Quant Estimation Interview Questions: Worked Examples

An estimation interview question asks you to build a defensible answer from incomplete information. The difficult part is often choosing what to count, deciding which assumptions matter and explaining how the estimate changes. Fast multiplication helps after you have built the model.

This guide uses original, hypothetical exercises. You can work through them with a notebook and then ask a partner to change an assumption. The goal is a calculation another person can inspect and improve.

For context, an Optiver intern's published account describes estimation questions and discussion of assumptions in his own interview experience. That is one person's account, rather than a current specification for every office. The underlying mathematical method is broader: MIT's estimation chapter treats units, dimensional analysis and order-of-magnitude calculations as connected problem-solving tools.

A method you can explain aloud

Start by defining the requested quantity. “How much storage?” could mean one day's text, one year's retained files or all replicated infrastructure. Those produce different answers even with identical arithmetic.

Write an equation before choosing numbers. For annual message payload, a possible model is:

People × messages per person per workday × bytes per message × workdays per year.

Then state your assumptions and compute with units attached. If your factors leave “bytes per person” when the target is “bytes per year,” something is missing. Finish by asking which assumption could move the answer most and whether an independent calculation gives a compatible result.

This is a practice routine, not a required interview script. Let the interviewer answer meaningful clarification questions; when information is unavailable, propose an assumption and continue.

Exercise 1: Estimate an organization's message storage

A hypothetical organization has 500 people. Estimate the annual storage needed for the text payload of internal messages. Exclude attachments, search indexes, backups, replication and message metadata. Use decimal units: 1 kB = 1,000 bytes and 1 GB = 1 billion bytes.

Choose your own assumptions first. For the worked example, assume 40 messages per person per workday, an average payload of 5 kB and 250 workdays per year. Count each message once when sent, regardless of the number of recipients.

Show answer and explanation

Daily messages = 500 × 40 = 20,000.

Daily payload = 20,000 × 5 kB = 100,000 kB = 100 MB.

Annual payload = 100 MB × 250 = 25,000 MB = 25 GB.

This answers the narrowly defined payload question. It does not size a production messaging system. Attachments or retention rules could dominate a different version of the problem.

A useful check calculates per person: 40 × 5 kB × 250 = 50 MB per year. Multiplying by 500 gives the same 25 GB. This check catches arithmetic slips, although it does not independently validate the assumptions.

Now the interviewer asks: “What is the peak message rate?” Can you obtain it from annual storage?

You need more assumptions. If messages arrive during an eight-hour active window, average traffic is 20,000 / 28,800, approximately 0.69 messages per second. If you stipulate that peak traffic is ten times that average, the estimate becomes 6.9 messages per second. The ten-times factor is a scenario assumption; annual totals alone cannot establish a peak.

Explain this distinction directly. An average can inform a model without containing all the information needed for its next question.

Exercise 2: Change the uncertain inputs

Keep 500 people and 250 workdays fixed. Consider three scenarios:

These are scenario bounds, not a statistical confidence interval. We have supplied no probability distribution or measured sampling uncertainty.

The model is multiplicative. Doubling messages while holding everything else fixed doubles storage. Doubling both messages and payload size multiplies storage by four. This tells you what to investigate next: actual message counts and payload sizes would be more useful than arguing about whether the year has 248 or 250 workdays.

Avoid presenting a detailed decimal answer when the inputs are rough. “About 25 GB of text payload under these assumptions” communicates more faithfully than “25.000 GB required.”

Exercise 3: Estimate venue throughput

A hypothetical event has 1,200 attendees. Sixty percent are expected to arrive in the busiest 20 minutes. Each check-in desk processes one attendee every 30 seconds, continuously. What is the minimum number of desks whose aggregate processing capacity matches that volume over 20 minutes?

Assume people can use any desk, staff have no breaks, there are no exceptions and throughput is deterministic. Solve this capacity question before discussing queues.

Show answer and explanation

Peak-window arrivals = 1,200 × 0.60 = 720 attendees.

One desk's capacity = 20 minutes × 2 attendees/minute = 40 attendees.

Required desks = 720 / 40 = 18.

A rate check gives the same answer: arrivals average 36 people/minute; each desk handles two per minute; 36 / 2 = 18.

Eighteen is an aggregate capacity floor, not a guarantee that everyone arriving during the window finishes before it ends. If all 720 people are waiting at the beginning, 18 continuously operating desks can finish them within 20 minutes under the stipulated service rate. If arrivals cluster near the end, some finish later. Waiting-time or completion-time predictions therefore require an arrival schedule as well as a service model. Variable service times, desk failures and exception handling could also require spare capacity.

Try two changes. At 40 seconds per attendee, each desk handles 30 people in the window, giving a capacity floor of 24 desks. If the original 30-second service time holds but 75% arrive in the window, the volume is 900 and 900 / 40 = 22.5, giving a capacity floor of 23 whole desks. State why you round up.

This exercise separates calculating a capacity floor from predicting service quality. Naming what a model leaves out is part of explaining its answer.

Make an estimate easier to challenge

A productive explanation exposes the important choices. Compare these openings:

“It's probably around 25 GB because there are 500 people.”

“I'll count text once per sent message, exclude attachments and use 250 working days. If each of 500 people sends 40 messages a day with a 5 kB average payload, I get about 25 GB annually. Message volume and payload size are the inputs I'd verify first.”

The second answer gives someone places to intervene. If they change the target to total retained storage, acknowledge that the original boundary has changed and add the missing components. Do not defend the first number after the question has changed.

If the calculations are slowing you down, use the separate mental-math drills. If choosing assumptions is the difficulty, spend more time writing models before working on speed. A quant consultation can help you discuss how those gaps fit your target role and preparation plan.

A worksheet for your next practice session

Ask your partner to alter one assumption after you finish. Repeat the calculation without rebuilding the entire explanation. Then ask them to change the scope, such as from average throughput to waiting time. That tests whether you notice when the existing equation stops answering the question.

Judge the session by whether the model was understandable, the units worked and the revision followed logically. Timing can help you compare your own attempts, but these exercises do not establish an employer's passing threshold. Add them to the four-week interview study plan alongside probability, coding and practice explaining your decisions.