Tell Me About a Time You Made a Mistake at Work
The question: "Tell me about a time when you made a mistake at work. How did you handle the situation and what did you learn from it?"
Also asked as: "What's the biggest mistake you've made in your career?" · "Tell me about a decision you got wrong." · "When did something go badly because of something you did?" · "Tell me about a bet you made that didn't pay off."
Who's asking: an Engineering Director or Senior Engineering Manager, in a behavioral round.
Executive Summary: The director already assumes you make mistakes. They're deciding whether they'd hear about the next one from you first, whether the damage would stay contained, and whether it would happen twice. What changes with level is the size of what you got wrong and of what you changed: one bug and a habit, one project decision and a team process, a cross-team bet and its unwinding, then a direction and a mechanism the org keeps.
What the director is really asking
The director is answering three questions:
- "Can they say it plainly?" They listen for a sentence that starts with "I" and names what you did.
- "How long between knowing and telling?" The mistake itself is usually forgivable. The stretch where you knew, or should have, and said nothing is what they worry about.
- "What's different now?" "I learned to be more careful" changes nothing they can rely on. They want a behavior, a process or a rule, and proof it has been used.
Pick the right kind of story. The question says you made a mistake, so it has to be your own act or decision, one you'd take back. If a project went wrong around you, "Describe a time when a project didn't go as planned" shows that shape. And it has to be real: if nothing happened to anybody, the director assumes you're hiding the real one.
It maps most closely to Amazon's "Ownership" and to "Earn Trust", for its line about being vocally self-critical; both approximate. Three ladder dimensions do most of the work: the failure story row, ownership, and self-awareness and coachability.
Level fit at a glance
| SDE | Senior SDE | Staff | Principal | |
|---|---|---|---|---|
| Asked at this level? | Core | Core | Core | Common |
| Story scope they expect | A bug or miss you caused, in your own feature | A decision you made on a project you led, which missed | A cross-team bet you pushed, which didn't pan out | A direction you set, with an org or business cost |
| The bar in one line | Say it was yours, tell someone fast, fix it, change one habit | Own the decision, repair it with those it hit, change a process for your team | State what it cost several teams, recommend stopping if that's right, unwind it with their managers | Name the direction as wrong, price it in business terms, build a mechanism the org keeps |
| Instant down-level | "The requirements weren't clear" | "We shipped a bug" | "My project slipped and a neighboring team had to wait" | "The market moved, so the strategy stopped making sense" |
At Principal it's Common because directors often fold it into strategy questions. The Staff trap is a Senior story with bigger numbers.
SDE: say it was yours, then change one habit
What meets the bar
- The bug or miss is yours, in your own feature, and you say so in a plain sentence.
- You told someone quickly, including anyone you'd sent the wrong way.
- You counted the damage and did the fix: how many users, what it cost them, and what the fix gave up.
- One habit changed, and you can say when you last used it.
- Level check: if it never reached anyone, it's too small to believe. If you need a team process change to sound serious, you're reaching for Senior. One honest habit is the bar.
Model answer
"I shipped a retry that billed drivers for charges the app said had failed, then blamed firmware for a day and a half. I worked on the driver app at an EV-charging network. Drivers complained that starting a charge sometimes took two tries, so I added an automatic retry: if the charger hadn't answered in ten seconds, the app sent the start command again. It shipped on a Tuesday.
By Thursday, support had about 40 tickets from drivers who'd been told 'charging failed' and were then billed for a session. The reports clustered on one newer charger model, so I decided it was firmware. I pulled logs and wrote it up for the engineer who works with the charger maker. She read it and asked why those chargers were getting two start commands.
That was my retry. That model sits mostly in parking garages, where a start can take 15 seconds. The first command had worked, my retry hit a busy charger, and the app told the driver it had failed. I'd added a retry to one call without asking what happens if the first try actually worked.
I told my lead within the hour that it was my change and that I'd sent us after the wrong cause, and I turned the retry off with its flag that afternoon. Then I pulled the numbers for support: 214 drivers had seen the false failure, about half of them after I'd started on firmware. Sixty-three had paid for a session they thought had failed, some with idle fees because they'd gone to find another charger. Support refunded about $1,900 and contacted those drivers from my list.
The real fix took me three days: the app sends one start command, then checks the session status for up to 30 seconds before it says anything. False failures in garages went from about 3 percent of starts, before my retry, to under 0.2. The cost is that a driver at a truly dead charger now waits 30 seconds to find out, not ten.
Before I add a retry now, I write in the pull request what happens if the first call worked. Last month that changed a payment retry I was writing."
Why this lands at SDE
- "I'd added a retry to one call without asking what happens if the first try actually worked": the level beat. One call in one feature, and they can say exactly what they skipped.
- "I told my lead within the hour that it was my change and that I'd sent us after the wrong cause": they admit both mistakes, the bug and the firmware theory.
- "about half of them after I'd started on firmware": they counted what the wrong theory cost.
Follow-ups to expect
- "How did you decide when to ask for help?": tests escalation → you worked a theory alone for a day and a half, so say what would send you to someone sooner now.
- "What did you say to the engineer who caught it?": tests how you treat people → what you actually said to her, and how you write those reports now.
- "What did you learn that you still use?": tests what stuck → the pull-request line, and what it changed in that payment retry.
Red flags
- ❌ "The review should have caught it." Blame, however politely it's put.
- ❌ "I learned to be more careful." That isn't a change anyone can see.
- ❌ "We caught it before it reached anyone." Too clean, so the director will ask for another.
Senior SDE: own the decision, then change how your team works
What meets the bar
- It's a decision you made on a project you led, with an option you turned down.
- You repaired it with the people it hit, and told them it was your call.
- You changed a process for your team, and someone other than you has used it.
- There's a system or product number, net of what the fix cost.
- Level check: a slip anyone writing the code could have made reads as SDE. If other teams' roadmaps moved on your word, it's becoming a Staff story, and you'd need to show their managers.
Model answer
"At our university's IT group, I led the team of five rebuilding the waitlist in course registration, for about 30,000 students. The old waitlist skipped about 2 percent of seat offers, so students who'd waited in line never heard. The decision I got wrong was the order.
The deputy registrar asked us to cut over after November registration. I argued for three weeks before it, because waiting meant another term of skipped students, and I put our load test, a replay of real opening-morning traffic, after the cutover, because it needed the new system's data in place. My manager backed me on my reasoning.
At 7 a.m. on seniors' opening day, about 6,000 students arrived in ten minutes. Registration stayed up, but the waitlist fell behind. Offers took up to 40 minutes, and by then other students had taken the seats directly. About 1,100 students saw wrong positions, and in nine sections we'd offered seats that no longer existed.
I froze waitlist offers by 7:45 and left registration running, because rolling back mid-morning would have lost enrollments. I called the deputy registrar myself, told her the date was my call, and had a list of affected students to her by noon. Two of my engineers spent three days working through it with her staff. Those nine sections went 17 seats over capacity, and four of her people lost most of a week.
Then we built the replay I'd put second. It took two engineers three weeks and pushed our roadmap back five weeks, and it found the lag and one more fault in its first day. At April's registration, offers stayed under two minutes at peak, and skipped offers went from 2 percent to about 0.1.
I changed how my team schedules cutovers: nothing goes live until we've replayed last term's opening morning against it, nothing goes live within six weeks of a registration window, and any engineer on the team can hold the date if the replay fails. That autumn, our newest engineer held a prerequisite-check change for a week on that rule. The deputy registrar now asks to see the replay results before she'll agree a date, and she should."
Why this lands at Senior SDE
- "I changed how my team schedules cutovers": the level beat. It's a process for the team they lead, with no reach beyond it.
- "The deputy registrar asked us to cut over after November registration. I argued for three weeks before it": a real decision, with the option they turned down and the person who was right.
- "told her the date was my call": they own it to the person it hit.
- "our newest engineer held a prerequisite-check change for a week": the process works without the candidate.
Follow-ups to expect
- "The registrar asked you to wait. What do you do now when someone who knows the calendar says that?": tests coachability → what you now ask before arguing past them, without turning it into "I always defer".
- "How did you get the team aligned on the new rule?": tests how you lead → who thought it was too slow, and what it has delayed since.
- "Who did you grow on this?": tests people growth → the engineer who held the date, and what you said when they did.
Red flags
- ❌ "The load was higher than anyone expected." It treats your decision as weather.
- ❌ "I added a checklist." Who has used it, and what has it stopped?
Quick check before moving on:
Staff: own the bet, say what it cost, and unwind it
What meets the bar
- It's a bet you pushed across several teams, whose managers gave up something on your word.
- You can say what it cost each team, in their terms.
- You led the unwinding: you re-priced it, made a recommendation, let each manager choose, and kept what earned its place.
- There's an org-level number, net, and something that wasn't recovered.
- Level check: if only your team and one neighbor changed plans, it reads as Senior. A business price and a mechanism the whole org keeps read as Principal.
Model answer
"At our hospital network, five product teams each read lab results from their own copy of the lab feed. When a lab corrected a result, the patient portal showed it in minutes, and the clinician mobile app could take six hours. After reviewing a near miss on a corrected potassium value, with nobody owning that gap, I proposed one shared results service for all five, and I got their managers to give up roadmap for it. My number was 60 engineer-weeks, taken from a chemistry pilot at one hospital.
That number was badly wrong. The pilot covered about 300 test codes. We had 14 labs and about 9,000 local codes, and lab directors had to sign off every mapping, because a wrong unit on a result can hurt someone. I'd never priced that. By month six we'd passed my 60, the third lab was four times over plan, and I told myself the rate would improve. By month nine, the teams had spent about 150 engineer-weeks, only chemistry had moved, on two apps, and the ICU team had pushed a sepsis alert back a quarter.
I redid the sum: about 400 more to finish. I took that to each of the five managers separately, with my original number beside it and what finishing would cost their team, and I recommended we stop. I told my director I should have done it three months earlier.
What could be saved: a stream announcing each corrected result, which addressed the near miss. Each manager took it onto their own feed, about 20 engineer-weeks in all. The ICU's manager took her people back that month. Portal's manager had already retired his parser, so he kept chemistry on the service, and his team took it over. Mobile spent five engineer-weeks moving back.
Corrected results now reach all five apps within about five minutes. That cost about 175 engineer-weeks, where the narrow version alone might have cost 30. I now price a shared bet from its strangest site, with the people paying for it, and we agree a stop number first. My next one stopped at its number. The gap stayed open about a year, the sepsis alert shipped a quarter late, and nobody can tell me what either meant on the wards."
Why this lands at Staff
- "I took that to each of the five managers separately, with my original number beside it and what finishing would cost their team, and I recommended we stop": the level beat. Five managers, no authority, and the candidate argues against their own bet in each manager's terms.
- "Portal's manager had already retired his parser, so he kept chemistry on the service": the managers choose for their own reasons, so the unwinding isn't imposed.
- "That cost about 175 engineer-weeks, where the narrow version alone might have cost 30": an org-level price, stated against themselves.
Follow-ups to expect
- "You saw the overrun in month six and told them in month nine. What was happening for those three months?": the most dangerous one → it's built to make you defend the delay. Don't. You wanted the rate to improve, and it cost about 50 engineer-weeks across five teams. The stop number exists because of it.
- "Those managers gave up roadmap on your number. Why would they believe your next one?": tests trust → what you do differently when you estimate, and what happened the next time you asked one of them for capacity.
- "Why was this your problem to solve?": tests ownership → you'd seen the near miss, nobody owned the gap, and say what you stopped doing for it.
- "What would have happened without you?": tests honesty → the gap might have stayed longer, and the five teams wouldn't have spent those weeks. Say both.
Red flags
- ❌ "The other teams didn't prioritize it." It was your number, and they gave up what you asked.
- ❌ "We pivoted to a narrower scope." It relabels a failure as a plan, with the cost removed.
Principal: name the direction as wrong, and build the mechanism
What meets the bar
- It's a direction you set and were accountable for, across organizations, with executives funding it on your case.
- The cost is in business terms: money, payback, revenue or risk.
- You told the executives yourself, including what you'd been telling them until then.
- The org kept a mechanism, which has since stopped or changed something, and you know what it doesn't catch.
- Level check: several teams and their managers, with no executive funding the bet on your case, reads as Staff.
Model answer
"At our credit-card issuer, I made the case for moving card authorization off our third-party processor onto our own platform, across four card lines. I took it to the CTO and CFO: $60 million over three years, then $24 million a year saved in processor fees. The engineering held up, and all four general managers said yes.
My mistake was reading that yes as capacity. Each line had to rebuild its products on the new platform, so each GM had to give up launches. Their bonuses ran on this year's new accounts and purchase volume. The savings landed in a central technology budget, years out. Every quarter a launch won, and that was the rational call for them. I read each deferral as a scheduling problem, and I kept telling the CTO the lines were aligned.
At month 20, we'd spent $41 million, moved 11 percent of volume, and were paying about $8 million a year to run both systems. I told the CTO and CFO the plan's central assumption was mine and wrong, and I showed them what each GM was paid on beside what I'd asked them to give up. I proposed moving the savings into the plans of lines giving up capacity. Consumer and small business took that deal. The co-brand and commercial GMs declined, because their contracts reward launch dates and they'd rather keep paying the processor. Those were fair calls, so I recommended we stop at two lines.
We moved 55 percent of volume, a year late, for $74 million including dual running. A renegotiated processor contract for the other two lines cut the cost of keeping both systems to about $4 million a year, so we save about $9 million net, not $24 million. Payback from go-live went from under three years to about eight.
I wrote a rule with the CFO's office: a platform bet over $5 million doesn't get funded until the leader giving up the capacity carries the savings in their own annual plan, and they check that at funding. It has stopped three proposals in two years, one of them mine. It has a hole I haven't closed: resilience work nobody can book as savings can't pass it, so that's still funded by argument at the CTO's table."
Why this lands at Principal
- "I wrote a rule with the CFO's office: a platform bet over $5 million doesn't get funded until the leader giving up the capacity carries the savings in their own annual plan": the level beat. The mechanism changes how the company funds work, and it outlives this program.
- "My mistake was reading that yes as capacity": the misjudgment is about how the org works, and "the rational call for them" leaves no villain.
- "Payback from go-live went from under three years to about eight": the price in business terms, beside the original promise. Have the first-dollar figure ready too: about 12 years against a planned 5.5.
- "one of them mine" and "It has a hole I haven't closed": they apply the rule to themselves and know where it stops.
Follow-ups to expect
- "You told the CTO the lines were aligned for 20 months. What did that do to your standing?": the most dangerous one at this level → say what it cost you, and how the CTO reads your numbers now. Don't claim it strengthened the relationship.
- "How did you bring the CFO along?": tests exec influence → what you showed, and why moving savings between budgets suited the CFO too.
- "What mechanism outlived your involvement?": tests lasting impact → the rule, who checks it, what it has stopped, and the hole.
Red flags
- ❌ "The market moved, so the strategy stopped making sense." Where's your part?
- ❌ "The GMs wouldn't commit." You called them aligned. Blaming them is the same misread again.
Quick check before moving on:
One story, four altitudes
| SDE | Senior SDE | Staff | Principal | |
|---|---|---|---|---|
| Situation | A retry on one app call, never asked what if the first try worked | A cutover date set before the replay | A shared service for five teams, priced from one lab | Leaving a processor, reading four GMs' yes as capacity |
| Your role | Wrote it and owned the bug | Led the team, and argued past the registrar's ask | Pushed the bet, with no authority over the teams | Made the case to the CTO and CFO |
| The key move | Told the lead within the hour, counted the drivers, fixed it | Froze offers, called the registrar, built the replay | Re-priced it for each manager and recommended stopping | Told the CTO and CFO, and moved savings into GMs' plans |
| Result | $1,900 refunded; false failures 3% to under 0.2% | 17 seats over; offers under two minutes; skipped offers 2% to 0.1% | 175 engineer-weeks; corrections in five minutes; alert a quarter late | $74M; about $9M a year net; payback about eight years from go-live |
| Reflection | Write down what happens if the first call worked | A replay and a window rule any engineer can enforce | Price from the strangest site, and agree a stop number | A funding rule the CFO's office checks |
Draft your answer
Step 1: Pick the right story
Run each candidate story through the test for your target level:
- SDE: Did a bug or miss that you caused reach someone, did you fix it, and did one of your habits change?
- Senior SDE: Did a decision you made on a project you led miss, with an option you'd turned down, and did your team's process change?
- Staff: Did several teams give up something on your word, did the bet fail, and did you lead the unwinding with their managers?
- Principal: Was it a direction you set, with an org or business cost, and did the org keep a mechanism because of it?
Then check that it passes all of these:
- It's from the last few years, and you'd take it back.
- You can say what you did in one sentence starting with "I".
- Someone paid a cost, and you can put a number on it.
- You can say when you first suspected it, and when you first told someone.
If you're a new grad. A bug from an internship, a part-time job or an open-source pull request counts, if it was yours and it reached someone: users, students or a release. A director expects a smaller blast radius from you, not a smaller admission. They still want when you told someone, the fix, and the habit that changed. "At my campus library job, an overdue-notice script I changed emailed 1,200 students fines they didn't owe."
Aiming a level above your current title? A Senior aiming at Staff should tell a two-team story as two teams, and use the ending to show what they can see: "If I ran that again, I'd ask who else was planning around my date, because I only checked with one team." That beats a bigger story that falls apart at "What did it cost each team?"
Step 2: Fill the skeleton
SITUATION (15s) — The work, your role, what was at stake: ______
THE MISTAKE (15s) — One sentence starting "I", naming what you did or
decided, plus any option you turned down: ______
THE COST (20s) — Who it hit, how many, how much: ______
HANDLING (45s) — When you knew, who you told, and how you repaired it:
2–3 moves, each "I …, because …": ______
LEVEL BEAT (inside HANDLING) — the line only your level could say: ______
SDE: exactly what you skipped, in your own code, and what you did next.
Senior: the decision, and the process you changed for your team.
Staff: what you took to which managers, and what each one chose.
Principal: what you told which executives, and what the org kept.
RESULT (15s) — The number, net of what the fix cost, and anything
that wasn't recovered: ______
WHAT YOU LEARNED (10s) — What you do now, and when it last mattered: ______
Aim for about two minutes, two and a half at most, with a 30-second cut ready (the mistake, the cost, who you told, what changed).
Step 3: Calibrate it
| If your answer says… | It reads as… | Fix |
|---|---|---|
| "I care too much" or "I worked too hard" | A humble-brag | A mistake that cost someone something |
| A Senior answer: "I shipped a bug" | SDE | The decision you made, and what you turned down |
| A Staff answer: "My project slipped and a neighbor waited" | Senior | Whose roadmap moved on your word, and how you unwound it |
| A Principal answer: "It cost 300 engineer-weeks" | Staff | The business price, and the mechanism |
Step 4: Rehearse the follow-ups
Prepare these five, and say each out loud once:
- "When did you first suspect it, and when did you tell anyone?" The most dangerous one. Give both times, say why you waited, and don't defend it.
- "How did your manager see it?" What they said, and what they watched more closely afterward.
- "What did it cost, and who paid?" The people outside your team, by role, with a number.
- "What do you do differently, and when did that last matter?" One recent example, or an honest "it hasn't come up yet".
- "Is that really your biggest mistake?" Say why this one, and have another ready that cost more.
Common mistakes
- Explaining before admitting. Context first sounds like a defense. Say the mistake early, then explain it.
- Spending the time on the rescue. A long recovery turns it into heroics.
- Ending on a win that cancels it. "It was the best thing that could have happened" erases the cost. Leave the cost standing.
Key takeaways
- It tests whether you'd hear about the next mistake from them first, and whether it would happen twice.
- The level shows in the size of what you got wrong, and of what you changed.
Test yourself on this question's key ideas: