I just passed Quora’s Data Scientist (DS) technical phone interview + OA, so I’m here to share valuable insights. As a giant in the Q&A community, Quora’s data roles place immense emphasis on experimental design and business acumen. The interview pace is steady yet challenging — a one-hour Case Study will thoroughly assess your capabilities from start to finish.
Interviewers are super friendly, but their questions are progressive, testing your logical and statistical skills. If you’re targeting DS positions at Quora or similar platforms, this article will help you avoid pitfalls and march straight to the VO (Virtual Onsite). Let’s dive in!
Interview Overview
About Quora
Founded in 2009 and headquartered in Mountain View, Quora is a knowledge-sharing platform with over 300 million users.
The DS role focuses on A/B testing, user behavior analysis, and product optimization, offering excellent benefits (generous RSU + remote flexibility).
I applied via HR referral after connecting on LinkedIn.
Phone Interview + OA Process (1 Hour Total)
OA Section:
- LeetCode-style medium-easy SQL + Python questions
- Examples: user retention calculation, data cleaning
- Used Pandas to process event logs → finished in 15 minutes
Technical Phone Interview (45 minutes):
- Pure Case Study on A/B testing
- Covers the entire workflow: from understanding the feature → experimental design → statistical testing
- Interviewer: Senior DS, fluent English, guides step-by-step discussion
Expected VO (4 Rounds):
- Statistics – modeling, hypothesis testing
- Metrics – indicator design and analysis
- Coding – Python/SQL practicals
- Product Sense – user journey and product logic
Core Case: Full Workflow of AB Test for a New Feature
The Scenario
Quora plans to introduce a window prompt on mobile web (mWeb) users with:
- “I want to download the app”
- “I already have the app”
- and an “X” to close
Goal: Measure success and evaluate whether the prompt improves conversion.
Step-by-Step Breakdown
1. Clarify the Feature — Don’t Rush
First, restate the feature clearly to align with the interviewer. Then ask:
- When is the prompt triggered? (first visit or specific pages?)
- How are users segmented? (new vs. existing, region?)
- Baseline metrics? (current download or bounce rate?)
Tip: Use a hypothesis-confirmation framework to show thoughtful, user-centric reasoning.
2. Choose Metrics — Primary + Guardrail + Secondary
| Type | Example | Why It Matters |
|---|---|---|
| Primary | App download conversion rate | Directly measures ROI |
| Guardrail | Time per user, sessions per user | Protects user experience |
| Secondary | Button CTR, X-click rate | Tracks interaction behavior |
Follow-ups:
- Weakness: “Time per user” may inflate artificially.
- Improvement: Exclude prompt display duration and use post-exposure metrics.
- Pro tip: Mention OKR/North Star frameworks to demonstrate strategic thinking.
3. Design the A/B Test
Setup:
- New vs. old prompt (old = plain download link)
- Randomly split 50/50 by user ID (to avoid session volatility)
- Test length: 2+ weeks to cover periodicity
Key Concept:
Why user-level bucketing? → Stable and prevents contamination.
Tip: Draw or verbally describe a simple flowchart:
user_id → exposure → metrics collection → analysis
4. Statistical Foundation — MDE, Power, Alpha
MDE (Minimum Detectable Effect):
MDE = Z * sqrt(2p(1-p)/n)
where p = baseline rate (e.g., 0.1)Variance Control:
Use stratified sampling to reduce confounding effects (device type, traffic source).Power & Sample Size:
Target Power = 80%, Alpha = 0.05
Formula:
n = (Zα + Zβ)^2 * 2σ^2 / δ^2
If variance is too high → increase sample size or reduce MDE.
Tip: Practice using Evan Miller’s calculators and Python power simulations.
5. Result Analysis
- For large samples (>30), use Z-test to compare mean differences.
-
Success criteria:
- p-value < 0.05
- CI does not include 0
- Effect size > MDE
Bonus point: Mention multi-armed bandit approaches to show advanced knowledge.
6. When Results Aren’t Significant
Stay calm — don’t kill the experiment immediately.
Steps:
- Check if sample size is sufficient.
- Look for data leakage or bucketing errors.
- Extend duration or refine variants.
If CI’s upper bound > 0 → try a pilot. Otherwise, conduct an After Action Review (AAR).
Final risk point: Cannibalization — app users returning to web after download.
Key Takeaways + VO Preparation Tips
Quora’s DS interview is friendly but analytical. You need both data rigor and product thinking.
- OA: Stay sharp with SQL + Pandas.
-
VO Topics:
- Statistics: hypothesis testing, Bayesian intuition
- Metrics: funnel optimization
- Coding: Numpy/Pandas logic questions
- Product Sense: user motivation, engagement drivers
OFFER PATHWAY: Let PROGRAMHELP Lead You There
PROGRAMHELP – the black-tech platform for conquering technical interviews, created by a team of returnees from Peking University and Oxford!
Our team (4 top-university Masters/PhDs + 3 senior engineers from Amazon/Google/Alibaba) provides direct expert guidance with zero intermediary fees.
Services include:
- OA ghostwriting – 100% pass rate for HackerRank
- Real-time VO assistance – North American experts give live voice prompts during your session
- Full-cycle offer support – from interview proxy to offer negotiation
⏰ 24-hour urgent support. Wherever you are, Programhelp helps you go from OA → VO → Offer — fast and flawless.
Top comments (0)