Technical leadership

Question 1

You lead the data platform team. A legacy orders table is read by about forty dashboards, and you want to retire it in favor of a new modeled table. What do you do first?

Accepted Answer

Pull query logs to list who actually reads the table and how, then contact those owners.

Question 2

Query logs show that 12 of the 40 dashboards on a table you are retiring have not been opened in six months. How do you treat them?

Accepted Answer

Tell their owners they are scheduled for archive with the table unless someone claims them by a date.

Question 3

Two months into retiring an old events table, one analytics team has not moved a single dashboard. Their manager says they have no time this quarter. What is your best next step?

Accepted Answer

Offer to pair on their top dashboards and agree a firm date for the rest.

Question 4

You plan to retire a widely used table. Which signal best tells you the old table is safe to turn off?

Accepted Answer

No reads from people or jobs for two weeks after a planned, announced brownout.

Question 5

Before turning off an old customer table for good, you want to find hidden readers. What do you propose?

Accepted Answer

Revoke read access for a few announced hours and watch who reports breakage.

Question 6

A finance director asks you to delay retiring a table because a quarterly board deck depends on it, and the deck is due in three weeks. The migration is otherwise on track. What do you do?

Accepted Answer

Grant a dated exception for that deck and move it after the board meeting.

Question 7

You are writing the plan to retire a heavily used table. What belongs in the announcement to dashboard owners?

Accepted Answer

The reason, the replacement, a column mapping, the dates, and where to get help.

Question 8

During a table retirement, a senior analyst says the new table drops a column her churn model needs. Nobody had flagged it. What do you do?

Accepted Answer

Confirm the need, then add the column or a mapped field before her deadline.

Question 9

Product teams keep renaming event fields, which breaks your pipelines every few weeks. You have no authority over them. What is the most effective first move?

Accepted Answer

Show their leads each breakage and what it cost, then propose a small first contract.

Question 10

One upstream team agrees to try a data contract for checkout events. How do you scope the first version?

Accepted Answer

A few critical fields with types, meaning, an owner, and how changes are announced.

Question 11

An upstream team signed a contract for its events, then shipped a breaking change without notice. Pipelines failed overnight. What do you do the next morning?

Accepted Answer

Fix forward, then meet their lead to see why notice failed and add a schema check to CI.

Question 12

A product engineering manager says data contracts will slow her team down. What argument is most likely to persuade her?

Accepted Answer

Show how often her team is pulled in to debug breaks a contract would prevent.

Question 13

Who should own a data contract for a mobile app's purchase events?

Accepted Answer

The producing app team owns it, with the data team as reviewer and consumer.

Question 14

Three months after rolling out contracts with two teams, you want more teams to join. What is the best way to spread adoption?

Accepted Answer

Publish the drop in breakages and on-call pages, and offer templates and help to the next teams.

Question 15

An upstream team wants to change a contracted field from dollars to cents. Six downstream jobs read it. What process do you push for?

Accepted Answer

Add a new field beside the old one, give a dated window, then retire the old.

Question 16

You are asked to present data contracts to product leadership. You have ten minutes. What do you lead with?

Accepted Answer

Last quarter's data breaks, what they cost, and the small ask to prevent them.

Question 17

Marketing and finance report different monthly active user counts to the same executive. Both pull from your warehouse. What do you do first?

Accepted Answer

Trace both queries side by side to find where the definitions diverge.

Question 18

You find that two teams count active users differently: one counts any login, the other counts a completed action. Both are reasonable. How do you resolve it?

Accepted Answer

Have the metric's business owner pick one, record it, and rename the other.

Question 19

Nobody owns the definition of net revenue. Three dashboards show three values. Who should define it?

Accepted Answer

A finance leader accountable for the number, with data engineering implementing it.

Question 20

After a metric definition is agreed, how do you stop the two versions from drifting apart again?

Accepted Answer

Put the definition in one shared, tested model that every dashboard reads from.

Question 21

An executive asks you, in a meeting, which of two conflicting revenue numbers is right. You have not looked into it yet. What do you say?

Accepted Answer

That you do not know yet, and will bring the cause and a recommendation by a set day.

Question 22

You change the agreed definition of a churn metric, which shifts it down by two points. What else must happen?

Accepted Answer

Announce it with its effect, restate history on the new basis, and date it.

Question 23

Two product teams each want the same metric defined their own way and both ask you to make theirs official. What do you do?

Accepted Answer

Lay out the trade-offs in a short doc and ask the shared business owner to decide by a date.

Question 24

A board deck shows a conversion rate that differs from the product dashboard by 0.4 points. The cause is a known, documented filter difference. What is the right move?

Accepted Answer

Add a clear label to both explaining the filter, and tell the deck owner before the meeting.

Question 25

The finance team flags that warehouse spend jumped 60 percent this month. You lead data engineering. What do you do first?

Accepted Answer

Break down spend by job, user, and table to find what changed before acting.

Question 26

You trace a warehouse cost spike to a new dashboard that a product team set to refresh every five minutes over two years of data. What do you do?

Accepted Answer

Talk to the owner about the real freshness need and help them switch to an incremental model.

Question 27

Warehouse costs keep growing about 8 percent a month and no single job is to blame. How do you make lasting progress?

Accepted Answer

Give each team a view of its own spend and agree on budgets with alerts per team.

Question 28

You want leadership to fund a quarter of work to cut warehouse costs. How do you frame the request?

Accepted Answer

Show the spend trend, expected monthly savings, and how fast the work pays back.

Question 29

During a cost spike, an engineer on your team proposes rewriting the whole transformation layer in a new framework to fix spending. What is your response?

Accepted Answer

Ask for the few biggest cost drivers to be fixed first and measure the effect.

Question 30

A cost spike came from a backfill that an engineer ran on the largest compute size by mistake. How do you handle it?

Accepted Answer

Treat it as a gap in safeguards and add size limits and alerts for large jobs.

Question 31

Your manager asks you to cut warehouse spend by 30 percent in one month. Your analysis shows 15 percent is realistic without hurting key reports. What do you do?

Accepted Answer

Show your manager the analysis, the 15 percent plan, and what a further 15 percent would cost.

Question 32

You are leading a move off a legacy pipeline scheduler that runs 600 jobs across five teams. How do you sequence the work?

Accepted Answer

Move a few low-risk jobs end to end first, then scale the pattern team by team.

Question 33

Halfway through a pipeline tool migration, you learn that 150 jobs have no clear owner. What do you do?

Accepted Answer

Publish the list, set a claim date, and plan to retire unclaimed jobs after a brownout.

Question 34

Your migration plan needs each team to spend about two weeks moving its own jobs. Two teams say they cannot fit it in. What is the best response?

Accepted Answer

Cut their effort with tooling that converts most jobs, then agree a date.

Question 35

Six months into a pipeline migration, 85 percent of jobs have moved. The last 15 percent are complex and owners keep slipping. What do you do?

Accepted Answer

Set a firm shutdown date with leadership and staff a team to help stragglers.

Question 36

Your team is choosing a replacement for a legacy pipeline tool. Two engineers strongly prefer different options. How do you get to a decision?

Accepted Answer

Agree on weighted criteria first, test both on one real pipeline, then decide by a set date.

Question 37

Leadership asks why the pipeline migration is worth two quarters of the team's time. What is the strongest case?

Accepted Answer

The old tool's failure rate, on-call hours, and license cost compared with the new one.

Question 38

During a pipeline migration, how do you confirm that a moved job produces the same results as before?

Accepted Answer

Run old and new side by side for a period and compare outputs automatically.

Question 39

You are three months into a migration that was planned for four. An honest estimate now says seven. What do you do?

Accepted Answer

Tell stakeholders now with the new estimate, the cause, and options to cut scope.

Question 40

A migration requires a freeze on new jobs in the old pipeline tool. A team asks for one exception to ship an urgent pipeline. What do you do?

Accepted Answer

Help them build it on the new tool, even if it takes slightly longer to set up.

Question 41

Analysts file about thirty requests a week for new tables and fixes. Your platform roadmap keeps slipping. What is your best structural fix?

Accepted Answer

Set a visible intake process and reserve a fixed share of capacity for requests.

Question 42

A vice president's analyst asks for a new table by Friday. It would push a platform reliability fix by a week. What do you do?

Accepted Answer

Explain the trade-off to the requester and your roadmap owner, and let them pick.

Question 43

Many analyst requests are small variations of the same table. How do you reduce the load over time?

Accepted Answer

Build a documented core model analysts can extend, and point requests there.

Question 44

You say no to an analyst team's request for a custom pipeline. How do you say it?

Accepted Answer

Give the reason, what you can offer instead, and when it could be reconsidered.

Question 45

Your manager wants the team to take every analyst request because partner satisfaction is a team goal. Platform debt is causing weekly incidents. What do you do?

Accepted Answer

Show your manager how the incidents hurt the same partners, and propose a split of capacity.

Question 46

An urgent analyst request arrives every week labeled critical. Most turn out to be optional. How do you fix the pattern?

Accepted Answer

Agree with partner leads on what counts as urgent and track how often each label holds up.

Question 47

A partner team asks your team to own a reporting pipeline they built. It has no tests and breaks weekly. What is your position?

Accepted Answer

Agree only once it meets your standards, and offer help getting it there.

Question 48

An analyst director complains that your team never delivers her requests. Records show her requests are done in about two weeks on average. What do you do?

Accepted Answer

Meet her, share the data, and ask what would make requests feel faster.

Question 49

On Monday morning, the chief executive's weekly report shows revenue double the real value because a join duplicated rows. You lead data engineering. What do you do first?

Accepted Answer

Tell the report's audience the number is wrong and a correction is coming, then fix it.

Question 50

After a bad number reached an executive report, you run the incident review. What is the main goal?

Accepted Answer

Find which checks and processes let the error through and fix those gaps.

Question 51

An executive asks you who caused the bad numbers in last week's report. What do you say?

Accepted Answer

That the process missed it, what is changing, and that you own the fix.

Question 52

Follow-ups from a data quality incident have been open for six weeks and nobody is working on them. What do you do?

Accepted Answer

Give each action one owner and a due date, and review them in planning.

Question 53

After an executive report error, leadership asks for a sign-off step where you personally approve every report before it is sent. What do you propose instead?

Accepted Answer

Automated checks on key metrics that block publishing when values move too far.

Question 54

During a live data incident, three engineers are debugging separately and stakeholders are messaging each of them. What do you do?

Accepted Answer

Name one incident lead and one communicator, and use a single channel.

Question 55

A data incident report is due. A junior engineer wrote the change that caused it and is worried. How do you involve them?

Accepted Answer

Ask them to help write the timeline and fixes, and make clear the review is blameless.

Question 56

A wrong number in the board report came from a dashboard that pulled from an unofficial table. How do you prevent this class of error?

Accepted Answer

Mark certified tables and require board reports to use only those sources.

Question 57

Your team had three data incidents in a month, each from a different cause. What is the best next step as the lead?

Accepted Answer

Look for a shared gap across all three, such as missing tests, and invest in that.

Question 58

You are proposing a new event ingestion layer used by four teams. When should you share the design doc?

Accepted Answer

Early, as a short draft with open questions, before code is written.

Question 59

What should the first page of your design doc for a new data platform component contain?

Accepted Answer

The problem, goals and non-goals, the proposal, and the key trade-offs.

Question 60

In a design review for a new warehouse model, a senior analyst raises a strong objection near the end of the meeting. What do you do?

Accepted Answer

Note the concern, set a time to dig into it, and hold the decision until then.

Question 61

A design review for your streaming pipeline has gone through four rounds of comments over six weeks. Most comments now are small preferences. What do you do?

Accepted Answer

Set a decision date, address blocking concerns, and log the rest as possible follow-ups.

Question 62

Your design for a shared metrics layer is approved, but a staff engineer on another team still disagrees. What is the right next step?

Accepted Answer

Record their concern in the doc, commit, and agree what would trigger a revisit.

Question 63

You are reviewing a junior engineer's design doc for a new ingestion job. It has a serious flaw in how it handles late data. How do you give feedback?

Accepted Answer

Ask a question that leads them to the late-data case, then discuss options together.

Question 64

Two senior engineers on your team argue in a design review about partitioning strategy. Both options are workable. How do you close it?

Accepted Answer

Pick one with a short written reason, since both work and delay costs more than the gap.