show me the evidence that decides it.”

Fix forward or roll back: each signal, and what Yes, Not sure and No can sound like
SignalYesNot sureNo
Asks for evidence of what is on those ports before changing them“Before I touch the second switch, what’s plugged into those fourteen ports?”“It’s probably the second switch.”Ask: What would you check before changing it?Moves the 14 ports to the controllers’ network to see whether they come up.
Decides to fix forward because the change is inside the approved intent and the abort criteria are untouched“Users are unaffected and none of the abort rules we set before the window has tripped. Fixing those 14 ports is inside what was approved, so I fix forward.”“I think we fix forward.”Ask: What does that rest on?“Something’s not right, so I’d roll the whole change back.”
Records the deviation in the change record, with what was changed and why“It goes in the change record tonight: the port range we missed, what I changed and why.”“I’ll note it.”Ask: Where, and what goes in?Fixes the ports and leaves the change record as approved.

Questions

What are good decision making interview questions?
Ones that give the candidate a real call with something missing, and a note of what a strong answer shows. “Ship v7, kill it, or extend the test” has no safe answer, and its plan says what to listen for: who owns the call, what waiting costs, and the rule that should have existed before the test started.
How do you judge a decision when there is no single right answer?
By how it was made. Most of the signals here reward the reasoning: who decides, what the call rests on, and what would reverse it. Two strong candidates can choose differently and both earn a Yes on those. Where a plan expects one choice, as the network change does, its signal says why.
Can you ask about a past decision instead?
Yes, if you write its signals first. “Tell me about a decision you made with part of the evidence missing” works when you know what to listen for: who owned it, what they knew and did not, and what would have changed their mind. The conflict resolution page has one about a decision the candidate wrote down. A decision people did not want

Start from a plan

Each question links to the plan it comes from, with that plan’s other questions and their signals. In Hirezen you can use a plan for one of your roles, change its questions, and mark its signals during the interview. The free plan covers one open role and up to ten people.