Instacart · Primly Community

Instacart data scientist interview (SQL + case + stats), my full notes

ds_dmitri · 6 replies

Went through Instacart's DS interview loop in March 2026 for a mid-senior DS role on their growth team. Want to be specific because most posts about DS interviews are either too vague or 3 years out of date.

Loop structure: SQL/coding screen (45 min, done async or live, mine was live) Case study / product analytics (60 min) Stats / probability (45 min) Behavioral (45 min)

SQL round: Three questions on a realistic-looking e-commerce schema (orders, order_items, users, shoppers). Difficulty: the first was a join + aggregate, something like compute average order value per user segment over the last 30 days. The second was a window function question: rank shoppers by fulfillment rate within each market. The third was trickier and involved a self-join to find users who placed two orders within 7 days of each other (retention signal). Know your window functions and self-joins cold.

Case study: They gave me a scenario: shopper pickup time has increased by 12% in the Southeast region. Diagnose it and recommend next steps. I walked through: is it a data issue, is it all categories or specific ones, is it new shoppers or experienced ones, did something change in store layouts or partner relationships. Then recommended a small experiment: A/B a guided picking route feature with experienced shoppers in one market. They asked for metrics, guardrails, and rough sample size. Spend time on metrics.

Stats round: Not super theoretical but not trivial either. Questions included: explain the difference between precision and recall and when you'd optimize for each. You're running an A/B test and your primary metric improved but a guardrail metric degraded. What do you do. Explain how you'd build a propensity model to identify shoppers at risk of churning. That last one got into feature selection and calibration.

Overall: Hard but fair. Takes about 3 weeks start to finish.

6 replies

analyst_ana

The self-join for 7-day retention pattern is a classic and I always mess it up under pressure. Do you remember whether they wanted it as a pure SQL problem or were you allowed to describe an approach in Python?

ds_dmitri

SQL only for that screen. They were testing whether you can actually write it, not just conceptualize it. Practice writing self-joins by hand without autocomplete. Sounds obvious but the syntax trips people up when they're nervous.

sdr_sky

What was the comp range for mid-senior DS at Instacart in 2026? Have some data points but they're from 2024.

ds_dmitri

My offer was around $175-185k base plus equity. RSUs, 4-year vest with 1-year cliff. Total comp in the $220-240k range depending on stock assumptions. SF Bay Area. Could be different in remote roles but I think the base is roughly consistent.

ml_mike

The precision/recall question and the 'guardrail metric degraded' scenario are both signs that they have real ML culture and not just 'we run SQL on data' culture. That's actually reassuring for DS roles there.

bootcamp_bri

3 weeks start to finish is pretty manageable. I was expecting 6+ weeks from other companies I'm in process with right now.