Question 5Discovery and scoping
A delivery company is losing money and hands you six months of order data. Where do you look first?
This is the classic case round: a vague business problem and a pile of data. They watch how you narrow it down.
“Run discovery and translate ambiguous business problems into clear requirements, scope, and acceptance criteria”Senior Forward Deployed Engineer, Applied AI · Juniper Square
The answerFour steps
- 1
Ask how they measure “losing money” today and since when
Say it like this
“I'd ask what “losing money” means to them, margin per order or total loss, and since which month.”
Why it scores
Pins down the target before touching data. A vague goal gives a vague analysis.
- 2
Split cost per delivery by region, time, driver and order size to find where the loss sits
Say it like this
“Then I'd break cost per delivery by region, hour, driver and order size, and look for where it moved.”
Why it scores
Splits along the cuts that drive delivery cost, to find where the loss sits.
- 3
Check the data for gaps that could fake a pattern (refunds, cancelled orders)
Say it like this
“Before I trust that, I check whether refunds and cancelled orders are in the data, because a missing refund table can make one area look worse than it is.”
Why it scores
Questions the data before trusting the pattern. Interviewers wait for this step.
- 4
Propose one fix to test in a single region, with the number that proves it worked
Say it like this
“If it holds, I'd propose one change, like a minimum order value in those suburbs after 6pm, tested for four weeks against margin per delivery.”
Why it scores
One change, one place, one number. Turns analysis into a decision.
StrongAll together
About a minute, out loud. The marked parts are the four steps; hover one to find it above.
I'd ask what “losing money” means to them, margin per order or total loss, and since which month.1 Say margin per delivery fell from $2.10 to $0.40 since March. Then I'd break cost per delivery by region, hour, driver and order size, and look for where it moved.2 Say it's evening orders in two suburbs. Before I trust that, I check whether refunds and cancelled orders are in the data, because a missing refund table can make one area look worse than it is.3 If it holds, I'd propose one change, like a minimum order value in those suburbs after 6pm, tested for four weeks against margin per delivery.4
WeakThe answer that fails
Sounds fine. Common. Scores low.
“I'd load the data into a notebook, do some exploratory analysis and charts, look for trends and outliers, and maybe train a model to predict which orders lose money.”
Interviewer's note
No question, no hypothesis, no check on the data. A model before anyone knows where the loss is.
PracticeTwo minutes, then they push back
Your turn
A delivery company is losing money and hands you six months of order data. Where do you look first?
Answer like you're in the room: about two minutes, no notes. Say it out loud, even alone. It feels different from thinking it. Then check your answer against what interviewers score.