Snowflake Data Scientist Interview Questions

Snowflake·Data Scientist·Onsite - Coding / Algorithms

Jul 2026

Snowflake Data Scientist interview that went deep into algorithmic problem solving, way deeper than I expected for a DS role. The whole session was essentially a multi-part coding and system design marathon centered on a single problem with escalating constraints. Left feeling like I'd been wrung out.

Implement an O(n^2) solution to find all unique triplets in an array that sum to zero, using sorting and two pointers. Prove your duplicate-skipping logic is correct on adversarial inputs like an array of all zeros.
Generalize the triplet sum problem to an arbitrary target T and then to k-sum for k >= 4 with pruning. What time complexity do you achieve, and how do you prevent integer overflow when summing large values?
If the array has 5 million elements and memory is capped at 512 MB, how would you approach finding all triplets using an external-memory strategy? What is the I/O complexity? And if exact enumeration is infeasible, how would you approximate the count of unique triplets using probabilistic sketches?
Given a frequency map of values rather than the raw array, compute the number of unique triplets summing to a target T in O(m^2) time where m is the number of distinct values. Handle edge cases where two or all three values in a triplet are the same.

“The basic two-pointer approach I knew cold.”

View Post

Snowflake·Data Scientist·Onsite - Cross-functional / Panel

Jun 2026

Snowflake data scientist loop, heavy on the case-study presentation format. You bring a real shipped project and they pick it apart for about an hour across metrics, mistakes, stakeholder drama, and cross-functional process.

Walk us through an end-to-end project you led that shipped to users: the problem, stakeholder goals, data sources, modeling approach, key decisions, results, and trade-offs. You have 10 minutes and up to 5 slides.
What was the single metric you optimized for, what guardrails did you put in place, and why? Tell us about a time your chosen metric conflicted with another stakeholder's metric and how you resolved it.
Describe one concrete mistake from the project: a wrong metric trade-off, a flawed assumption, anything that actually hurt the outcome. What did you change afterward?
Leadership pushes back on your proposal because the metrics are moving in opposite directions, say retention is up but revenue is down. How do you redesign the follow-up experiment or rollout to address that without blowing up the timeline?
Give a specific example of working with data engineering, product, or design under time pressure. How did you negotiate scope or data model changes when you didn't have the luxury of time?

“This is the anchor for the whole session so whatever you pick sets the ceiling for every follow-up.”

View Post

Snowflake·Data Scientist·Technical Phone Screen

Jun 2026

Snowflake DS interview that was basically a full experiment design case study. One long question broken into five parts covering everything from sample size math to ramp decisions. Dense but fair if you've actually run A/B tests before.

You're replacing a feed ranker with a new model. Baseline CTR is 2%, you expect a +5% relative lift, and you want 90% power at alpha 0.05. Daily traffic is 1M users on a 50/50 user-level split. Compute the required sample size and minimum test duration, adjusting for 5% bot traffic and 1% sample ratio mismatch.
What guardrail metrics would you set for this experiment, and how would you use sequential monitoring with alpha-spending to allow early stopping without inflating Type I error?
How would you use CUPED to address novelty effects and day-of-week variation? What exact covariate would you use, and how do you validate that variance reduction isn't introducing bias?
How do you prevent interference and contamination between experiment arms, and when would you consider a geo-based or holdout design instead of user-level bucketing?
After the experiment concludes, how do you verify that you actually achieved the intended power, check for heterogeneous treatment effects across user cohorts, and decide when to ramp to 100%?

“This is where I spent most of my mental energy.”

View Post

Snowflake·Data Scientist·Onsite - Behavioral / Leadership

Jun 2026

Snowflake data scientist interview with a behavioral question that was way more layered than it looked on the surface. The core of it was about reading between the lines when someone keeps rephrasing the same question at you.

Walk me through a time a stakeholder or interviewer kept circling back to the same question in different forms, apparently looking for a specific answer. Cover the context, how you figured out the real concern, how you adjusted your communication style, how you confirmed you'd actually addressed it, what you sacrificed under time pressure, and what the result was. What would you change next time?

“This one is sneaky because it's asking you to meta-reflect on a communication breakdown while also demonstrating you can do the thing it's asking about in real time.”

View Post

Snowflake·Data Scientist·Onsite - System Design / Architecture

Jun 2026

Snowflake data scientist interview that was basically one long ML system design case study on purchase prediction. Every sub-question peeled back another layer and by part D I was running out of steam. Dense but fair.

Design an end-to-end training and evaluation pipeline for a 7-day purchase classifier that avoids label leakage given a 10-day label delay. Specify your time-based cross-validation scheme including fold boundaries and feature/label windows.
Which offline metrics would you use for this imbalanced classifier, how would you calibrate the model's probabilities, and what formula would you use to pick a decision threshold that maximizes expected profit given the stated costs?
How would you detect and respond to distribution shift in both input features and model calibration over time? Describe the monitoring dashboard and any guardrails you'd put in place.
Given a 50ms p95 latency budget and 64MB RAM per request, what model architecture and featurization strategy would you choose for deployment, and what fallback rule would you use if the model is unavailable?
How would you explain this model and its threshold choice to a non-technical stakeholder, and what would you do if they pushed back and insisted on using a different threshold?

“This is where I spent the most time and also made the most mistakes early in my answer.”

View Post

Snowflake·Data Scientist·Technical Phone Screen

Jun 2026

Snowflake data scientist interview that went way deeper into finance fundamentals than I expected. The whole thing centered on one long multi-part question about cash flow vs net income, and I genuinely was not prepared for how granular it got.

An asset-heavy company shows rising net income but negative free cash flow for three consecutive quarters. Walk through why FCF can be a better signal than net income for valuation, solvency, and dividend safety, and reconcile the two numerically including non-cash charges, working capital changes, and capex.
Give at least two specific situations where net income looks healthy but free cash flow is signaling real risk. Also describe one scenario where FCF itself can be temporarily misleading.
If you could only access one financial statement, which would you pick for each of these three goals: assessing credit risk, evaluating profitability trends, and estimating cash runway? For each choice, name three critical risks or gaps you can't assess without the other statements, and explain how you'd approximate them.

“I fumbled the numeric reconciliation part.”

View Post

Snowflake·Data Scientist·Technical Phone Screen

Jun 2026

Technical screen for a Data Scientist role at Snowflake. The whole thing was one big multi-part question combining BigQuery SQL, cohort analysis, and a Streamlit app build. Dense and exhausting, but kind of interesting if you're into that stuff.

Write a single BigQuery query using window functions with PARTITION BY that produces a cohort retention heatmap table with columns for signup week, week index, country, cohort size, retained users, and retention rate. Explain how you avoid double-counting users across weeks and handle users with late-arriving events.
Extend the cohort retention SQL to compute weekly ARPU by cohort and country, ensuring users with no purchases still count in the cohort size but contribute zero to revenue.
Build a Streamlit app that lets a user filter by country and toggle between a retention heatmap and an ARPU line chart, treats missing future weeks as NA rather than zero, masks cohorts below 50 users, and includes a UTC offset selector that re-buckets events by timezone without duplicating users across weeks.
Describe two data quality checks you would build into the app, such as verifying cohort size monotonicity across observed weeks and guarding against clock skew that produces negative week index values.

“This is the kind of question where you feel fine until you actually try to write it.”

View Post

Snowflake·Data Scientist·Technical Phone Screen

May 2026

Snowflake data scientist interview that was basically a probability and randomness gauntlet. Three connected problems all building on each other, and by part C I was definitely improvising. Tough but fair, the kind of interview where you can tell they actually care about whether you understand the math.

You have a function that returns a uniform random integer from 1 to 5. Implement a function that returns a uniform random integer from 1 to 7 using only that function. Your solution must be unbiased, always terminate, and close to optimal in how many calls it makes. Prove it's uniform and compute the exact expected number of calls.
Generalize the previous problem: given a function returning a uniform integer from 1 to M, implement one that returns a uniform integer from 1 to N for any positive integers M and N. Derive an upper bound on the expected number of calls in terms of M and N, and discuss how the approach changes when N is much larger than M versus when M is much larger than N.
Given a biased coin that flips heads with some unknown probability p, construct a fair coin flip. Prove it works and find the expected number of flips as a function of p. Then extend this to sampling uniformly from 1 to K using only the biased coin.

“I knew the rejection sampling trick from somewhere in my past but proving it rigorously on the spot was harder than I expected.”

View Post

Snowflake·Data Scientist·Onsite - System Design / Architecture

May 2026

This was a deep technical design round for a Data Scientist role at Snowflake, basically a full system design session focused on analytics infrastructure at scale. Four sub-questions, all interconnected, and the level of DDL and SQL detail expected was way higher than I anticipated.

Design a fact and dimension table schema for a warehouse handling 50M events per day. Include DDL-level details: partitioning, clustering keys, surrogate keys, and how you'd handle slowly changing dimensions for user country and app version.
Events can arrive up to 14 days late and out of order, with occasional duplicates on the same (user_id, event_ts, event_type) combination. What's your deduplication key and upsert strategy, and how do you handle late data reprocessing without double-counting cohorts? Include a backfill plan.
Define the grain for a derived metrics table covering weekly retention by signup cohort and ARPU by cohort. Show the SQL pattern you'd use, explain how incremental materialization works, and describe data quality checks you'd put in place.
Estimate table sizes for this 50M events/day system, justify your clustering key choices for typical queries like top-N countries over the last 8 weeks and rolling DAU/WAU/MAU, and walk through cost controls including partition pruning, approximate distinct counts, and result caching.

“I knew Type 2 SCDs conceptually but fumbled when they asked me to actually write the DDL on the spot.”

View Post

Snowflake·Data Scientist·Technical Phone Screen

May 2026

Snowflake Data Scientist interview with a pretty deep algorithms question that felt more like a software engineering screen than anything data-related. One question, lots of moving parts, and they wanted both a recursive and iterative solution plus a full explanation of design choices.

Implement both a recursive and an iterative DFS on a directed graph that returns a topological order if the graph is a DAG, or detects and returns a cycle if not. Must also track discovery and finish timestamps for every node, handle disconnected graphs, self-loops, and parallel edges, all in O(N+M) time and space. Output as a dict with keys for is_dag, topo_order, cycle, discover, and finish.

“This was a lot.”

View Post