Practise the guesstimate questions Google, Meta, Amazon, and McKinsey actually ask.
Capacity planning. Market sizing. Cost estimation. The structured-thinking questions that separate offer-getters from rejections, graded by AI against the same four-dimension rubric a real interviewer uses, in seconds, with feedback that tells you exactly what was missing.
Free account · No credit card · AI feedback in seconds
Estimation questions appear in PM rounds at Meta, Google, Amazon, and Microsoft, in system-design rounds across FAANG, and in nearly every McKinsey, BCG, and Bain case interview. The pattern is everywhere. The preparation isn't.
Company names and logos are the trademarks of their owners, shown here only to say where this style of question is asked. Journey Uncommon is not affiliated with, endorsed by, or sponsored by any of these companies.
Most interview-prep platforms hand you a list of questions and a stopwatch. We hand you a real interviewer.
Generic flashcards train memorisation. Mock interviews cost real money per session. Here you type a real, structured guesstimate answer and get it graded, with calibrated 0-100 feedback, in under three seconds. Reading the questions and rubrics costs nothing.
What you get
- AI grader trained on the same four-dimension rubric MBB and FAANG interviewers use.
- Feedback that names exactly what was missing: assumptions, arithmetic, sanity check.
- 1 AI-graded attempt a day on Free, 30 a day on Pro. The grader doesn’t get tired of you.
What you won’t find
- Generic memorisation drills with no feedback.
- Pay-per-mock-interview pricing.
- Vague “good answer” scoring that doesn’t tell you why.
Three seconds from submit
- 0 to 100 score against the rubric.
- Natural-language feedback on each dimension.
- A reference answer to compare against.
Why interviewers love guesstimates
Real consulting and product work means making decisions with incomplete data. Guesstimates test whether you can stay structured under uncertainty and whether your math holds up when someone pushes back. They reveal more in 4 minutes than a behavioral question does in 30.
Structure over speed
Interviewers grade the framework, not the final number. We train you to slow down and show your work.
Calibrated assumptions
Every assumption you state is a chance to lose or earn points. Practice making them explicit.
Defendable answers
When the interviewer pushes back ("why 2x not 5x?") you should already know your weak link.
Same rubric a real interviewer uses.
Type your reasoning into the answer box. Our AI scores it against a hand-written rubric, the same four dimensions a McKinsey consultant or a Google PM would mark you on. You get a 0 to 100 score and targeted feedback in seconds.
Problem framing
Did you restate the problem and pick the right proxy to estimate from?
Assumptions
Are your numbers explicit (DAU, sessions/day, requests/session) rather than hand-waved?
Arithmetic
Do the numbers actually multiply correctly to your stated answer?
Sanity check
Did you compare your answer against a reference, or break out peak vs. average?
Estimation isn't a quirky interview gimmick.
It's how engineers and PMs at the most demanding companies actually work, under uncertainty and with incomplete data. We trained our AI grader on the same dimensions interviewers grade you on.
“back-of-the-envelope calculations are estimates you create using a combination of thought experiments and common performance numbers to get a good feel for which designs will meet your requirements”
“Estimation questions test your ability to size a market, approximate usage, or calculate scale with incomplete information. Interviewers care more about your reasoning structure and assumptions than whether you arrive at the exact number.”
Quotes are verbatim from public industry sources. Links above lead to the original article so you can verify in one click. Platform stats are auditable.
Try a guesstimate on the house.
Six real questions across capacity planning, frontend perf, JVM tuning, cloud cost, and storage growth. Type your answer; see your score in seconds. Sign up to track your scores and unlock more.
Estimate HTTPS requests per second for a mid-traffic weather app on a normal Tuesday.
Hint: Start from DAU. Walk through requests-per-session. Multiply through.
How many concurrent WebSocket connections can one Node.js process hold on a 4-vCPU, 8GB box?
Hint: What's the real bottleneck: file descriptors, memory, or the event loop?
5M page views/day, 500-byte log lines. How much disk for a year, and what storage would you pick?
Hint: Raw vs. compressed. Hot vs. cold. Tier the retention.
Your React SPA ships 2 MB gzipped. How long until interactive on a mid-tier Android over 4G, and what's the real fix?
Hint: Network is only ~30% of the cost. Parse and hydrate dominate.
Sizing a JVM microservice: 500 RPS, 50K sessions, 10K cache entries. What's the right -Xmx, and why not the whole pod?
Hint: GC pause time scales with live heap. Bigger isn't always better.
Monthly AWS bill for a 100K-MAU B2B SaaS on Next.js + RDS. Show the line items.
Hint: At this scale, egress and RDS dominate. EC2 is not the problem.
Frequently asked
Which companies actually ask guesstimate questions?
Meta, Google, Amazon, and Microsoft ask estimation questions in PM and system-design rounds. McKinsey, BCG, and Bain build them into case interviews. The format varies (market sizing for PMs, capacity, cost, and scaling math for engineers) but the muscle is the same: structure your assumptions, do the math, sanity-check the answer.
What's the right way to approach a guesstimate?
Five steps: (1) clarify the question, (2) pick top-down or bottom-up, (3) state your assumptions explicitly, (4) walk through the arithmetic without rushing, (5) sanity-check against a reference. Interviewers care way more about the structure than the final number. A clean answer that lands 2× off scores higher than a confident guess that happens to be right.
How does the AI grader work?
Every question on this page has a hand-written rubric across four dimensions: problem framing, assumptions, arithmetic, sanity check. Our AI grader (Claude Haiku) scores your written answer against the rubric and gives you targeted feedback on what was missing. It's strict but fair. Vague hand-waving scores below 0.3, a complete answer with shown work scores 0.85+.
Is this free, and what are the limits?
Reading every question, every rubric, and every reference answer is free and always will be, with no account. AI grading costs us a model call per answer, so it is capped: 1 graded answer per day on the free plan, 30 per day on Pro (₹199 a month). Logged-out visitors get a small daily allowance per IP address. We would rather print the real number than call it unlimited.
Are these the exact questions companies ask?
No interview platform should claim that. Interviewers vary their questions and many are confidential. What we do guarantee: every question on this page is in the same format, difficulty range, and reasoning style as what you'll face. Practising these will train the structure interviewers grade you on.
The cheapest mock interviewer who never sleeps.
Type a real answer. Get a real score in three seconds. No credit card, and no paywalled rubric: every guesstimate on this page and its grading rubric are free to read before you start. Free covers 1 graded answer a day.