Okta · Primly Community

Okta data scientist interview (SQL + case + stats): breakdown of what they actually test

analyst_ana · 5 replies

went through okta's DS interview process for a mid-level role on their customer analytics team. sharing this because there wasn't much online about the DS-specific loop and i had to piece it together from scattered sources.

round structure: recruiter screen (30 min) hiring manager screen (45 min) SQL/data skills round (60 min) case/product analytics round (60 min) stats and probability round (45 min) behavioral (45 min)

SQL round: this was harder than i expected. they use a tool like CoderPad with a mock schema (user activity table, events table, something like a product log). questions i got: find users who performed action A but not action B within 7 days of signup (classic funnel, window functions) calculate 30-day rolling retention by cohort write a query to flag anomalous sessions based on session length percentile

if you're rusty on window functions (RANK, ROW_NUMBER, LAG, LEAD), practice those specifically. CTEs are expected. performance considerations came up: 'how would you optimize this if the events table had 10B rows'.

case/product analytics round: something like: okta wants to understand why enterprise customers are not adopting MFA for all their users. how would you approach this? they want: hypothesis generation, metric selection, analysis plan, and then 'what would you actually do with the findings'.

stats round: explain p-values to a non-technical stakeholder how would you design an experiment to test a new onboarding flow when you can only run it on 20% of new customers describe a situation where a model you built was technically correct but the business acted on it wrong

the last one is a great question and worth having a real answer for.

overall: okta's DS bar feels like a serious mid-level analytics bar, not research-heavy. less ML theory than i expected, more applied SQL and business case judgment. the 'why does this metric move' type thinking matters more than knowing SVMs cold.

5 replies

analyst_ana

the rolling retention query is one of those things where i know the concept but always fumble the SQL. do you remember if they let you look things up or was it fully from memory?

hardware_hugo

fully from memory in CoderPad. no docs. syntax errors were fine as long as the logic was right. they cared more about whether you knew you needed a self-join or a window function than whether your parentheses were perfect.

de_derek

the 10B row optimization question is almost more of a data engineering question than DS. did you feel like they wanted a DS answer (sample first, then analyze) or an eng answer (partition the table, push down filters)?

alex_design

both, actually. i gave the DS answer first (work with a sample, validate the query, then run on full data) and then added the eng layer (why you'd want date-partitioned tables, how pushdown predicates help). they seemed to like that i knew both layers existed.

brand_ben

what's the comp range for this role? ML/DS at identity-focused companies varies a lot. are they paying SaaS-mid-market comp or closer to faang-adjacent?