Some analyst interviews include a question with no dataset at all: "Estimate the number of …" or "Sales dropped last week; what would you do?". These test structured thinking. There is rarely one right answer; there is definitely a right way to approach it.
Guesstimates
A guesstimate asks you to estimate a quantity using logic and reasonable assumptions.
The approach
- Clarify the question: what exactly, where, over what time?
- Choose a structure: top-down (start from a population and narrow down) or bottom-up (start from a unit, such as one shop, and scale up).
- Break it into factors you can estimate.
- State each assumption aloud and round numbers to keep maths simple.
- Calculate step by step.
- Sanity-check the result: does it seem plausible?
Worked example: cups of tea sold daily at a college canteen
Structure it bottom-up:
- Students and staff on campus on a typical day: assume a round number, say a few thousand.
- Share who buy tea at the canteen: assume around a quarter.
- Cups per buyer per day: assume between one and two, say 1.5 on average.
Multiply the factors to get the estimate, then sanity-check against the canteen's capacity: how many cups can the counter serve in the busy hours?
The numbers are invented assumptions; the interviewer is judging your structure, your clarity and whether you check the result, not the final figure.
Say your assumptions as assumptions: "I'll assume roughly a quarter of people buy tea there; we could check this with sales data." That shows analytical honesty.
Business case questions
These sound like real problems: "Our app's daily orders dropped 15% last week. How would you investigate?"
A structure that works
- Clarify: Compared with what? Which platforms, cities, customer types? Is the data itself reliable?
- Check the data first: a tracking or pipeline problem can look like a business problem.
- Break down the metric: orders = visitors × conversion rate. Which part fell? Then go further: visitors by channel; conversion by step (browse, cart, payment).
- Segment: by city, platform, new vs returning customers, category.
- Form hypotheses: an app release bug, a payment gateway problem, a competitor offer, a price change, a holiday or weather effect, a supply problem.
- Prioritise: check the quickest and most likely explanations first.
- Recommend: what you would do once you find the cause, and how you would measure the fix.
Metric questions
"What metrics would you track for a food delivery app?" Organise by stage:
- Acquisition: installs, sign-ups, cost per acquisition.
- Engagement: sessions, orders per user.
- Conversion: cart-to-order rate, payment success rate.
- Operations: delivery time, late-delivery share, cancellations.
- Retention: repeat order rate, cohort retention.
- Money: average order value, revenue, margin per order.
Then name the one or two you would treat as the main metric and why.
What interviewers are really checking
- Do you clarify before you dive in?
- Do you structure the problem rather than listing random ideas?
- Do you think about data quality?
- Do you end with a recommendation?
Practise
Practise structured thinking on realistic problems with context: our guided projects and live projects are built around investigations like these. Then rehearse explaining your approach aloud in a mock interview.
Practise what you just read
Reading helps; writing the code yourself is what sticks. Start here: