Interview Leaks · Primly Community

data science technical interview at a mid-size startup vs big tech, the gap is real

ds_dmitri · 5 replies

Been doing a parallel track of interviews: two big tech companies and three Series B/C startups. Wanted to share what the actual question types look like because the gap is bigger than I expected.

Big tech (one FAANG, one large fintech):

Both had dedicated SQL rounds. These weren't 'write a select statement' tests. We're talking multi-step analytical questions: build a 7-day rolling retention metric, identify drop-off points in a funnel with messy event data, handle duplicate rows from a pipeline bug. One company gave me a 30-minute take-home SQL problem before the loop even started.

ML system design was real. I had to design an experimentation platform for a two-sided marketplace: how do you detect network effects in A/B tests, what's your variance reduction strategy, how do you set up guardrail metrics. This required actual knowledge, not just ML jargon.

Probability and stats were light in both. One question on expected value, one on Bayesian updating. Less than I expected.

Startups (all Series B or C, 150-600 employees):

No dedicated ML system design rounds. One startup asked me to walk through a model I had built and how I would productionize it, but it was conversational, not structured.

SQL was simpler but they added a 'now tell me what you'd do with this result' step that big tech didn't do explicitly. They want you to connect the output to a business decision in real time.

Product sense was heavy at all three. At one company the founder joined the last interview and asked me to diagnose a drop in a key metric we'd never discussed. Just from the numbers. No context.

One startup asked me about causal inference in the context of a pricing experiment. That surprised me.

Comp rough ranges: Big tech: $160k-$215k base, MLE-adjacent data science roles. The spread is huge. Startups: $130k-$165k base, meaningful equity that mayor may not matter.

Happy to answer specific questions about any of the rounds.

5 replies

ml_mike

The causal inference question at a startup surprised me when I went through something similar last year. It's more common than people expect now, especially if the startup has done any growth/pricing experimentation and doesn't want to hire someone who will just run bad A/B tests. Did they go deep into DiD vs RDD territory or keep it at the 'do you understand why this experiment is hard' level?

analyst_ana

The 7-day rolling retention in SQL is a classic that trips people up. Did they expect you to use window functions or were you allowed to self-join? I've seen interviewers have strong opinions on which approach they want.

ds_dmitri

Window functions only was the implicit expectation. When I started sketching a self-join approach for the first sub-question the interviewer just said 'can you do this with a window function instead' and waited. I think the window function approach is now table stakes at big tech for anything retention or cohort related.

sdr_sky

The comp spread at big tech is real and location-dependent in a way startups rarely are. Did the big tech offers have target bonus on top of base, or was it all in total cash plus RSU?

market_realist

Good comparison. The product sense requirement at startups is because the team is smaller and they need you to own the question 'so what does this mean for the business.' At big tech that role belongs to a PM and the DS can be more narrowly technical. Both make sense in context.