Tell Me About a Time Your Technical Plan Hit Unexpected Obstacles
The question: "Tell me about a time when you had a plan for approaching a technical problem, but encountered unexpected obstacles. How did you adapt your approach and what was the outcome?"
Also asked as: "Tell me about a project where your original technical plan didn't survive." · "Tell me about a time you realized partway through that your approach wasn't going to work." · "Tell me about a time you had to change direction on a plan other people were depending on."
Who's asking: an Engineering Director or Senior Engineering Manager, in a behavioral round.
Executive Summary: The director is deciding whether they'd hear the truth from you early when a plan with a date on it starts to fail. Clever fixes don't answer that. The signal is when you knew, what you gave up, and who you told. As the level rises, the obstacle stops being technical in itself: the plan spans teams, and adapting means re-negotiating what other people committed to.
What the director is really asking
Underneath the story, the director is answering four questions of their own:
- "When did they know?" They want a moment and the evidence for it. "We hit some obstacles along the way" has neither.
- "What did they give up?" If the adaptation cost nothing, nothing was at stake, and the director stops believing the story.
- "Who did they tell, and how fast?" Bad news sat on for two weeks is the risk a manager is screening for.
- "Whose fault is it, in their telling?" If the obstacle is another team's bad data or a PM's changing mind, you've told them how you'd describe them next year.
Pick the right kind of story. Choose one where something real was lost: a promise, a date, a week of your own work. A plan that bends and still lands on time isn't this question.
It maps most closely to Amazon's "Deliver Results" and "Earn Trust"; the mapping is approximate. Three ladder dimensions do most of the work: ownership (how much of the plan was yours to change?), influence (whose commitments did you have to reopen?), and how impact is measured (what did the changed plan deliver, net?).
Level fit at a glance
| SDE | Senior SDE | Staff | Principal | |
|---|---|---|---|---|
| Asked at this level? | Core | Common | Common | Rare |
| Story scope they expect | A plan for a task or feature, held with your lead | A project plan your team owned, touching one adjacent team | A plan across several teams, with dates other managers had committed to | A multi-org bet you got funded, with a number executives had already used |
| The bar in one line | Notice your plan is wrong, tell your lead that day, and finish something real | Re-size it, choose what to give up, and bring your manager and PM a re-plan | Go to each person who committed, own your part, and re-negotiate what they give up | Reprice your own bet before the people who funded it find out, and take yourself out of the decision |
| Instant down-level | "The data was bad, so I worked around it" | "We worked weekends and still hit the date" | "The other teams slipped, so I escalated" | "The teams worked through the obstacles" |
The climb is in who pays. At SDE it's your week and your lead's sprint. By Staff, other managers give up what they promised, and you have to ask them yourself.
SDE: notice it, say so that day, and finish something real
What meets the bar
- You can name the moment you knew and what showed you.
- You can say what you'd assumed. The surprise lives in your plan, not in someone else's work.
- You told your lead quickly, with options, and the plan stayed in their hands.
- You gave something up, and you shipped something real.
- Check the level: if your lead found the problem, or you only fixed a bug, it falls short. If you re-planned your team's quarter and moved other people's work, you may be telling a Senior story.
Model answer
"At a used board-game marketplace, I took a ticket to let sellers import a whole collection from a spreadsheet. Our biggest sellers list 300 games or more, one at a time, and the PM had told them it would be one click. My plan, at two weeks, was to match each row's title to our catalog and publish it. A week in, I found I'd spent that week fixing the wrong thing.
My test on three volunteer sellers' 500 rows left 31 percent unmatched. I assumed typos and messy formatting, so I spent four days on fuzzy matching and got to 88 percent. Then I asked one of those sellers to check their 60 newly matched rows, and they found nine wrong: a German edition matched to the English one, an expansion matched to the base game. A buyer would get the wrong box. So I finally read the failures properly. Spelling was a small part of it. The same title names several products, so even my exact matches could be wrong, and my better match rate was just making wrong answers look right.
Before I went home, I told my lead the plan itself was wrong, not just late. I showed the nine rows and gave two options: three more weeks to add publisher codes, which would only help sellers whose spreadsheets have them, or give up one click: single-edition titles publish immediately, and for the rest the seller picks from a short list. I recommended the second, and my lead agreed. That meant throwing away most of the four days. I told the PM the promise was gone, and drafted the note to sellers.
We shipped in three weeks, one late. In two months, 70 sellers imported about 11,000 listings, and wrong-edition complaints ran about 0.6 percent of orders, the same as hand-typed listings. Sellers still click on about a third of their rows, and two of the biggest told the PM it wasn't what they'd been promised.
Now I read a sample of failures before I tune anything. On my next ticket, rejected seller addresses, reading 30 of them took an hour and showed one country's postcode format was most of the problem."
Why this lands at SDE
- "Before I went home, I told my lead the plan itself was wrong, not just late": the level beat. It's one ticket, the news moves the same day, and the decision stays with the lead.
- "I assumed typos and messy formatting": the wrong turn is theirs, and the catalog isn't the villain.
- "Sellers still click on about a third of their rows": the adaptation cost something, and they say so.
Follow-ups to expect
- "When did you first suspect the plan was wrong, and how long before you told your lead?": the most dangerous one at this level. It tests whether you sit on bad news → the same day for the news, but four days were spent on a guess. Say so.
- "What did giving up one click cost the PM?": tests whether you see other people's costs → the PM went back to sellers with a smaller promise. Say what you did to help.
- "What did you learn that you still use?": tests whether it stuck → the rule, plus the postcode ticket as proof.
Red flags
- ❌ "The catalog data was a mess, so I had to work around it." That's blame, and you never say what you'd assumed.
- ❌ "I stayed late and still shipped on time." Nothing was given up, and your lead never learned there was a risk.
Senior SDE: re-size it, choose the loss, and bring a re-plan
What meets the bar
- You re-sized the plan yourself from new evidence, and you can give the old number and the new one.
- You owned the miss, including any warning you didn't act on.
- You brought your manager and PM a recommendation with its cost, not a problem to solve.
- You took the hard conversation yourself, with the adjacent team or customer it hit.
- Check the level: if your manager did the re-planning, it reads as SDE. If several other managers reopened their commitments because of you, you may have a Staff story.
Model answer
"At a company selling school-bus routing software to about 140 districts, I was tech lead on the route planner, a team of eight. A planning run took about three hours, so planners could try one version a night, and every August they have to publish routes before school opens. In January I committed to a new engine by August, on a 16-week estimate. In week six I found that estimate was off by nearly a factor of three.
I'd sized it from five mid-size districts and 40 routing rules. Our implementation lead had told me in planning that big districts were 'a different product'. I wrote that down as a risk and never checked it. Then our first large district's rules wouldn't fit, so I pulled every district's settings. The 128 smaller ones fit; the 12 largest needed about 300 more: no kindergartner dropped across a busy road, wheelchair buses. Most were variations on about a dozen kinds, so I re-sized it at 45 weeks, and August was week 30. If routes aren't out, kids don't get picked up.
That week I took it to my manager and the PM, with the new number and a split I was recommending: the 128 smaller districts move by August, the 12 largest stay on the old engine until spring, and we run both until then. The PM pushed back, because those 12 were about 40 percent of revenue. So I agreed to two more things: I'd pull two engineers off driver-delay alerts to bring spring forward to January, and I'd go with the implementation lead to the two districts we'd promised, and tell them myself. One was angry. Both renewed, the angry one for one year, not three.
We got the 128 live on time. Runs dropped from three hours to about 40 minutes, planners tried around four versions, and empty bus miles fell about 4 percent. The 12 largest moved in January, five months after the date I'd given. I now size from the largest customer, and on our next migration that doubled the estimate before we committed. The price of my estimate was about one engineer running both engines until January, and driver-delay alerts, the feature parents asked for most, sliding from fall to the next spring."
Why this lands at Senior SDE
- "with the new number and a split I was recommending": the level beat. They own the project's outcome, and the re-plan arrives with its cost, inside one team and one neighbor.
- "I wrote that down as a risk and never checked it": the warning was given, and the miss is theirs.
- "The price of my estimate was about one engineer running both engines until January": the result is net, and they name who paid.
Follow-ups to expect
- "Your implementation lead warned you. Why didn't you check?": the most dangerous one at this level. It tests self-awareness → no excuse. You had a date you wanted, and a bigger number was unwelcome. Say what you do with warnings now.
- "Why split it, and not slip everything or pull more people?": tests judgment → what each option cost, why August ruled out the first, and why two engineers was the limit.
- "Where did the PM push back, and were they right?": tests how you treat a partner → they had a point on revenue, and it's why you added the engineers and the district visits. Those districts' first August on it came a year late; say so.
Red flags
- ❌ "The requirements turned out to be far more complex than anyone knew." Someone did know, and it reads as blame in the passive.
- ❌ "We worked weekends and hit the date." It's heroics, and your manager never got to choose.
Quick check before moving on:
Staff: go to the people who committed, and re-negotiate what they give up
What meets the bar
- The plan crossed teams, and you can say why fixing it fell to you and what you stopped doing.
- You knew which of your own calls broke it, and said so to the people it cost.
- You went to each manager yourself, knowing what they'd promised and what you were asking them to drop.
- Someone said no, and the plan changed because of it.
- Check the level: if the teams all report to your manager, or the slip was your own team's project, it reads as Senior. If executives changed what the company funds, and you can price that in business terms, it's moving toward Principal.
Model answer
"At a company running baggage tracking for about 30 airports, transfer bags were missing connections at about 11 per 1,000, while all four teams, in three reporting lines, showed healthy numbers. Bags got lost between their handoffs. I led sorter integrations, and my director gave me half my time to plan one shared bag record over 18 months, so I handed off my roadmap. All four managers signed up to dates. Five months in, I knew my order was wrong, and two of them would pay for it.
I'd put passenger alerts first, because two airlines were paying for them, and handheld scans last. The handheld manager had asked in planning whether alerts should wait for them, and I'd said the money came first. These days I test a skeptic's question before I overrule it. An airline's operations lead then sent me 40 missed-bag alerts for bags that had made it. One in nine was false, because loaders' handhelds upload in batches, often after the plane has gone.
That week I took my director a new order, priced by team, then went to the managers myself. I told the alerts manager the order was my mistake and asked them to pause the alerts and give up a date they'd promised two airlines. They agreed if I told both airlines myself. One stayed, and one left the pilot. I asked the handheld manager to start six months early and drop an offline feature. They refused to touch devices before the summer peak, so we moved to autumn, and my own manager lent two engineers, pushing our sorter work back a quarter.
The four teams took 24 months. At the 12 airports on the shared record, misses fell from about 11 to 8 per 1,000: 27,000 fewer bags a year, saving the airlines about $2.7 million at roughly $100 each. Our return was about $900,000 a year in renewal price, against a roughly $4 million program. False alerts fell to one in 200. I got the managers to publish how late each team's data arrives, and the team that auto-rebooks missed bags now acts only on fresh scans. The other 18 are still near 11, because their handlers own their devices and nobody has budgeted new ones."
Why this lands at Staff
- "I told the alerts manager the order was my mistake and asked them to pause the alerts and give up a date they'd promised two airlines": the level beat. Adapting here is a conversation with someone they have no authority over, about a promise already made.
- "They refused to touch devices before the summer peak, so we moved to autumn": a real no, and the plan moved to fit it.
- "the team that auto-rebooks missed bags now acts only on fresh scans": a team outside the plan changed what it built.
Follow-ups to expect
- "Two managers paid for your ordering mistake. Would they work with you again?": the most dangerous one at this level. It tests trust → say what each actually said to you. Don't claim the relationship came out stronger.
- "The handheld manager raised this in planning. Why did you overrule them?": tests judgment → the funding argument was real, but you never tested it. Say what you ask a skeptical manager now, and whether the 18 were ever in the plan.
- "Why was this your problem to solve?": tests ownership → the losses sat between the teams' handoffs, your director funded the time, and you handed off your roadmap.
- "How did you get them to reopen what they'd committed to?": tests influence → what each one would lose, what you offered, and where you gave way.
Red flags
- ❌ "The other teams couldn't hit their dates, so I escalated and we re-planned." The action is escalation, and they're the villains.
- ❌ "My team's project slipped, so I re-planned it." However large, it's a Senior story.
Principal: reprice your own bet, and take yourself out of the decision
What meets the bar
- Why it's rare: directors usually get this signal from strategy and failed-bet questions. When it comes up, they're checking whether you'd reprice your own bet before the people who funded it find out. This short answer shows the repricing more than the roadmap, so be ready to say what the org did next.
- You told executives the number was wrong, including a number they'd already used.
- You can name how you were part of the problem, and the mechanism the org kept, with one real use.
- Check the level: if several teams changed plans but no executive heard a new price, it reads as Staff.
Model answer
"Our telehealth company paid a vendor about $14 million a year for video visits. I got a three-year, $18 million bet funded to replace it, with $9 million a year of savings in the board plan. In month 14, visits that dropped before the doctor finished ran at 9 percent in our first two states, against 3 percent on the vendor, mostly older patients on weak connections. That was a care problem first.
I'd made it worse. I'd insisted on approving each state's cutover personally, so clinical operations, engineering and finance all waited on me, and the numbers, held for my review, reached the VP of clinical operations six weeks late. That VP told me so.
That month I went to the CTO and CFO with a new plan and a $4 million savings floor: pause at two states, and move weak connections, a quarter of visits, back to the vendor. Then I wrote a cutover rule with that VP, who took over every decision: under 3 percent dropped across a state's visits, vendor included, for four weeks, or the state doesn't move. That VP held three states back on it the next year.
We finished a year late, at about $24 million. Overall drops are back under 3 percent. Savings are about $5 million a year, not nine, so payback went from two years to nearly five. The CFO, who had to take that to the board, told me the fallback goes in the first number next time."
Why this lands at Principal
- "That month I went to the CTO and CFO with a new plan and a $4 million savings floor": the level beat. They reopen their own funded bet with executives, and the price goes with it.
- "I'd insisted on approving each state's cutover personally": they name how they slowed the bad news down, at a level where people rarely do.
- "That VP held three states back on it": the mechanism has an owner and a real use.
- "so payback went from two years to nearly five": the overrun is in the sum, so it survives a second question from a CFO.
Follow-ups to expect
- "If that VP hadn't told you, how long would you have kept approving those cutovers?": the most dangerous one at this level. It tests self-awareness → admit you might have gone on for months, and say what the rule does that you didn't. The six weeks was yours; name it.
- "Why not kill the bet in month 14?": tests judgment → what cancelling would have cost, against a smaller saving, and what would have made you stop.
- "How did you bring the CFO along on a number the board had seen?": tests executive trust → you went that month, with the new number and a floor.
Red flags
- ❌ "We hit obstacles in the migration, and the teams worked through them." There's no bet and no price, so where are you in it?
- ❌ A result in engineering terms only. Without patients and dollars, it reads as Staff.
Quick check before moving on:
One story, four altitudes
| SDE | Senior SDE | Staff | Principal | |
|---|---|---|---|---|
| Situation | A spreadsheet import, one ticket | A new routing engine, a project the team owned | One bag record across four teams | Replacing a video vendor, a three-year bet |
| Your role | SDE on the ticket; the lead holds the plan | Tech lead; owns the project's estimate | Leads one team; no authority over the rest | Got the bet funded; accountable for it |
| The key move | Told the lead the same day, with two options | Re-sized it and brought a split with its cost | Went to each manager and reopened their promises | Took a lower number to the CFO and handed cutovers to the VP |
| Result | A week late, one click gone, complaints level with hand-typed | 128 districts on time, 12 five months late, alerts slid | Misses 11 to 8 per 1,000, 24 months, one airline lost | $5M a year not $9M, $24M not $18M, a year late |
| Reflection | Read failures before tuning anything | Size from the largest customer | Test a skeptic's question before overruling it | The fallback goes in the first number |
Draft your answer
Step 1: Pick the right story
Run each candidate story through the test for your target level:
- SDE: Did a plan you made for a task turn out wrong, and did you tell your lead before they found out?
- Senior SDE: Did you re-size a project your team owned, choose what to give up, and bring the re-plan to your manager and PM?
- Staff: Did at least two managers you don't report to reopen something they'd committed to, because you went and asked?
- Principal: Did you take a new price on your own multi-org bet to executives, and did the org keep a mechanism that doesn't need you?
Then check all of these:
- It ended in the last three years, though a Staff or Principal plan may have started earlier.
- You can name the moment you knew, something you gave up and someone you told.
- The obstacle includes something you assumed, not only something others did.
- There's a number for the result, net of what the change cost.
If you're a new grad. A hackathon, a sponsored team project or an internship works if something real was lost, such as a demo feature, and the wrong assumption was yours. A small loss is fine. Skipping the moment you knew isn't: say it, then what you gave up and how soon you told your team. "Twelve hours into a hackathon, I found the transit feed I'd promised my team was live only updated hourly."
Aiming a level above your current title? Tell your team's project as your team's project, and use the ending to show you can see further: "If I ran that again, I'd ask which other teams had built dates on my estimate." An honest Senior story with that awareness beats an inflated one that falls apart at the first follow-up.
Step 2: Fill the skeleton
SCOPE + PLAN (15s) — Where, how big, what you'd promised and to whom: ______
THE BREAK (15s) — The moment you knew, what showed you, and what it meant
for customers or the team, in words a non-engineer
follows: ______
YOUR PART (15s) — What you'd assumed, or the warning you passed over: ______
WHO YOU TOLD (15s) — Who, how soon, and what options you took them: ______
WHAT YOU GAVE (20s) — What you gave up, and who else paid: ______
LEVEL BEAT (20s) — The line only your level could say: ______
SDE: what you told your lead, that day, and what they decided (merge with
WHO YOU TOLD).
Senior: the re-sized plan you recommended, and its cost.
Staff: which managers you went to, what they gave up, and who said no.
Principal: the new price executives heard, and the mechanism the org kept.
RESULT (20s) — One number, net of the cost, and what it left unfixed: ______
REFLECTION (10s) — What you do now, with one later use, or what's still open: ______
Aim for a little over two minutes, with the break in your first 30 seconds and a 30-second cut ready (the plan, the break, what you gave up, the number).
Step 3: Calibrate it
| If your answer says… | It reads as… | Fix |
|---|---|---|
| "We hit some obstacles along the way" | No moment of knowing | Date it, and name the evidence |
| "We adapted and still hit the date" | Nothing was at stake | What you gave up, and who paid |
| "The other team's data turned out to be bad" | Blame, down a level anywhere | What you assumed and didn't check |
| A Staff answer: "My team's project slipped and I re-planned it" | A big Senior story | Whose commitments you reopened, and how |
| A Principal answer: "The teams worked through it" | Staff or below, with no bet | The new price, who heard it, and the mechanism |
Step 4: Rehearse the follow-ups
Prepare these five, and say each out loud once:
- "When did you first know the plan was failing, and how long before you told anyone?" The most dangerous one. Give the real gap, and its cost.
- "What did you give up?" Name something you wanted to keep, and who else lost by it.
- "Who saw this coming before you did?" Give them credit, and say why you didn't listen or ask.
- "What did the person who paid for this say to you?" Give their real reaction, not the flattering version.
- "How would you plan it differently?" Give a check you could have run in week one, and where you've used it since.
Common mistakes
- Making the obstacle someone else's fault. Find the assumption you made about their work.
- Describing the fix at length. Other rounds grade the design. Give it one clause, and spend your time on the decision and who it reached.
Key takeaways
- It tests what you do when the plan meets reality: when you knew, what you gave up, and who you told.
- An adaptation with no cost isn't a story. Neither is an obstacle that's all someone else's fault.
- The level shows in who pays, and who you have to ask.
Test yourself on this question's key ideas: