Yelp · Primly Community

Yelp data scientist interview (SQL, case, and stats) what the loop looks like

de_derek · 5 replies

Went through the Yelp DS interview earlier this year, targeting a mid-level analytics DS role on their local ads team. This is the breakdown I wish I'd had.

Phone screen (30 min with recruiter): Standard. Background, why Yelp, salary range check. Nothing technical.

Technical screen (1 hr with DS hiring manager): This was heavier than I expected for a first technical round. About half was SQL and half was product analytics discussion.

The SQL problem: given a table of reviews with columns like userid, businessid, rating, date, they asked me to write a query to find businesses with a statistically significant drop in average rating over the last 90 days versus the prior 90 days. Not just the SQL syntax, they wanted me to think about what "significant" means with a small sample. So it turned into a mini statistics conversation.

Onsite (virtual, 4 rounds): SQL deep dive: more complex aggregations, window functions, they had me debug a query with subtle grouping logic that was producing double-counts. Good prep is writing and reading complex SQL, not just memorizing syntax. Stats and probability: sampling bias, A/B test design, p-value interpretation. One question on how you'd set up an experiment to measure whether a new review prompt changes review submission rates. This is standard DS interview territory but they went deeper than usual on the confounding variables discussion. Case study: given a dashboard showing a 15% drop in Yelp restaurant reviews in a metro area over two months, walk through how you'd diagnose the root cause. This was my favorite round. They wanted a structured problem breakdown: data quality checks first, then seasonality, then product changes, then competitive factors. Being systematic matters more than finding the "right" answer. Behavioral: similar to what others have posted, failures and cross-functional work came up.

Level and comp: For mid-level DS in SF, I was offered $175k base. Total comp with bonus and equity landed around $210k. I didn't negotiate aggressively because it matched my target.

Bottom line: Yelp's DS loop is SQL-heavy and product-analytics-focused. If your background is more ML engineering or modeling-heavy, it's worth asking the recruiter how the role breaks down before you get deep into prep.

5 replies

analyst_ana

The SQL for statistical significance is a level up from what I usually prep. Did they actually want you to compute a t-test in SQL or just flag the approach conceptually?

sre_sol

Mostly conceptual. They wanted to see that I'd bring up sample size as a concern before calling something significant. I sketched the idea of grouping by business and filtering for minimum review count. They weren't asking me to write the full test in SQL.

hardware_hugo

The double-count debugging question is a rite of passage. GROUP BY with a JOIN that fans out rows is how you end up with numbers that are exactly 2x reality and you spend three hours not knowing why.

finance_faye

Good flag on the modeling vs. analytics split. Yelp's DS roles skew product analytics. If you want to be building recsys or ML infra, their MLE track is probably more relevant.

growth_gabe

The review drop case study is exactly the kind of product sense question that trips up people who prep only for coding. Being able to walk through a metric movement systematically is its own skill.